It now uses custom alloc_aligned() wrapper for all allocations,
therefore all allocations are larger than expected by (64 + sizeof(void*)).
Further, they are seen as allocations of 1 element. Relevant calculations
were adjusted to reflect this, and state allocation is now protected
with a flag to avoid misinterpreting other allocations as the zlib
deflate_state allocation.
Further, it no longer forces window bits to 13 on compression level 1,
so the comment was adjusted to reflect this.
When establishing a connection to the backend, nginx blocks reading
from the client with ngx_mail_proxy_block_read(). Previously, such
events were lost, and in some cases this resulted in connection hangs.
Notably, this affected mail_imap_ssl.t on Windows, since the test
closes connections after requesting authentication, but without
waiting for any responses (so the connection close events might be
lost).
Fix is to post an event to read from the client after connecting to
the backend if there were blocked events.
SSL context is not present if the default server has neither certificates nor
ssl_reject_handshake enabled. Previously, this led to null pointer dereference
before it would be caught with configuration checks.
Additionally, non-default servers with distinct SSL contexts need to initialize
compatibility layer in order to complete a QUIC handshake.
This ensures that errors which happen during logging to syslog are logged
with proper context, such as "while logging to syslog" and the server name.
Prodded by Safar Safarly.
During initial startup the ngx_cycle->hostname is not available, and
previously this resulted in incorrect logging. Instead, hostname from the
configuration being parsed is now preserved in the syslog peer structure
and then used during logging.
Similarly, ngx_cycle->log might not match the configuration where the
syslog peer is defined if the configuration is not yet fully applied,
and previously this resulted in unexpected logging of syslog errors
and debug information. Instead, cf->cycle->new_log is now referenced
in the syslog peer structure and used for logging, similarly to how it
is done in other modules.
Similarly to ticket #274 (7354:1812f1d79d84), early request finalization
without calling ngx_http_run_posted_requests() resulted in a connection
hang (a socket leak) if the 400 (Bad Request) error was generated in
ngx_http_v2_state_process_header() due to invalid request headers and
"return 444" was used in error_page 400.
As tested with tlsfuzzer with LibreSSL 3.7.0, the following errors are
certainly client-related:
SSL_do_handshake() failed (SSL: error:14026073:SSL routines:ACCEPT_SR_CLNT_HELLO:bad packet length)
SSL_do_handshake() failed (SSL: error:1402612C:SSL routines:ACCEPT_SR_CLNT_HELLO:ssl3 session id too long)
SSL_do_handshake() failed (SSL: error:140380EA:SSL routines:ACCEPT_SR_KEY_EXCH:tls rsa encrypted value length is wrong)
Accordingly, the SSL_R_BAD_PACKET_LENGTH ("bad packet length"),
SSL_R_SSL3_SESSION_ID_TOO_LONG ("ssl3 session id too long"),
SSL_R_TLS_RSA_ENCRYPTED_VALUE_LENGTH_IS_WRONG ("tls rsa encrypted value
length is wrong") errors are now logged at the "info" level.
To further differentiate client-related errors and adjust logging levels
of various SSL errors, nginx was tested with tlsfuzzer with multiple
OpenSSL versions (3.1.0-beta1, 3.0.8, 1.1.1t, 1.1.0l, 1.0.2u, 1.0.1u,
1.0.0s, 0.9.8zh).
The following errors were observed during tlsfuzzer runs with OpenSSL 3.0.8,
and are clearly client-related:
SSL_do_handshake() failed (SSL: error:0A000092:SSL routines::data length too long)
SSL_do_handshake() failed (SSL: error:0A0000A0:SSL routines::length too short)
SSL_do_handshake() failed (SSL: error:0A000124:SSL routines::bad legacy version)
SSL_do_handshake() failed (SSL: error:0A000178:SSL routines::no shared signature algorithms)
Accordingly, the SSL_R_DATA_LENGTH_TOO_LONG ("data length too long"),
SSL_R_LENGTH_TOO_SHORT ("length too short"), SSL_R_BAD_LEGACY_VERSION
("bad legacy version"), and SSL_R_NO_SHARED_SIGNATURE_ALGORITHMS
("no shared signature algorithms", misspelled as "sigature" in OpenSSL 1.0.2)
errors are now logged at the "info" level.
Additionally, the following errors were observed with OpenSSL 3.0.8 and
with TLSv1.3 enabled:
SSL_do_handshake() failed (SSL: error:0A00006F:SSL routines::bad digest length)
SSL_do_handshake() failed (SSL: error:0A000070:SSL routines::missing sigalgs extension)
SSL_do_handshake() failed (SSL: error:0A000096:SSL routines::encrypted length too long)
SSL_do_handshake() failed (SSL: error:0A00010F:SSL routines::bad length)
SSL_read() failed (SSL: error:0A00007A:SSL routines::bad key update)
SSL_read() failed (SSL: error:0A000125:SSL routines::mixed handshake and non handshake data)
Accordingly, the SSL_R_BAD_DIGEST_LENGTH ("bad digest length"),
SSL_R_MISSING_SIGALGS_EXTENSION ("missing sigalgs extension"),
SSL_R_ENCRYPTED_LENGTH_TOO_LONG ("encrypted length too long"),
SSL_R_BAD_LENGTH ("bad length"), SSL_R_BAD_KEY_UPDATE ("bad key update"),
and SSL_R_MIXED_HANDSHAKE_AND_NON_HANDSHAKE_DATA ("mixed handshake and non
handshake data") errors are now logged at the "info" level.
Additionally, the following errors were observed with OpenSSL 1.1.1t:
SSL_do_handshake() failed (SSL: error:14094091:SSL routines:ssl3_read_bytes:data between ccs and finished)
SSL_do_handshake() failed (SSL: error:14094199:SSL routines:ssl3_read_bytes:too many warn alerts)
SSL_read() failed (SSL: error:1408F0C6:SSL routines:ssl3_get_record:packet length too long)
SSL_read() failed (SSL: error:14094085:SSL routines:ssl3_read_bytes:ccs received early)
Accordingly, the SSL_R_CCS_RECEIVED_EARLY ("ccs received early"),
SSL_R_DATA_BETWEEN_CCS_AND_FINISHED ("data between ccs and finished"),
SSL_R_PACKET_LENGTH_TOO_LONG ("packet length too long"), and
SSL_R_TOO_MANY_WARN_ALERTS ("too many warn alerts") errors are now logged
at the "info" level.
Additionally, the following errors were observed with OpenSSL 1.0.2u:
SSL_do_handshake() failed (SSL: error:1407612A:SSL routines:SSL23_GET_CLIENT_HELLO:record too small)
SSL_do_handshake() failed (SSL: error:1408C09A:SSL routines:ssl3_get_finished:got a fin before a ccs)
Accordingly, the SSL_R_RECORD_TOO_SMALL ("record too small") and
SSL_R_GOT_A_FIN_BEFORE_A_CCS ("got a fin before a ccs") errors are now
logged at the "info" level.
No additional client-related errors were observed while testing with
OpenSSL 3.1.0-beta1, OpenSSL 1.1.0l, OpenSSL 1.0.1u, OpenSSL 1.0.0s,
and OpenSSL 0.9.8zh.
In some cases there might be multiple errors in the OpenSSL error queue,
notably when a libcrypto call fails, and then the SSL layer generates
an error itself. For example, the following errors were observed
with OpenSSL 3.0.8 with TLSv1.3 enabled:
SSL_do_handshake() failed (SSL: error:02800066:Diffie-Hellman routines::invalid public key error:0A000132:SSL routines::bad ecpoint)
SSL_do_handshake() failed (SSL: error:08000066:elliptic curve routines::invalid encoding error:0A000132:SSL routines::bad ecpoint)
SSL_do_handshake() failed (SSL: error:0800006B:elliptic curve routines::point is not on curve error:0A000132:SSL routines::bad ecpoint)
In such cases it seems to be better to determine logging level based on
the last error in the error queue (the one added by the SSL layer,
SSL_R_BAD_ECPOINT in all of the above example example errors). To do so,
the ngx_ssl_connection_error() function was changed to use
ERR_peek_last_error().
An UTF-8 octet sequence cannot start with a 11111xxx byte (above 0xf8),
see https://datatracker.ietf.org/doc/html/rfc3629#section-3. Previously,
such bytes were accepted by ngx_utf8_decode() and misinterpreted as 11110xxx
bytes (as in a 4-byte sequence). While unlikely, this can potentially cause
issues.
Fix is to explicitly reject such bytes in ngx_utf8_decode().
Just a drive letter might not correctly represent file system being used,
notably when using symlinks (as created by "mklink /d"). As such, instead
of trying to call GetDiskFreeSpace() with just a drive letter, we now always
use GetDiskFreeSpace() with full path.
Further, it looks like the code to use just a drive letter never worked,
since it tried to test name[2] instead of name[1] to be ':'.
This ensures that ngx_win32_rename_file() will support non-ASCII names
when supported by the wrappers.
Notably, this is used by PUT requests in the dav module when overwriting
existing files with non-ASCII names (ticket #1433).
Previously, ngx_win32_rename_file() retried on all errors returned by
MoveFile() to a temporary name. It only make sense, however, to retry
when the destination file already exists, similarly to the condition
when ngx_win32_rename_file() is called. Retrying on other errors is
meaningless and might result in an infinite loop.
This makes it possible to create directories under prefix with non-ASCII
characters, as well as makes it possible to create directories with non-ASCII
characters when using the dav module (ticket #1433).
To ensure that the dav module operations are restricted similarly to
other file operations (in particular, short names are not allowed), the
ngx_win32_check_filename() function is used. It improved to support
checking of just dirname, and now can be used to check paths when creating
files or directories.
Notably, ngx_open_dir() now supports opening directories with non-ASCII
characters, and directory entries returned by ngx_read_dir() are properly
converted to UTF-8.
To ensure proper target selection the NGX_MACHINE variable is now set
based on the MSVC compiler output, and the OpenSSL target is set based
on it.
This is not important as long as "no-asm" is used (as in misc/GNUmakefile
and win32 build instructions), but might be beneficial if someone is trying
to build OpenSSL with assembler code.
Previously, NGX_MACHINE was not set when crossbuilding, resulting in
NGX_ALIGNMENT=16 being used in 32-bit builds (if not explicitly set to a
correct value). This in turn might result in memory corruption in
ngx_palloc() (as there are no usable aligned allocator on Windows, and
normal malloc() is used instead, which provides 8 byte alignment on
32-bit platforms).
To fix this, now i386 machine is set when crossbuilding, so nginx won't
assume strict alignment requirements.
Output examples in English, Russian, and Spanish:
Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 16.00.30319.01 for 80x86
Оптимизирующий 32-разрядный компилятор Microsoft (R) C/C++ версии 16.00.30319.01 для 80x86
Compilador de optimización de C/C++ de Microsoft (R) versión 16.00.30319.01 para x64
Since most of the words are translated, instead of looking for the words
"Compiler Version" we now search for "C/C++" and the version number.
This is expected to help with clients using pipelining with some constant
depth, such as apt[1][2].
When downloading many resources, apt uses pipelining with some constant
depth, a number of requests in flight. This essentially means that after
receiving a response it sends an additional request to the server, and
this can result in requests arriving to the server at any time. Further,
additional requests are sent one-by-one, and can be easily seen as such
(neither as pipelined, nor followed by pipelined requests).
The only safe approach to close such connections (for example, when
keepalive_requests is reached) is with lingering. To do so, now nginx
monitors if pipelining was used on the connection, and if it was, closes
the connection with lingering.
[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=973861#10
[2] https://mailman.nginx.org/pipermail/nginx-devel/2023-January/ZA2SP5SJU55LHEBCJMFDB2AZVELRLTHI.html
Since 4611:2b6cb7528409 responses from the gzip static, flv, and mp4 modules
can be used with subrequests, though empty files were not properly handled.
Empty gzipped, flv, and mp4 files thus resulted in "zero size buf in output"
alerts. While valid corresponding files are not expected to be empty, such
files shouldn't result in alerts.
Fix is to set b->sync on such empty subrequest responses, similarly to what
ngx_http_send_special() does.
Additionally, the static module, the ngx_http_send_response() function, and
file cache are modified to do the same instead of not sending the response
body at all in such cases, since not sending the response body at all is
believed to be at least questionable, and might break various filters
which do not expect such behaviour.
The "listen" directive in the http module can be used multiple times
in different server blocks. Originally, it was supposed to be specified
once with various socket options, and without any parameters in virtual
server blocks. For example:
server { listen 80 backlog=1024; server_name foo; ... }
server { listen 80; server_name bar; ... }
server { listen 80; server_name bazz; ... }
The address part of the syntax ("address[:port]" / "port" / "unix:path")
uniquely identifies the listening socket, and therefore is enough for
name-based virtual servers (to let nginx know that the virtual server
accepts requests on the listening socket in question).
To ensure that listening options do not conflict between virtual servers,
they were allowed only once. For example, the following configuration
will be rejected ("duplicate listen options for 0.0.0.0:80 in ..."):
server { listen 80 backlog=1024; server_name foo; ... }
server { listen 80 backlog=512; server_name bar; ... }
At some point it was, however, noticed, that it is sometimes convenient
to repeat some options for clarity. In nginx 0.8.51 the "ssl" parameter
was allowed to be specified multiple times, e.g.:
server { listen 443 ssl backlog=1024; server_name foo; ... }
server { listen 443 ssl; server_name bar; ... }
server { listen 443 ssl; server_name bazz; ... }
This approach makes configuration more readable, since SSL sockets are
immediately visible in the configuration. If this is not needed, just the
address can still be used.
Later, additional protocol-specific options similar to "ssl" were
introduced, notably "http2" and "proxy_protocol". With these options,
one can write:
server { listen 443 ssl backlog=1024; server_name foo; ... }
server { listen 443 http2; server_name bar; ... }
server { listen 443 proxy_protocol; server_name bazz; ... }
The resulting socket will use ssl, http2, and proxy_protocol, but this
is not really obvious from the configuration.
To emphasize such misleading configurations are discouraged, nginx now
warns as long as the "listen" directive is used with options different
from the options previously used if this is potentially confusing.
In particular, the following configurations are allowed:
server { listen 8401 ssl backlog=1024; server_name foo; }
server { listen 8401 ssl; server_name bar; }
server { listen 8401 ssl; server_name bazz; }
server { listen 8402 ssl http2 backlog=1024; server_name foo; }
server { listen 8402 ssl; server_name bar; }
server { listen 8402 ssl; server_name bazz; }
server { listen 8403 ssl; server_name bar; }
server { listen 8403 ssl; server_name bazz; }
server { listen 8403 ssl http2; server_name foo; }
server { listen 8404 ssl http2 backlog=1024; server_name foo; }
server { listen 8404 http2; server_name bar; }
server { listen 8404 http2; server_name bazz; }
server { listen 8405 ssl http2 backlog=1024; server_name foo; }
server { listen 8405 ssl http2; server_name bar; }
server { listen 8405 ssl http2; server_name bazz; }
server { listen 8406 ssl; server_name foo; }
server { listen 8406; server_name bar; }
server { listen 8406; server_name bazz; }
And the following configurations will generate warnings:
server { listen 8501 ssl http2 backlog=1024; server_name foo; }
server { listen 8501 http2; server_name bar; }
server { listen 8501 ssl; server_name bazz; }
server { listen 8502 backlog=1024; server_name foo; }
server { listen 8502 ssl; server_name bar; }
server { listen 8503 ssl; server_name foo; }
server { listen 8503 http2; server_name bar; }
server { listen 8504 ssl; server_name foo; }
server { listen 8504 http2; server_name bar; }
server { listen 8504 proxy_protocol; server_name bazz; }
server { listen 8505 ssl http2 proxy_protocol; server_name foo; }
server { listen 8505 ssl http2; server_name bar; }
server { listen 8505 ssl; server_name bazz; }
server { listen 8506 ssl http2; server_name foo; }
server { listen 8506 ssl; server_name bar; }
server { listen 8506; server_name bazz; }
server { listen 8507 ssl; server_name bar; }
server { listen 8507; server_name bazz; }
server { listen 8507 ssl http2; server_name foo; }
server { listen 8508 ssl; server_name bar; }
server { listen 8508; server_name bazz; }
server { listen 8508 ssl backlog=1024; server_name foo; }
server { listen 8509; server_name bazz; }
server { listen 8509 ssl; server_name bar; }
server { listen 8509 ssl backlog=1024; server_name foo; }
The basic idea is that at most two sets of protocol options are allowed:
the main one (with socket options, if any), and a shorter one, with options
being a subset of the main options, repeated for clarity. As long as the
shorter set of protocol options is used, all listen directives except the
main one should use it.
Now "listen" directve has a new "quic" parameter which enables QUIC protocol
for the address. Further, to enable HTTP/3, a new directive "http3" is
introduced. The hq-interop protocol is enabled by "http3_hq" as before.
Now application protocol is chosen by ALPN.
Previously used "http3" parameter of "listen" is deprecated.
A QUIC handshake failure breaks down into several cases:
- a handshake error which leads to a send_alert call
- an error triggered by the add_handshake_data callback
- internal errors (allocation etc)
Previously, in the first case, only error code was set in the send_alert
callback. Now the "handshake failed" reason phrase is set there as well.
In the second case, both code and reason are set by add_handshake_data.
In the last case, setting reason phrase is removed: returning NGX_ERROR
now leads to closing the connection with just INTERNAL_ERROR.
Reported by Jiuzhou Cui.
Previously, since 3550b00d9dc8, the token was allocated on stack, to get
rid of pool usage. Now the token is allocated by ngx_quic_copy_buffer()
in QUIC buffers, also used for STREAM, CRYPTO and ACK frames.
Previously, location prefix length in ngx_http_location_tree_node_t was
stored as "u_char", and therefore location prefixes longer than 255 bytes
were handled incorrectly.
Fix is to use "u_short" instead. With "u_short", prefixes up to 65535 bytes
can be safely used, and this isn't reachable due to NGX_CONF_BUFFER, which
is 4096 bytes.
In contrast to on-the-fly gzipping with gzip filter, static gzipped
representation as returned by gzip_static is persistent, and therefore
the same binary representation is available for future requests, making
it possible to use range requests.
Further, if a gzipped representation is re-generated with different
compression settings, it is expected to result in different ETag and
different size reported in the Content-Range header, making it possible
to safely use range requests anyway.
As such, ranges are now allowed for files returned by gzip_static.
In nginx source code the inttypes.h include, if available, is used to define
standard integer types. Changed the SO_COOKIE configure test to follow this.
Specifically, now it is kept unset until streams are initialized.
Notably, this unbreaks OCSP with client certificates after 35e27117b593.
Previously, the read event could be posted prematurely via ngx_quic_set_event()
e.g., as part of handling a STREAM frame.
Previously, streams were initialized in early keys handler. However, client
transport parameters may not be available by then. This happens, for example,
when using QuicTLS. Now streams are initialized in ngx_quic_crypto_input()
after calling SSL_do_handshake() for both 0-RTT and 1-RTT.
Previously, there was no timeout for a request stream blocked on insert count,
which could result in infinite wait. Now client_header_timeout is set when
stream is first blocked.
Now, when RESET_STREAM is sent or received, or when streams are closed,
stream connection error flag is set. Previously, only stream state was
changed, which resulted in setting the error flag only after calling
recv()/send()/send_chain(). However, there are cases when none of these
functions is called, but it's still important to know if the stream is being
closed. For example, when an HTTP/3 request stream is blocked on insert count,
receiving RESET_STREAM should trigger stream closure, which was not the case.
The change also fixes ngx_http_upstream_check_broken_connection() and
ngx_http_test_reading() with QUIC streams.
Previously, stream events were added and deleted by ngx_handle_read_event() and
ngx_handle_write_event() in a way similar to level-triggered events. However,
QUIC stream events are effectively edge-triggered and can stay active all time.
Moreover, the events are now active since the moment a stream is created.
Previously, start_time wasn't set for a new stream.
The fix is to derive it from the parent connection.
Also it's used to simplify tracking keepalive_time.
As per RFC 9204, section 3.2.2, a new entry can reference an entry in the
dynamic table that will be evicted when adding this new entry into the dynamic
table.
Previously, such inserts resulted in use-after-free since the old entry was
evicted before the insertion (ticket #2431). Now it's evicted after the
insertion.
This change fixes Insert with Name Reference and Duplicate encoder instructions.
Ports difference must be respected when checking addresses for duplicates,
otherwise configurations like this are broken:
listen 127.0.0.1:6000-6005
It was broken by 4cc2bfeff46c (nginx 1.23.3).
Fixed event flags handling edge cases in ngx_wsarecv() and ngx_wsarecv_chain(),
notably to always reset rev->ready in case of errors (which wasn't the case
after ngx_socket_nread() errors), and after EOF (rev->ready was not cleared
if due to a misconfiguration a zero-sized buffer was used for reading).
With this change, behaviour of ngx_ssl_recv() now matches ngx_unix_recv(),
which used to always reset c->read->ready to 0 when returning errors.
This fixes an infinite loop in unbuffered SSL proxying if writing to the
client is blocked and an SSL error happens (ticket #2418).
With this change, the fix for a similar issue in the stream module
(6868:ee3645078759), which used a different approach of explicitly
testing c->read->error instead, is no longer needed and was reverted.
Casts are believed to be not needed, since memcmp() has "const void *"
arguments since introduction of the "void" type in C89. And on pre-C89
platforms nginx is unlikely to compile without warnings anyway, as there
are no casts in memcpy() and memmove() calls.
These casts were added in 1648:89a47f19b9ec without any details on why they
were added, and Igor does not remember details either. The most plausible
explanation is that they were copied from ngx_strcmp() and were not really
needed even at that time.
Prodded by Alejandro Colomar.
Binary upgrades are not supported without master process, but it is,
however, possible, that nginx running with master process is asked
to upgrade binary, and the configuration file as available on disk
at this time includes "master_process off;".
If this happens, listening sockets inherited from the previous binary
will have ls[i].previous set. But the old cycle on initial process
startup, including startup after binary upgrade, is destroyed by
ngx_init_cycle() once configuration parsing is complete. As a result,
an attempt to dereference ls[i].previous in ngx_event_process_init()
accesses already freed memory.
Fix is to avoid looking into ls[i].previous if the old cycle is already
freed.
With this change it is also no longer needed to clear ls[i].previous in
worker processes, so the relevant code was removed.
Cloning of listening sockets for each worker process does not make sense
when working without master process, and causes some of the connections
not to be accepted if worker_processes is set to more than one and there
are listening sockets configured with the reuseport flag. Fix is to
disable cloning when master process is disabled.
Due to the glibc bug[1], getaddrinfo("localhost") with AI_ADDRCONFIG
on a typical host with glibc and without IPv6 returns two 127.0.0.1
addresses, and therefore "listen localhost:80;" used to result in
"duplicate ... address and port pair" after 4f9b72a229c1.
Fix is to explicitly filter out duplicate addresses returned during
resolution of a name.
[1] https://sourceware.org/bugzilla/show_bug.cgi?id=14969
Previously, if an event was posted by a read event handler, called by
ngx_close_idle_connections(), that event was not processed until the next
event loop iteration, which could happen after a timeout.
As the SSI parser always uses the context from the main request for storing
variables and blocks, that context should always exist for subrequests using
SSI, even though the main request does not necessarily have SSI enabled.
However, `ngx_http_get_module_ctx(r->main, ...)` is getting NULL in such cases,
resulting in the worker crashing SIGSEGV when accessing its attributes.
This patch links the first initialized context to the main request, and
upgrades it only when main context is initialized.
The check is not expected to fail unless there is a bug in the calling
code. But given the check is here, it should log an alert if it fails
instead of silently closing the connection.
Maximum size for reading the PROXY protocol header is increased to 4096 to
accommodate a bigger number of TLVs, which are supported since cca4c8a715de.
Maximum size for writing the PROXY protocol header is not changed since only
version 1 is currently supported.
Previously, keepalive timer was deleted in ngx_http_v3_wait_request_handler()
and set in request cleanup handler. This worked for HTTP/3 connections, but not
for hq connections. Now keepalive timer is deleted in
ngx_http_v3_init_request_stream() and set in connection cleanup handler,
which works both for HTTP/3 and hq.
It's called after handshake completion or prior to the first early data stream
creation. The callback should initialize application-level data before
creating streams.
HTTP/3 callback implementation sets keepalive timer and sends SETTINGS.
Also, this allows to limit max handshake time in ngx_http_v3_init_stream().
Most atoms should not appear more than once in a container. Previously,
this was not enforced by the module, which could result in worker process
crash, memory corruption and disclosure.
Now it properly detects invalid shared zone configuration with omitted size.
Previously it used to read outside of the buffer boundary.
Found with AddressSanitizer.
OpenSSL with TLSv1.3 updates the session creation time on session
resumption and keeps the session timeout unmodified, making it possible
to maintain the session forever, bypassing client certificate expiration
and revocation. To make sure session timeouts are actually used, we
now update the session creation time and reduce the session timeout
accordingly.
BoringSSL with TLSv1.3 ignores configured session timeouts and uses a
hardcoded timeout instead, 7 days. So we update session timeout to
the configured value as soon as a session is created.
Instead of syncing keys with shared memory on each ticket operation,
the code now does this only when the worker is going to change expiration
of the current key, or going to switch to a new key: that is, usually
at most once per second.
To do so without races, the code maintains 3 keys: current, previous,
and next. If a worker will switch to the next key earlier, other workers
will still be able to decrypt new tickets, since they will be encrypted
with the next key.
As long as ssl_session_cache in shared memory is configured, session ticket
keys are now automatically generated in shared memory, and rotated
periodically. This can be beneficial from forward secrecy point of view,
and also avoids increased CPU usage after configuration reloads.
This also helps BoringSSL to properly resume sessions in configurations
with multiple worker processes and no ssl_session_ticket_key directives,
as BoringSSL tries to automatically rotate session ticket keys and does
this independently in different worker processes, thus breaking session
resumption between worker processes.
Given the present typical SSL session sizes, on 32-bit platforms it is
now beneficial to store all data in a single allocation, since rbtree
node + session id + ASN1 representation of a session takes 256 bytes of
shared memory (36 + 32 + 150 = about 218 bytes plus SNI server name).
Storing all data in a single allocation is beneficial for SNI names up to
about 40 characters long and makes it possible to store about 4000 sessions
in one megabyte (instead of about 3000 sessions now). This also slightly
simplifies the code.
Session ids are not expected to be longer than 32 bytes, but this is
theoretically possible with TLSv1.3, where session ids are essentially
arbitrary and sent as session tickets. Since on 64-bit platforms we
use fixed 32-byte buffer for session ids, added an explicit length check
to make sure the buffer is large enough.
Session cache allocations might fail as long as the new session is different
in size from the one least recently used (and freed when the first allocation
fails). In particular, it might not be possible to allocate space for
sessions with client certificates, since they are noticeably bigger than
normal sessions.
To ensure such allocation failures won't clutter logs, logging level changed
to "warn", and logging is now limited to at most one warning per second.
OpenSSL tries to save TLSv1.3 sessions into session cache even when using
tickets for stateless session resumption, "because some applications just
want to know about the creation of a session". To avoid trashing session
cache with useless data, we do not save such sessions now.
The variables have prefix $proxy_protocol_tlv_ and are accessible by name
and by type. Examples are: $proxy_protocol_tlv_0x01, $proxy_protocol_tlv_alpn.
Previously, all received user input was logged. If a multi-line text was
received from client and logged, it could reduce log readability and also make
it harder to parse nginx log by scripts. The change brings to PROXY protocol
the same behavior that exists for HTTP request line in
ngx_http_log_error_handler().
SSL_sendfile() expects integer file descriptor as an argument, but nginx
uses OS file handles (HANDLE) to work with files on Windows, and passing
HANDLE instead of an integer correctly results in build failure. Since
SSL_sendfile() is not expected to work on Windows anyway, the code is now
disabled on Windows with appropriate compile-time checks.
Multiple C4306 warnings (conversion from 'type1' to 'type2' of greater size)
appear during 64-bit compilation with MSVC 2010 (and older) due to extensively
used constructs like "(void *) -1", so they were disabled.
In newer MSVC versions C4306 warnings were replaced with C4312 ones, and
these are not generated for such trivial type casts.
In 2014ed60f17f, "#if SSL_CTRL_SET_ECDH_AUTO" test was incorrectly used
instead of "#ifdef SSL_CTRL_SET_ECDH_AUTO". There is no practical
difference, since SSL_CTRL_SET_ECDH_AUTO evaluates to a non-zero numeric
value when defined, but anyway it's better to correctly test if the value
is defined.
All these events are created in context of a client connection and are deleted
when the connection is closed. Setting ev->cancelable could trigger premature
connection closure and a socket leak alert.
Now main QUIC connection for HTTP/3 always has c->idle flag set. This allows
the connection to receive worker shutdown notification. It is passed to
application level via a new conf->shutdown() callback.
The HTTP/3 shutdown callback sends GOAWAY to client and gracefully shuts down
the QUIC connection.
Previously, stream was kept alive until all its data is sent. This resulted
in disabling retransmission of final part of stream when QUIC connection
was closed right after closing stream connection.
The connection is automatically switched to this mode by transport layer when
there are no non-cancelable streams. Currently, cancelable streams are
HTTP/3 encoder/decoder/control streams.
Previously, ngx_quic_finalize_connection() closed the connection with NGX_ERROR
code, which resulted in immediate connection closure. Now the code is NGX_OK,
which provides a more graceful shutdown with a timeout.
Previously, zero was used for this purpose. However, NGX_QUIC_ERR_NO_ERROR is
zero too. As a result, NGX_QUIC_ERR_NO_ERROR was changed to
NGX_QUIC_ERR_INTERNAL_ERROR when closing a QUIC connection.
If a client packet carrying a stream data frame is not acked due to packet loss,
the stream data is retransmitted later by client. It's also possible that the
retransmitted range is bigger than before due to more stream data being
available by then. If the original data was read out by the application,
there would be no read event triggered by the retransmitted frame, even though
it contains new data.
They are not supported by MSVC till 2012.
SSL_QUIC_METHOD initialization is moved to run-time to preserve portability
among SSL library implementations, which allows to reduce its visibility.
Note using of a static storage to keep SSL_set_quic_method() reference valid.
Previously, ngx_quic_hkdf_t variables used declaration with assignment
in the middle of a function, which is not supported by MSVC 2010.
Fixing this also required to rewrite the ngx_quic_hkdf_set macro
and to switch to an explicit array size.
Previously, HTTP/3 stream connection didn't inherit the servername regex
from the main QUIC connection saved when processing SNI and using regular
expressions in server names. As a result, it didn't execute to set regex
captures when choosing the virtual server while parsing HTTP/3 headers.
The type field was added in 7999d3fbb765 at early stages of QUIC implementation
and was not initialized for default listen. Missing initialization resulted in
default listen socket creation error.
SSL_CIPHER_get_protocol_id() appeared in BoringSSL somewhere between
BORINGSSL_API_VERSION 12 and 13 for compatibility with OpenSSL 1.1.1.
It was adopted without a proper macro test, which remained unnoticed.
This justifies that such old BoringSSL API isn't widely used and its
support can be dropped.
While here, removed SSL_set_quic_use_legacy_codepoint() that became
useless after the default was flipped in BoringSSL over a year ago.
Setting QUIC methods is converted to use C99 designated initializers
for simplicity, as LibreSSL 3.6.0 has different SSL_QUIC_METHOD layout.
Additionally, only set_read_secret/set_write_secret callbacks are set.
Although they are preferred in LibreSSL over set_encryption_secrets,
better be on a safe side as LibreSSL has unexpectedly incompatible
set_encryption_secrets calling convention expressed in passing read
and write secrets split in separate calls, unlike this is documented
in old BoringSSL sources. To avoid introducing further changes for
the old API, it is simply disabled.
This function is present in QuicTLS only. After SSL_READ_EARLY_DATA_SUCCESS
became visible in LibreSSL together with experimental QUIC API, this required
to revise the conditional compilation test to use more narrow macros.
After BoringSSL aligned[1] with OpenSSL on TLS1_3_CK_* macros, and
LibreSSL uses OpenSSL naming, our own variants can be dropped now.
Compatibility is preserved with libraries that lack these macros.
Additionally, transition to SSL_CIPHER_get_id() fixes build error
with LibreSSL that doesn't implement SSL_CIPHER_get_protocol_id().
[1] https://boringssl.googlesource.com/boringssl/+/dfddbc4ded
The SSL_R_BAD_RECORD_TYPE ("bad record type") errors are reported by
OpenSSL 1.1.1 or newer when using TLSv1.3 if the client sends a record
with unknown or unexpected type. These errors are now logged at the
"info" level.
When client DATA frame header and its content come in different QUIC packets,
it may happen that only the header is processed by the first
ngx_http_v3_request_body_filter() call. In this case an empty request body
buffer is added to r->request_body->bufs, which is later reused in a
subsequent ngx_http_v3_request_body_filter() call without being removed from
the body chain. As a result, rb->request_body->bufs ends up with two copies of
the same buffer.
The fix is to avoid adding empty request body buffers to r->request_body->bufs.
When reading exactly rev->available bytes, rev->available might become 0
after FIONREAD usage introduction in efd71d49bde0. On the next call of
ngx_readv_chain() on systems with EPOLLRDHUP this resulted in return without
any actions, that is, with rev->ready set, and this in turn resulted in no
timers set in event pipe, leading to socket leaks.
Fix is to reset rev->ready in ngx_readv_chain() when returning due to
rev->available being 0 with EPOLLRDHUP, much like it is already done in
ngx_unix_recv(). This ensures that if rev->available will become 0, on
systems with EPOLLRDHUP support appropriate EPOLLRDHUP-specific handling
will happen on the next ngx_readv_chain() call.
While here, also synced ngx_readv_chain() to match ngx_unix_recv() and
reset rev->ready when returning due to rev->available being 0 with kqueue.
This is mostly cosmetic change, as rev->ready is anyway reset when
rev->available is set to 0.
Some servers might emit Content-Range header on 200 responses, and this
does not seem to contradict RFC 9110: as per RFC 9110, the Content-Range
header has no meaning for status codes other than 206 and 416. Previously
this resulted in duplicate Content-Range headers in nginx responses handled
by the range filter. Fix is to clear pre-existing headers.
Starting with OpenSSL 1.1.1, various additional errors can be reported
by OpenSSL in case of client-related issues, most notably during TLSv1.3
handshakes. In particular, SSL_R_BAD_KEY_SHARE ("bad key share"),
SSL_R_BAD_EXTENSION ("bad extension"), SSL_R_BAD_CIPHER ("bad cipher"),
SSL_R_BAD_ECPOINT ("bad ecpoint"). These are now logged at the "info"
level.
To ensure optimal use of memory, SSL contexts for proxying are now
inherited from previous levels as long as relevant proxy_ssl_* directives
are not redefined.
Further, when no proxy_ssl_* directives are redefined in a server block,
we now preserve plcf->upstream.ssl in the "http" section configuration
to inherit it to all servers.
Similar changes made in uwsgi, grpc, and stream proxy.
Similar to 70e65bf8dfd7, the change is made to ensure that the ability to
cancel resolver tasks is fully controlled by the caller. As mentioned in the
referenced commit, it is safe to make this timer cancelable because resolve
tasks can have their own timeouts that are not cancelable.
The scenario where this may become a problem is a periodic background resolve
task (not tied to a specific request or a client connection), which receives a
response with short TTL, large enough to warrant fallback to a TCP query.
With each event loop wakeup, we either have a previously set write timer
instance or schedule a new one. The non-cancelable write timer can delay or
block graceful shutdown of a worker even if the ngx_resolver_ctx_t->cancelable
flag is set by the API user, and there are no other tasks or connections.
We use the resolver API in this way to maintain the list of upstream server
addresses specified with the 'resolve' parameter, and there could be third-party
modules implementing similar logic.
When we generate the last_buf buffer for an UDP upstream recv error, it does
not contain any data from the wire. ngx_stream_write_filter attempts to forward
it anyways, which is incorrect (e.g., UDP upstream ECONNREFUSED will be
translated to an empty packet).
This happens because we mark the buffer as both 'flush' and 'last_buf', and
ngx_stream_write_filter has special handling for flush with certain types of
connections (see d127837c714f, 32b0ba4855a6). The flags are meant to be
mutually exclusive, so the fix is to ensure that flush and last_buf are not set
at the same time.
Reproduction:
stream {
upstream unreachable {
server 127.0.0.1:8880;
}
server {
listen 127.0.0.1:8998 udp;
proxy_pass unreachable;
}
}
1 0.000000000 127.0.0.1 → 127.0.0.1 UDP 47 45588 → 8998 Len=5
2 0.000166300 127.0.0.1 → 127.0.0.1 UDP 47 51149 → 8880 Len=5
3 0.000172600 127.0.0.1 → 127.0.0.1 ICMP 75 Destination unreachable (Port
unreachable)
4 0.000202400 127.0.0.1 → 127.0.0.1 UDP 42 8998 → 45588 Len=0
Fixes d127837c714f.
Both "count" and "duration" variables are 32-bit, so their product might
potentially overflow. It is used to reduce 64-bit start_time variable,
and with very large start_time this can result in incorrect seeking.
Found by Coverity (CID 1499904).
Now, if the directive is given an empty string, such configuration cancels
loading of certificates, in particular, if they would be otherwise inherited
from the previous level. This restores previous behaviour, before variables
support in certificates was introduced (3ab8e1e2f0f7).
Previously, if caching was disabled due to Expires in the past, nginx
failed to cache the response even if it was cacheable as per subsequently
parsed Cache-Control header (ticket #964).
Similarly, if caching was disabled due to Expires in the past,
"Cache-Control: no-cache" or "Cache-Control: max-age=0", caching was not
used if it was cacheable as per subsequently parsed X-Accel-Expires header.
Fix is to avoid disabling caching immediately after parsing Expires in
the past or Cache-Control, but rather set flags which are later checked by
ngx_http_upstream_process_headers() (and cleared by "Cache-Control: max-age"
and X-Accel-Expires).
Additionally, now X-Accel-Expires does not prevent parsing of cache control
extensions, notably stale-while-revalidate and stale-if-error. This
ensures that order of the X-Accel-Expires and Cache-Control headers is not
important.
Prodded by Vadim Fedorenko and Yugo Horie.
If a module adds multiple WWW-Authenticate headers (ticket #485) to the
response, linked in r->headers_out.www_authenticate, all headers are now
cleared if another module later allows access.
This change is a nop for standard modules, since the only access module which
can add multiple WWW-Authenticate headers is the auth request module, and
it is checked after other standard access modules. Though this might
affect some third party access modules.
Note that if a 3rd party module adds a single WWW-Authenticate header
and not yet modified to set the header's next pointer to NULL, attempt to
clear such a header with this change will result in a segmentation fault.
When using auth_request with an upstream server which returns 401
(Unauthorized), multiple WWW-Authenticate headers from the upstream server
response are now properly copied to the response.
When using proxy_intercept_errors and an error page for error 401
(Unauthorized), multiple WWW-Authenticate headers from the upstream server
response are now properly copied to the response.
Most of the known duplicate upstream response headers are now ignored
with a warning.
If syntax permits multiple headers, these are now properly linked to
the lists, notably Vary and WWW-Authenticate. This makes it possible
to further handle such lists where it makes sense.
With this change, duplicate Content-Length and Transfer-Encoding headers
are now rejected. Further, responses with invalid Content-Length or
Transfer-Encoding headers are now rejected, as well as responses with both
Content-Length and Transfer-Encoding.
The h->next pointer properly provided as NULL in all cases where known
output headers are added.
Note that there are 3rd party modules which might not do this, and it
might be risky to rely on this for arbitrary headers.
Since introduction of offset handling in ngx_http_upstream_copy_header_line()
in revision 573:58475592100c, the ngx_http_upstream_copy_content_encoding()
function is no longer needed, as its behaviour is exactly equivalent to
ngx_http_upstream_copy_header_line() with appropriate offset. As such,
the ngx_http_upstream_copy_content_encoding() function was removed.
Further, the u->headers_in.content_encoding field is not used anywhere,
so it was removed as well.
Further, Content-Encoding handling no longer depends on NGX_HTTP_GZIP,
as it can be used even without any gzip handling compiled in (for example,
in the charset filter).
As all known input headers are now linked lists, these are now handled
identically. In particular, this makes it possible to access properly
combined values of headers not specifically handled previously, such
as "Via" or "Connection".
The ngx_http_process_multi_header_lines() function is removed, as it is
exactly equivalent to ngx_http_process_header_line(). Similarly,
ngx_http_variable_header() is used instead of ngx_http_variable_headers().
Multi headers are now using linked lists instead of arrays. Notably,
the following fields were changed: r->headers_in.cookies (renamed
to r->headers_in.cookie), r->headers_in.x_forwarded_for,
r->headers_out.cache_control, r->headers_out.link, u->headers_in.cache_control
u->headers_in.cookies (renamed to u->headers_in.set_cookie).
The r->headers_in.cookies and u->headers_in.cookies fields were renamed
to r->headers_in.cookie and u->headers_in.set_cookie to match header names.
The ngx_http_parse_multi_header_lines() and ngx_http_parse_set_cookie_lines()
functions were changed accordingly.
With this change, multi headers are now essentially equivalent to normal
headers, and following changes will further make them equivalent.
Previously, $http_*, $sent_http_*, $sent_trailer_*, $upstream_http_*,
and $upstream_trailer_* variables returned only the first header (with
a few specially handled exceptions: $http_cookie, $http_x_forwarded_for,
$sent_http_cache_control, $sent_http_link).
With this change, all headers are returned, combined together. For
example, $http_foo variable will be "a, b" if there are "Foo: a" and
"Foo: b" headers in the request.
Note that $upstream_http_set_cookie will also return all "Set-Cookie"
headers (ticket #1843), though this might not be what one want, since
the "Set-Cookie" header does not follow the list syntax (see RFC 7230,
section 3.2.2).
The uwsgi specification states that "The uwsgi block vars represent a
dictionary/hash". This implies that no duplicate headers are expected.
Further, provided headers are expected to follow CGI specification,
which also requires to combine headers (RFC 3875, section "4.1.18.
Protocol-Specific Meta-Variables"): "If multiple header fields with
the same field-name are received then the server MUST rewrite them
as a single value having the same semantics".
SCGI specification explicitly forbids headers with duplicate names
(section "3. Request Format"): "Duplicate names are not allowed in
the headers".
Further, provided headers are expected to follow CGI specification,
which also requires to combine headers (RFC 3875, section "4.1.18.
Protocol-Specific Meta-Variables"): "If multiple header fields with
the same field-name are received then the server MUST rewrite them
as a single value having the same semantics".
FastCGI responder is expected to receive CGI/1.1 environment variables
in the parameters (see section "6.2 Responder" of the FastCGI specification).
Obviously enough, there cannot be multiple environment variables with
the same name.
Further, CGI specification (RFC 3875, section "4.1.18. Protocol-Specific
Meta-Variables") explicitly requires to combine headers: "If multiple
header fields with the same field-name are received then the server MUST
rewrite them as a single value having the same semantics".
Previously, the r->header_in->connection pointer was never set despite
being present in ngx_http_headers_in, resulting in incorrect value returned
by $r->header_in("Connection") in embedded perl.
In 7583:efd71d49bde0 (nginx 1.17.5) along with introduction of the
ioctl(FIONREAD) support proper handling of systems without EPOLLRDHUP
support in the kernel (but with EPOLLRDHUP in headers) was broken.
Before the change, rev->available was never set to 0 unless
ngx_use_epoll_rdhup was also set (that is, runtime test for EPOLLRDHUP
introduced in 6536:f7849bfb6d21 succeeded). After the change,
rev->available might reach 0 on systems without runtime EPOLLRDHUP
support, stopping further reading in ngx_readv_chain() and ngx_unix_recv().
And, if EOF happened to be already reported along with the last event,
it is not reported again by epoll_wait(), leading to connection hangs
and timeouts on such systems.
This affects Linux kernels before 2.6.17 if nginx was compiled
with newer headers, and, more importantly, emulation layers, such as
DigitalOcean's App Platform's / gVisor's epoll emulation layer.
Fix is to explicitly check ngx_use_epoll_rdhup before the corresponding
rev->pending_eof tests in ngx_readv_chain() and ngx_unix_recv().
Previously, QUIC used the existing UDP framework, which was created for UDP in
Stream. However the way QUIC connections are created and looked up is different
from the way UDP connections in Stream are created and looked up. Now these
two implementations are decoupled.
Previously, last buffer was tracked by keeping a pointer to the previous
chain link "next" field. When the previous buffer was split and then removed,
the pointer was no longer valid. Writing at this pointer resulted in broken
data chains.
Now last buffer is tracked by keeping a direct pointer to it.
The object is used instead of ngx_chain_t pointer for buffer operations like
ngx_quic_write_chain() and ngx_quic_read_chain(). These functions are renamed
to ngx_quic_write_buffer() and ngx_quic_read_buffer().
Such fatal errors are reported by OpenSSL 1.1.1, and similarly by BoringSSL,
if application data is encountered during SSL shutdown, which started to be
observed on the second SSL_shutdown() call after SSL shutdown fixes made in
09fb2135a589 (1.19.2). The error means that the client continues to send
application data after receiving the "close_notify" alert (ticket #2318).
Previously it was reported as SSL_shutdown() error of SSL_ERROR_SYSCALL.
Now ngx_quic_stream_t is decoupled from ngx_connection_t in a way that it
can persist after connection is closed by application. During this period,
server is expecting stream final size from client for correct flow control.
Also, buffered output is sent to client as more flow control credit is granted.
As shown in RFC 8446, section 2.2, Figure 3, and further specified in
section 4.6.1, BoringSSL releases session tickets in Application Data
(along with Finished) early, based on a precalculated client Finished
transcript, once client signalled early data in extensions.
Initially, frames are genereated and stored in ctx->frames.
Next, ngx_quic_output() collects frames to be sent in in ctx->sending.
On failure, ngx_quic_revert_sned() returns frames into ctx->frames.
On success, the ngx_quic_commit_send() moves ack-eliciting frames into
ctx->sent and frees non-ack-eliciting frames.
This function also updates in-flight bytes counter, so only actually sent
frames are accounted.
The counter is decremented in the following cases:
- acknowledgment is received
- packet was declared lost
- we are discarding context completely
In each of this cases frame is removed from ctx->sent queue and in-flight
counter is accordingly decremented.
The patch fixes the case of discarding context - only removing frames
from ctx->sent must be followed by in-flight bytes counter decrement,
otherwise cg->in_flight could experience type underflow.
The issue appeared in b1676cd64dc9.
The cd8018bc81a5 fixed unintended send of non-padded initial packets,
but failed to restore context properly: only processed contexts need
to be restored. As a consequence, a packet number could be restored
from uninitialized value.
Previously, the flag could be reset after send_chain() with a limit, even
though there was room for more data. The application then started waiting for
a write event notification, which never happened.
Now the wev->ready flag is only reset when flow control is exhausted.
With large http2_max_concurrent_streams or http2_max_concurrent_pushes, more
than 255 ngx_http_v2_node_t structures might be allocated, eventually leading
to h2c->closed_nodes overflow when closing corresponding streams. This will
in turn result in additional allocations in ngx_http_v2_get_node_by_id().
While mostly harmless, it can result in excessive memory usage by a HTTP/2
connection, notably in configurations with many keepalive_requests allowed.
Fix is to use ngx_uint_t for h2c->closed_nodes instead of unsigned:8.
The switch happens when received byte counter reaches stream final size.
Previously, this state was skipped. The stream went from SIZE_KNOWN to
DATA_READ when all bytes were read by application.
The change prevents STOP_SENDING frames from being sent when all data is
received from client, but not yet fully read by application.
Previously, size was calculated based on the number of input bytes processed
by the function. Now only the copied bytes are considered. This prevents
overlapping buffers from contributing twice to the overall written size.
Response headers can be buffered in the SSL buffer. But stream's fake
connection buffered flag did not reflect this, so any attempts to flush
the buffer without sending additional data were stopped by the write filter.
It does not seem to be possible to reflect this in fc->buffered though, as
we never known if main connection's c->buffered corresponds to the particular
stream or not. As such, fc->buffered might prevent request finalization
due to sending data on some other stream.
Fix is to implement handling of flush buffers when the c->need_flush_buf
flag is set, similarly to the existing last buffer handling. The same
flag is now used for UDP sockets in the stream module instead of explicit
checking of c->type.
Previously, non-padded initial packet could be sent as a result of the
following situation:
- initial queue is not empty (so padding to 1200 is required)
- handshake queue is not empty (so padding is to be added after h/s packet)
- path is limited
If serializing handshake packet would violate path limit, such packet was
omitted, and the non-padded initial packet was sent.
The fix is to avoid sending the packet at all in such case. This follows the
original intention introduced in c5155a0cb12f.
During configuration reload two cache managers might exist for a short
time. If both tried to delete the same cache node, the "ignore long locked
inactive cache entry" alert appeared in logs. Additionally,
ngx_http_file_cache_forced_expire() might be also called by worker
processes, with similar results.
Fix is to ignore cache nodes being deleted, similarly to how it is
done in ngx_http_file_cache_expire() since 3755:76e3a93821b1. This
was somehow missed in 7002:ab199f0eb8e8, when ignoring long locked
cache entries was introduced in ngx_http_file_cache_forced_expire().
- wording in log->action is adjusted to match function names.
- connection close steps are made obvious and start with "quic close" prefix:
*1 quic close initiated rc:-4
*1 quic close silent drain:0 timedout:1
*1 quic close resumed rc:-1
*1 quic close resumed rc:-1
*1 quic close resumed rc:-4
*1 quic close completed
this makes it easy to understand if particular "close" record is an initial
cause or lasting process, or the final one.
- cases of close without quic connection now logged as "packet rejected":
*14 quic run
*14 quic packet rx long flags:ec version:1
*14 quic packet rx hs len:61
*14 quic packet rx dcid len:20 00000000000002c32f60e4aa2b90a64a39dc4228
*14 quic packet rx scid len:8 81190308612cd019
*14 quic expected initial, got handshake
*14 quic packet done rc:-1 level:hs decr:0 pn:0 perr:0
*14 quic packet rejected rc:-1, cleanup connection
*14 reusable connection: 0
this makes it easy to spot early packet rejection and avoid confuse with
quic connection closing (which in fact was not even created).
- packet processing summary now uses same prefix "quic packet done rc:"
- added debug to places where packet was rejected without any reason logged
The NGX_DECLINED is replaced with NGX_DONE to match closer to return code
of ngx_quic_handle_packet() and ngx_quic_close_connection() rc argument.
The ngx_quic_close_connection() rc code is used only when quic connection
exists, thus anything goes if qc == NULL.
The ngx_quic_handle_datagram() does not return NG_OK in cases when quic
connection is not yet created.
Previously, closure detection for server-initiated uni streams was not properly
implemented. Instead, HTTP/3 code relied on QUIC code posting the read event
and setting rev->error when it needed to close the stream. Then, regular
uni stream read handler called c->recv() and received error, which closed the
stream. This was an ad-hoc solution. If, for whatever reason, the read
handler was called earlier, c->recv() would return 0, which would also close
the stream.
Now server-initiated uni streams have a separate read event handler for
tracking stream closure. The handler calls c->recv(), which normally returns
0, but may return error in case of closure.
Sending the instruction is delayed until the end of the current event cycle.
Delaying the instruction is allowed by quic-qpack-21, section 2.2.2.3.
The goal is to reduce the amount of data sent back to client by accumulating
several inserts in one instruction and sometimes not sending the instruction at
all, if Section Acknowledgement was sent just before it.
Operations like ngx_quic_open_stream(), ngx_http_quic_get_connection(),
ngx_http_v3_finalize_connection(), ngx_http_v3_shutdown_connection() used to
receive a QUIC stream connection. Now they can receive the main QUIC
connection as well. This is useful when calling them from a stream context.
This is to ease transition with oldish BoringSSL versions,
the default for SSL_set_quic_use_legacy_codepoint() has been
flipped in BoringSSL a1d3bfb64fd7ef2cb178b5b515522ffd75d7b8c5.
Chrome only uses TLS session tickets once with TLS 1.3, likely following
RFC 8446 Appendix C.4 recommendation. With OpenSSL, this works fine with
built-in session tickets, since these are explicitly renewed in case of
TLS 1.3 on each session reuse, but results in only two connections being
reused after an initial handshake when using ssl_session_ticket_key.
Fix is to always renew TLS session tickets in case of TLS 1.3 when using
ssl_session_ticket_key, similarly to how it is done by OpenSSL internally.
Line continuation as used in the syntax file might be broken if "compatible"
is set or "C" is added to cpoptions. Fix is to set the "cpoptions" option
to vim default value at script start and restore it later, see
":help use-cpo-save".
Previously, when input ended on a QUIC buffer boundary, input chain was not
advanced to the next buffer. As a result, ngx_quic_write_chain() returned
a chain with an empty buffer instead of NULL. This broke HTTP write filter,
preventing it from closing the HTTP request and eventually timing out.
Now input chain is always advanced to a buffer that has data, before checking
QUIC buffer boundary condition.
RFC 9000, 9.3. Responding to Connection Migration:
An endpoint only changes the address to which it sends packets in
response to the highest-numbered non-probing packet.
The patch extends this requirement to probing packets. Although it may
seem excessive, it helps with mitigation of reply attacks (when an off-path
attacker has copied packet with PATH_CHALLENGE and uses different
addresses to exhaust available connection ids).
The quic connection now holds active, backup and probe paths instead
of sockets. The number of migration paths is now limited and cannot
be inflated by a bad client or an attacker.
The client id is now associated with path rather than socket. This allows
to simplify processing of output and connection ids handling.
New migration abandons any previously started migrations. This allows to
free consumed client ids and request new for use in future migrations and
make progress in case when connection id limit is hit during migration.
A path now can be revalidated without losing its state.
The patch also fixes various issues with NAT rebinding case handling:
- paths are now validated (previously, there was no validation
and paths were left in limited state)
- attempt to reuse id on different path is now again verified
(this was broken in 40445fc7c403)
- former path is now validated in case of apparent migration
The function splits a buffer at given offset. The function is now
called from ngx_quic_read_chain() and ngx_quic_write_chain(), which
simplifies both functions.
After 9ae239d2547d, ngx_quic_handle_write_event() no longer runs into
ngx_send_lowat() for QUIC connections, so the check became excessive.
It is assumed that external modules operating with SO_SNDLOWAT
(I'm not aware of any) should do this check on their own.
Previously, ngx_quic_write_chain() treated each input buffer as a memory
buffer, which is not always the case. Special buffers were not skipped, which
is especially important when hitting the input byte limit.
The issue manifested itself with ngx_quic_write_chain() returning a non-empty
chain consisting of a special last_buf buffer when called from QUIC stream
send_chain(). In order for this to happen, input byte limit should be equal to
the chain length, and the input chain should end with an empty last_buf buffer.
An easy way to achieve this is the following:
location /empty {
return 200;
}
When this non-empty chain was returned from send_chain(), it signalled to the
caller that input was blocked, while in fact it wasn't. This prevented HTTP
request from finalization, which prevented QUIC from sending STREAM FIN to
the client. The QUIC stream was then reset after a timeout.
Now special buffers are skipped and send_chain() returns NULL in the case
above, which signals to the caller a successful operation.
Also, original byte limit is now passed to ngx_quic_write_chain() from
send_chain() instead of actual chain length to make sure it's never zero.
Previously, when a STREAM FIN frame with no data bytes was received after all
prior stream data were already read by the application layer, the frame was
ignored and eof was not reported to the application.
When a worker process is shutting down, keepalive is not used: this is checked
before the ngx_http_set_keepalive() call in ngx_http_finalize_connection().
Yet the "Connection: keep-alive" header was still sent, even if we know that
the worker process is shutting down, potentially resulting in additional
requests being sent to the connection which is going to be closed anyway.
While clients are expected to be able to handle asynchronous close events
(see ticket #1022), it is certainly possible to send the "Connection: close"
header instead, informing the client that the connection is going to be closed
and potentially saving some unneeded work.
With this change, we additionally check for worker process shutdown just
before sending response headers, and disable keepalive accordingly.
Linux with EPOLLEXCLUSIVE usually notifies only the process which was first
to add the listening socket to the epoll instance. As a result most of the
connections are handled by the first worker process (ticket #2285). To fix
this, we re-add the socket periodically, so other workers will get a chance
to accept connections.
The SF_NOCACHE flag, introduced in FreeBSD 11 along with the new non-blocking
sendfile() implementation by glebius@, makes it possible to use sendfile()
along with the "directio" directive.
Starting with FreeBSD 11, there is no need to use AIO operations to preload
data into cache for sendfile(SF_NODISKIO) to work. Instead, sendfile()
handles non-blocking loading data from disk by itself. It still can, however,
return EBUSY if a page is already being loaded (for example, by a different
process). If this happens, we now post an event for the next event loop
iteration, so sendfile() is retried "after a short period", as manpage
recommends.
The limit of the number of EBUSY tolerated without any progress is preserved,
but now it does not result in an alert, since on an idle system event loop
iteration might be very short and EBUSY can happen many times in a row.
Instead, SF_NODISKIO is simply disabled for one call once the limit is
reached.
With this change, sendfile(SF_NODISKIO) is now used automatically as long as
sendfile() is enabled, and no longer requires "aio on;".
It was mostly copy of the ngx_quic_listen(). Now ngx_quic_listen() no
longer generates server id and increments seqnum. Instead, the server
id is generated when the socket is created.
The ngx_quic_alloc_socket() function is renamed to ngx_quic_create_socket().
Notably, NAXSI is known to misuse ngx_regex_compile() with rc.options set
to PCRE_CASELESS | PCRE_MULTILINE. With PCRE2 support, and notably binary
compatibility changes, it is no longer possible to set PCRE[2]_MULTILINE
option without using proper interface. To facilitate correct usage,
this change adds the NGX_REGEX_MULTILINE option.
With this change, dynamic modules using nginx regex interface can be used
regardless of the variant of the PCRE library nginx was compiled with.
If a module is compiled with different PCRE library variant, in case of
ngx_regex_exec() errors it will report wrong function name in error
messages. This is believed to be tolerable, given that fixing this will
require interface changes.
The PCRE2 library is now used by default if found, instead of the
original PCRE library. If needed for some reason, this can be disabled
with the --without-pcre2 configure option.
To make it possible to specify paths to the library and include files
via --with-cc-opt / --with-ld-opt, the library is first tested without
any additional paths and options. If this fails, the pcre2-config script
is used.
Similarly to the original PCRE library, it is now possible to build PCRE2
from sources with nginx configure, by using the --with-pcre= option.
It automatically detects if PCRE or PCRE2 sources are provided.
Note that compiling PCRE2 10.33 and later requires inttypes.h. When
compiling on Windows with MSVC, inttypes.h is only available starting
with MSVC 2013. In older versions some replacement needs to be provided
("echo '#include <stdint.h>' > pcre2-10.xx/src/inttypes.h" is good enough
for MSVC 2010).
The interface on nginx side remains unchanged.
If a configuration parsing fails for some reason, ngx_regex_module_init()
is not called, and ngx_pcre_studies remained set despite the fact that
the pool it was allocated from is already freed. This might result in
a segmentation fault during runtime regular expression compilation, such
as in SSI, for example, in the single process mode, or if a worker process
died and was respawned from a master process in such an inconsistent state.
Fix is to clear ngx_pcre_studies from the pool cleanup handler (which is
anyway used to free JIT-compiled patterns).
Previously, buffer lists was used to track used buffers. Now reference
counter is used instead. The new implementation is simpler and faster with
many buffer clones.
They are replaced with ngx_quic_write_chain() and ngx_quic_read_chain().
These functions represent the API to data buffering.
The first function adds data of given size at given offset to the buffer.
Now it returns the unwritten part of the chain similar to c->send_chain().
The second function returns data of given size from the beginning of the buffer.
Its second argument and return value are swapped compared to
ngx_quic_split_bufs() to better match ngx_quic_write_chain().
Added, returned and stored data are regular ngx_chain_t/ngx_buf_t chains.
Missing data is marked with b->sync flag.
The functions are now used in both send and recv data chains in QUIC streams.
Previously, when a few bytes were send to a QUIC stream by the application, a
4K buffer was allocated for these bytes. Then a STREAM frame was created and
that entire buffer was used as data for that frame. The frame with the buffer
were in use up until the frame was acked by client. Meanwhile, when more
bytes were send to the stream, more buffers were allocated and assigned as
data to newer STREAM frames. In this scenario most buffer memory is unused.
Now the unused part of the stream output buffer is available for further
stream output while earlier parts of the buffer are waiting to be acked.
This is achieved by splitting the output buffer.
The output is always sent to the active path, which is stored in the
quic connection. There is no need to pass it in arguments.
When output has to be send to to a specific path (in rare cases, such as
path probing), a separate method exists (ngx_quic_frame_sendto()).
The path validation status and anti-amplification limit status is actually
two different variables. It is possible that validating path should not
be limited (for example, when re-validating former path).
Previously, path was considered valid during arbitrary selected 10m timeout
since validation. This is quite not what RFC 9000 says; the relevant
part is:
An endpoint MAY skip validation of a peer address if that
address has been seen recently.
The patch considers a path to be 'recently seen' if packets were received
during idle timeout. If a packet is received from the path that was seen
not so recently, such path is considered new, and anti-amplification
restrictions apply.
After creation, a client stream is added to qc->streams.uninitialized queue.
After initialization it's removed from the queue. If a stream is never
initialized, it is freed in ngx_quic_close_streams(). Stream initializer
is now set as read event handler in stream connection.
Previously qc->streams.uninitialized was used only for delayed stream
initialization.
The change makes it possible not to handle separately the case of a new stream
in stream-related frame handlers. It makes these handlers simpler since new
streams and existing streams are now handled by the same code.
With sendfile() in threads ("aio threads; sendfile on;"), client connection
can block on writing, waiting for sendfile() to complete. In HTTP/2 this
might result in the request hang, since an attempt to continue processing
in thread event handler will call request's write event handler, which
is usually stopped by ngx_http_v2_send_chain(): it does nothing if there
are no additional data and stream->queued is set. Further, HTTP/2 resets
stream's c->write->ready to 0 if writing blocks, so just fixing
ngx_http_v2_send_chain() is not enough.
Can be reproduced with test suite on Linux with:
TEST_NGINX_GLOBALS_HTTP="aio threads; sendfile on;" prove h2*.t
The following tests currently fail: h2_keepalive.t, h2_priority.t,
h2_proxy_max_temp_file_size.t, h2.t, h2_trailers.t.
Similarly, sendfile() with AIO preloading on FreeBSD can block as well,
with similar results. This is, however, harder to reproduce, especially
on modern FreeBSD systems, since sendfile() usually does not return EBUSY.
Fix is to modify ngx_http_v2_send_chain() so it actually tries to send
data to the main connection when called, and to make sure that
c->write->ready is set by the relevant event handlers.
With sendfile in threads, "task already active" alerts might appear in logs
if a write event happens on the main HTTP/2 connection, triggering a sendfile
in threads while another thread operation is already running. Observed
with "aio threads; aio_write on; sendfile on;" and with thread event handlers
modified to post a write event to the main HTTP/2 connection (though can
happen without any modifications).
Similarly, sendfile() with AIO preloading on FreeBSD can trigger duplicate
aio operation, resulting in "second aio post" alerts. This is, however,
harder to reproduce, especially on modern FreeBSD systems, since sendfile()
usually does not return EBUSY.
Fix is to avoid starting a sendfile operation if other thread operation
is active by checking r->aio in the thread handler (and, similarly, in
aio preload handler). The added check also makes duplicate calls protection
redundant, so it is removed.
The SSL_OP_ENABLE_MIDDLEBOX_COMPAT option is provided by QuicTLS and enabled
by default in the newly created SSL contexts. SSL_set_quic_method() is used
to clear it, which is required for SSL handshake to work on QUIC connections.
Switching context in the ngx_http_ssl_servername() SNI callback overrides SSL
options from the new SSL context. This results in the option set again.
Fix is to explicitly clear it when switching to another SSL context.
Initially reported here (in Russian):
http://mailman.nginx.org/pipermail/nginx-ru/2021-November/063989.html
ngx_http_v3_tables.h and ngx_http_v3_tables.c are renamed to
ngx_http_v3_table.h and ngx_http_v3_table.c to better match HTTP/2 code.
ngx_http_v3_streams.h and ngx_http_v3_streams.c are renamed to
ngx_http_v3_uni.h and ngx_http_v3_uni.c to better match their content.
Directives that set transport parameters are removed from the configuration.
Corresponding values are derived from the quic configuration or initialized
to default. Whenever possible, quic configuration parameters are taken from
higher-level protocol settings, i.e. HTTP/3.
A new variable $http3 is added. The variable equals to "h3" for HTTP/3
connections, "hq" for hq connections and is an empty string otherwise.
The variable $quic is eliminated.
The new variable is similar to $http2 variable.
RFC 9000 19.16
The sequence number specified in a RETIRE_CONNECTION_ID frame MUST NOT
refer to the Destination Connection ID field of the packet in which the
frame is contained.
Before the patch, the RETIRE_CONNECTION_ID frame was sent before switching
to the new client id. If retired client id was currently in use, this lead
to violation of the spec.
The c->udp->dgram may be NULL only if the quic connection was just
created: the ngx_event_udp_recvmsg() passes information about datagrams
to existing connections by providing information in c->udp.
If case of a new connection, c->udp is allocated by the QUIC code during
creation of quic connection (it uses c->sockaddr to initialize qsock->path).
Thus the check for qsock->path is excessive and can be read wrong, assuming
that other options possible, leading to warnings from clang static analyzer.
Removed sending CLOSE_CONNECTION directly to avoid duplicate frames,
since it is sent later again in SSL_do_handshake() error handling.
As such, removed redundant settings of error fields set elsewhere.
While here, improved debug message.
All open sockets are stored in a queue. There is no need to close some
of them separately. If it happens that active and backup point to same
socket, double close may happen (leading to possible segfault).
The RFC 9000 allows a packet from known CID arrive from unknown path:
These requirements regarding connection ID reuse apply only to the
sending of packets, as unintentional changes in path without a change
in connection ID are possible. For example, after a period of
network inactivity, NAT rebinding might cause packets to be sent on a
new path when the client resumes sending.
Before the patch, such packets were rejected with an error in the
ngx_quic_check_migration() function. Removing the check makes the
separate function excessive - remaining checks are early migration
check and "disable_active_migration" check. The latter is a transport
parameter sent to client and it should not be used by server.
The server should send "disable_active_migration" "if the endpoint does
not support active connection migration" (18.2). The support status depends
on nginx configuration: to have migration working with multiple workers,
you need bpf helper, available on recent Linux systems. The patch does
not set "disable_active_migration" automatically and leaves it for the
administrator. By default, active migration is enabled.
RFC 900 says that it is ok to migrate if the peer violates
"disable_active_migration" flag requirements:
If the peer violates this requirement,
the endpoint MUST either drop the incoming packets on that path without
generating a Stateless Reset
OR
proceed with path validation and allow the peer to migrate. Generating a
Stateless Reset or closing the connection would allow third parties in the
network to cause connections to close by spoofing or otherwise manipulating
observed traffic.
So, nginx adheres to the second option and proceeds to path validation.
Note:
The ngtcp2 may be used for testing both active migration and NAT rebinding:
ngtcp2/client --change-local-addr=200ms --delay-stream=500ms <ip> <port> <url>
ngtcp2/client --change-local-addr=200ms --delay-stream=500ms --nat-rebinding \
<ip> <port> <url>
Single UDP datagram may contain multiple QUIC datagrams. In order to
facilitate handling of such cases, 'first' flag in the ngx_quic_header_t
structure is introduced.
Previously, the retired socket was not closed if it didn't match
active or backup.
New sockets could not be created (due to count limit), since retired socket
was not closed before calling ngx_quic_create_sockets().
When replacing retired socket, new socket is only requested after closing
old one, to avoid hitting the limit on the number of active connection ids.
Together with added restrictions, this fixes an issue when a current socket
could be closed during migration, recreated and erroneously reused leading
to null pointer dereference.
Previously the frame was not handled and connection was closed with an error.
Now, after receiving this frame, global flow control is updated and new
flow control credit is sent to client.
Previously, after receiving STREAM_DATA_BLOCKED, current flow control limit
was sent to client. Now, if the limit can be updated to the full window size,
it is updated and the new value is sent to client, otherwise nothing is sent.
The change lets client update flow control credit on demand. Also, it saves
traffic by not sending MAX_STREAM_DATA with the same value twice.
The reasons why a stream may not be created by server currently include hitting
worker_connections limit and memory allocation error. Previously in these
cases the entire QUIC connection was closed and all its streams were shut down.
Now the new stream is rejected and existing streams continue working.
To reject an HTTP/3 request stream, RESET_STREAM and STOP_SENDING with
H3_REQUEST_REJECTED error code are sent to client. HTTP/3 uni streams and
Stream streams are not rejected.
The variable contains a negotiated curve used for the handshake key
exchange process. Known curves are listed by their names, unknown
ones are shown in hex.
Note that for resumed sessions in TLSv1.2 and older protocols,
$ssl_curve contains the curve used during the initial handshake,
while in TLSv1.3 it contains the curve used during the session
resumption (see the SSL_get_negotiated_group manual page for
details).
The variable is only meaningful when using OpenSSL 3.0 and above.
With older versions the variable is empty.
Without this change, aio used with HTTP/2 can result in connection hang,
as observed with "aio threads; aio_write on;" and proxying (ticket #2248).
The problem is that HTTP/2 updates buffers outside of the output filters
(notably, marks them as sent), and then posts a write event to call
output filters. If a filter does not call the next one for some reason
(for example, because of an AIO operation in progress), this might
result in a state when the owner of a buffer already called
ngx_chain_update_chains() and can reuse the buffer, while the same buffer
is still sitting in the busy chain of some other filter.
In the particular case a buffer was sitting in output chain's ctx->busy,
and was reused by event pipe. Output chain's ctx->busy was permanently
blocked by it, and this resulted in connection hang.
Fix is to change ngx_chain_update_chains() to skip buffers from other
modules unconditionally, without trying to wait for these buffers to
become empty.
The "sendfile_max_chunk" directive is important to prevent worker
monopolization by fast connections. The 2m value implies maximum 200ms
delay with 100 Mbps links, 20ms delay with 1 Gbps links, and 2ms on
10 Gbps links. It also seems to be a good value for disks.
Previously, connections to upstream servers used sendfile() if it was
enabled, but never honored sendfile_max_chunk. This might result
in worker monopolization for a long time if large request bodies
are allowed.
On Linux starting with 2.6.16, sendfile() silently limits all operations
to MAX_RW_COUNT, defined as (INT_MAX & PAGE_MASK). This incorrectly
triggered the interrupt check, and resulted in 0-sized writev() on the
next loop iteration.
Fix is to make sure the limit is always checked, so we will return from
the loop if the limit is already reached even if number of bytes sent is
not exactly equal to the number of bytes we've tried to send.
Previously, it was checked that sendfile_max_chunk was enabled and
almost whole sendfile_max_chunk was sent (see e67ef50c3176), to avoid
delaying connections where sendfile_max_chunk wasn't reached (for example,
when sending responses smaller than sendfile_max_chunk). Now we instead
check if there are unsent data, and the connection is still ready for writing.
Additionally we also check c->write->delayed to ignore connections already
delayed by limit_rate.
This approach is believed to be more robust, and correctly handles
not only sendfile_max_chunk, but also internal limits of c->send_chain(),
such as sendfile() maximum supported length (ticket #1870).
Previously, 1 millisecond delay was used instead. In certain edge cases
this might result in noticeable performance degradation though, notably on
Linux with typical CONFIG_HZ=250 (so 1ms delay becomes 4ms),
sendfile_max_chunk 2m, and link speed above 2.5 Gbps.
Using posted next events removes the artificial delay and makes processing
fast in all cases.
The directive enables including all frames from start time to the most recent
key frame in the result. Those frames are removed from presentation timeline
using mp4 edit lists.
Edit lists are currently supported by popular players and browsers such as
Chrome, Safari, QuickTime and ffmpeg. Among those not supporting them properly
is Firefox[1].
Based on a patch by Tracey Jaquith, Internet Archive.
[1] https://bugzilla.mozilla.org/show_bug.cgi?id=1735300
The function updates the duration field of mdhd atom. Previously it was
updated in ngx_http_mp4_read_mdhd_atom(). The change makes it possible to
alter track duration as a result of processing track frames.
As per quic-qpack-21:
When a stream is reset or reading is abandoned, the decoder emits a
Stream Cancellation instruction.
Previously the instruction was not sent. Now it's sent when closing QUIC
stream connection if dynamic table capacity is non-zero and eof was not
received from client. The latter condition means that a trailers section
may still be on its way from client and the stream needs to be cancelled.
A QUIC stream connection is treated as reusable until first bytes of request
arrive, which is also when the request object is now allocated. A connection
closed as a result of draining, is reset with the error code
H3_REQUEST_REJECTED. Such behavior is allowed by quic-http-34:
Once a request stream has been opened, the request MAY be cancelled
by either endpoint. Clients cancel requests if the response is no
longer of interest; servers cancel requests if they are unable to or
choose not to respond.
When the server cancels a request without performing any application
processing, the request is considered "rejected." The server SHOULD
abort its response stream with the error code H3_REQUEST_REJECTED.
The client can treat requests rejected by the server as though they had
never been sent at all, thereby allowing them to be retried later.
When an HTTP/3 function returns an error in context of a QUIC stream, it's
this function's responsibility now to finalize the entire QUIC connection
with the right code, if required. Previously, QUIC connection finalization
could be done both outside and inside such functions. The new rule follows
a similar rule for logging, leads to cleaner code, and allows to provide more
details about the error.
While here, a few error cases are no longer treated as fatal and QUIC connection
is no longer finalized in these cases. A few other cases now lead to
stream reset instead of connection finalization.
The PATH_RESPONSE frame must be expanded to 1200, except the case
when anti-amplification limit is in effect, i.e. on unvalidated paths.
Previously, the anti-amplification limit was always applied.
If client ID was never used, its refcount is zero. To keep things simple,
the ngx_quic_unref_client_id() function is now aware of such IDs.
If client ID was used, the ngx_quic_replace_retired_client_id() function
is supposed to find all users and unref the ID, thus ngx_quic_unref_client_id()
should not be called after it.
Previously, it was not enforced in the stream module.
Now, since b9e02e9b2f1d it is possible to specify protocols.
Since ALPN is always required, the 'require_alpn' setting is now obsolete.
The "min" and "max" arguments refer to UDP datagram size. Generating payload
requires to account properly for header size, which is variable and depends on
payload size and packet number.
After fe919fd63b0b, processing QUIC streams was postponed until after handshake
completion, which means that 0-RTT is effectively off. With ssl_ocsp enabled,
it could be further delayed. This differs from how OCSP validation works with
SSL_read_early_data(). With this change, processing QUIC streams is unlocked
when obtaining 0-RTT secret.
The sent queue is sorted by packet number. It is possible to avoid
traversing full queue while handling ack ranges. It makes sense to
start traversing from the queue head (i.e. check oldest packets first).
With this patch, all traffic over a QUIC connection is compared to traffic
over QUIC streams. As long as total traffic is many times larger than stream
traffic, we consider this to be a flood.
With this patch, all traffic over HTTP/3 bidi and uni streams is counted in
the h3c->total_bytes field, and payload traffic is counted in the
h3c->payload_bytes field. As long as total traffic is many times larger than
payload traffic, we consider this to be a flood.
Request header traffic is counted as if all fields are literal. Response
header traffic is counted as is.
Checking the reset after encryption avoids false positives. More importantly,
it avoids the check entirely in the usual case where decryption succeeds.
RFC 9000, 10.3.1 Detecting a Stateless Reset
Endpoints MAY skip this check if any packet from a datagram is
successfully processed.
As per RFC 9000:
An endpoint that receives a STOP_SENDING frame MUST send a RESET_STREAM
frame if the stream is in the "Ready" or "Send" state.
An endpoint SHOULD copy the error code from the STOP_SENDING frame to
the RESET_STREAM frame it sends, but it can use any application error code.
The flag indicates that the entire response was sent to the socket up to the
last_buf flag. The flag is only usable for protocol implementations that call
ngx_http_write_filter() from header filter, such as HTTP/1.x and HTTP/3.
Similar to the previous change, a segmentation fault occurres when evaluating
SSL certificates on a QUIC connection due to an uninitialized stream session.
The fix is to adjust initializing the QUIC part of a connection until after
it has session and variables initialized.
Similarly, this appends logging error context for QUIC connections:
- client 127.0.0.1:54749 connected to 127.0.0.1:8880 while handling frames
- quic client timed out (60: Operation timed out) while handling quic input
A QUIC connection doesn't have c->log->data and friends initialized to sensible
values. Yet, a request can be created in the certificate callback with such an
assumption, which leads to a segmentation fault due to null pointer dereference
in ngx_http_free_request(). The fix is to adjust initializing the QUIC part of
a connection such that it has all of that in place.
Further, this appends logging error context for unsuccessful QUIC handshakes:
- cannot load certificate .. while handling frames
- SSL_do_handshake() failed .. while sending frames
OpenSSL library QUIC support cannot be tested at configure time when
using the --with-openssl option so assume it's present if requested.
While here, fixed the error message in case QUIC support is missing.
Previously the counter was not incremented for HTTP/3 streams, but still
decremented in ngx_http_close_connection(). There are two solutions here, one
is to increment the counter for HTTP/3 streams, and the other one is not to
decrement the counter for HTTP/3 streams. The latter solution looks
inconsistent with ngx_stat_reading/ngx_stat_writing, which are incremented on a
per-request basis. The change adds ngx_stat_active increment for HTTP/3
request and push streams.
Notably, it is to avoid setting the TCP_NODELAY flag for QUIC streams
in ngx_http_upstream_send_response(). It is an invalid operation on
inherently SOCK_DGRAM sockets, which leads to QUIC connection close.
The change reduces diff to the default branch in stream content phase.
This function was only referenced from ngx_http_v3_create_push_request() to
initialize push connection log. Now the log handler is copied from the parent
request connection.
The change reduces diff to the default branch.
The functions ngx_quic_handle_read_event() and ngx_quic_handle_write_event()
are added. Previously this code was a part of ngx_handle_read_event() and
ngx_handle_write_event().
The change simplifies ngx_handle_read_event() and ngx_handle_write_event()
by moving QUIC-related code to a QUIC source file.
Previously it had -1 as fd. This fixes proxying, which relies on downstream
connection having a real fd. Also, this reduces diff to the default branch for
ngx_close_connection().
The request body filter chain is no longer called after processing
a DATA frame. Instead, we now post a read event to do this. This
ensures that multiple small DATA frames read during the same event loop
iteration are coalesced together, resulting in much faster processing.
Since rb->buf can now contain unprocessed data, window update is no
longer sent in ngx_http_v2_state_read_data() in case of flow control
being used due to filter buffering. Instead, window will be updated
by ngx_http_v2_read_client_request_body_handler() in the posted read
event.
Following rb->filter_need_buffering changes, request body reading is
only finished after the filter chain is called and rb->last_saved is set.
As such, with r->request_body_no_buffering, timer on fc->read is no
longer removed when the last part of the body is received, potentially
resulting in incorrect behaviour.
The fix is to call ngx_http_v2_process_request_body() from the
ngx_http_v2_read_unbuffered_request_body() function instead of
directly calling ngx_http_v2_filter_request_body(), so the timer
is properly removed.
In the body read handler, the window was incorrectly calculated
based on the full buffer size instead of the amount of free space
in the buffer. If the request body is buffered by a filter, and
the buffer is not empty after the read event is generated by the
filter to resume request body processing, this could result in
"http2 negative window update" alerts.
Further, in the body ready handler and in ngx_http_v2_state_read_data()
the buffer wasn't cleared when the data were already written to disk,
so the client might stuck without window updates.
If a MAX_DATA frame was received before any stream was created, then the worker
process would crash in nginx_quic_handle_max_data_frame() while traversing the
stream tree. The issue is solved by adding a check that makes sure the tree is
not empty.
If a filter wants to buffer the request body during reading (for
example, to check an external scanner), it can now do so. To make
it possible, the code now checks rb->last_saved (introduced in the
previous change) along with rb->rest == 0.
Since in HTTP/2 this requires flow control to avoid overflowing the
request body buffer, so filters which need buffering have to set
the rb->filter_need_buffering flag on the first filter call. (Note
that each filter is expected to call the next filter, so all filters
will be able set the flag if needed.)
It indicates that the last buffer was received by the save filter,
and can be used to check this at higher levels. To be used in the
following changes.
If due to an error ngx_http_request_body_save_filter() is called
more than once with rb->rest == 0, this used to result in a segmentation
fault. Added an alert to catch such errors, just in case.
Previously, fully preread unbuffered requests larger than client body
buffer size were saved to disk, despite the fact that "unbuffered" is
expected to imply no disk buffering.
The save body filter saves the request body to disk once the buffer is full.
Yet in HTTP/2 this might happen even if there is no need to save anything
to disk, notably when content length is known and the END_STREAM flag is
sent in a separate empty DATA frame. Workaround is to provide additional
byte in the buffer, so saving the request body won't be triggered.
This fixes unexpected request body disk buffering in HTTP/2 observed after
the previous change when content length is known and the END_STREAM flag
is sent in a separate empty DATA frame.
In particular, now the code always uses a buffer limited by
client_body_buffer_size. At the cost of an additional copy it
ensures that small DATA frames are not directly mapped to small
write() syscalls, but rather buffered in memory before writing.
Further, requests without Content-Length are no longer forced
to use temporary files.
With SSL it is possible that an established connection is ready for
reading after the handshake. Further, events might be already disabled
in case of level-triggered event methods. If this happens and
ngx_http_upstream_send_request() blocks waiting for some data from
the upstream, such as flow control in case of gRPC, the connection
will time out due to no read events on the upstream connection.
Fix is to explicitly check the c->read->ready flag if sending request
blocks and post a read event if it is set.
Note that while it is possible to modify ngx_ssl_handshake() to keep
read events active, this won't completely resolve the issue, since
there can be data already received during the SSL handshake
(see 573bd30e46b4).
Hash initialization ignores elements with key.data set to NULL.
Nevertheless, the initial hash bucket size check didn't skip them,
resulting in unnecessary restrictions on, for example, variables with
long names and with the NGX_HTTP_VARIABLE_NOHASH flag.
Fix is to update the initial hash bucket size check to skip elements
with key.data set to NULL, similarly to how it is done in other parts
of the code.
Requires OpenSSL 3.0 compiled with "enable-ktls" option. Further, KTLS
needs to be enabled in kernel, and in OpenSSL, either via OpenSSL
configuration file or with "ssl_conf_command Options KTLS;" in nginx
configuration.
On FreeBSD, kernel TLS is available starting with FreeBSD 13.0, and
can be enabled with "sysctl kern.ipc.tls.enable=1" and "kldload ktls_ocf"
to load a software backend, see man ktls(4) for details.
On Linux, kernel TLS is available starting with kernel 4.13 (at least 5.2
is recommended), and needs kernel compiled with CONFIG_TLS=y (with
CONFIG_TLS=m, which is used at least on Ubuntu 21.04 by default,
the tls module needs to be loaded with "modprobe tls").
While clock_gettime(CLOCK_MONOTONIC_COARSE) is faster than
clock_gettime(CLOCK_MONOTONIC), the latter is fast enough on Linux for
practical usage, and the difference is negligible compared to other costs
at each event loop iteration. On the other hand, CLOCK_MONOTONIC_COARSE
causes various issues with typical CONFIG_HZ=250, notably very inaccurate
limit_rate handling in some edge cases (ticket #1678) and negative difference
between $request_time and $upstream_response_time (ticket #1965).
This is a recommended behavior by RFC 7301 and is useful
for mitigation of protocol confusion attacks [1].
To avoid possible negative effects, list of supported protocols
was extended to include all possible HTTP protocol ALPN IDs
registered by IANA [2], i.e. "http/1.0" and "http/0.9".
[1] https://alpaca-attack.com/
[2] https://www.iana.org/assignments/tls-extensiontype-values/
In b87b7092cedb (nginx 1.21.1), logging level of "upstream sent invalid
header" errors was accidentally changed to "info". This change restores
the "error" level, which is a proper logging level for upstream-side
errors.
The u->keepalive flag is initialized early if the response has no body
(or an empty body), and needs to be reset if there are any extra data,
similarly to how it is done in ngx_http_proxy_copy_filter(). Missed
in 83c4622053b0.
The "proxy_half_close" directive enables handling of TCP half close. If
enabled, connection to proxied server is kept open until both read ends get
EOF. Write end shutdown is properly transmitted via proxy.
Do this only when the entire request body is empty and
r->request_body_in_file_only is set.
The issue manifested itself with missing warning "a client request body is
buffered to a temporary file" when the entire rb->buf is full and all buffers
are delayed by a filter.
This adds new Auth-SSL-Protocol and Auth-SSL-Cipher headers to
the mail proxy auth protocol when SSL is enabled.
This can be useful for detecting users using older clients that
negotiate old ciphers when you want to upgrade to newer
TLS versions of remove suppport for old and insecure ciphers.
You can use your auth backend to notify these users before the
upgrade that they either need to upgrade their client software
or contact your support team to work out an upgrade path.
To load old/weak server or client certificates it might be needed to adjust
the security level, as introduced in OpenSSL 1.1.0. This change ensures that
ciphers are set before loading the certificates, so security level changes
via the cipher string apply to certificate loading.
Export ciphers are forbidden to negotiate in TLS 1.1 and later protocol modes.
They are disabled since OpenSSL 1.0.2g by default unless explicitly configured
with "enable-weak-ssl-ciphers", and completely removed in OpenSSL 1.1.0.
A new behaviour was introduced in OpenSSL 1.1.1e, when a peer does not send
close_notify before closing the connection. Previously, it was to return
SSL_ERROR_SYSCALL with errno 0, known since at least OpenSSL 0.9.7, and is
handled gracefully in nginx. Now it returns SSL_ERROR_SSL with a distinct
reason SSL_R_UNEXPECTED_EOF_WHILE_READING ("unexpected eof while reading").
This leads to critical errors seen in nginx within various routines such as
SSL_do_handshake(), SSL_read(), SSL_shutdown(). The behaviour was restored
in OpenSSL 1.1.1f, but presents in OpenSSL 3.0 by default.
Use of the SSL_OP_IGNORE_UNEXPECTED_EOF option added in OpenSSL 3.0 allows
to set a compatible behaviour to return SSL_ERROR_ZERO_RETURN:
https://git.openssl.org/?p=openssl.git;a=commitdiff;h=09b90e0
See for additional details: https://github.com/openssl/openssl/issues/11381
The OPENSSL_SUPPRESS_DEPRECATED macro is used to suppress deprecation warnings.
This covers Session Tickets keys, SSL Engine, DH low level API for DHE ciphers.
Unlike OPENSSL_API_COMPAT, it works well with OpenSSL built with no-deprecated.
In particular, it doesn't unhide various macros in OpenSSL includes, which are
meant to be hidden under OPENSSL_NO_DEPRECATED.
The only consumer is a callback function for SSL_CTX_set_tmp_rsa_callback()
deprecated in OpenSSL 1.1.0. Now the function is conditionally compiled too.
The latest HTTP/1.1 draft describes Transfer-Encoding in HTTP/1.0 as having
potentially faulty message framing as that could have been forwarded without
handling of the chunked encoding, and forbids processing subsequest requests
over that connection: https://github.com/httpwg/http-core/issues/879.
While handling of such requests is permitted, the most secure approach seems
to reject them.
The c->read->ready and c->write->ready flags might be reset during
the handshake, and not set again if the handshake was finished on
the other event. At the same time, some data might be read from
the socket during the handshake, so missing c->read->ready flag might
result in a connection hang, for example, when waiting for an SMTP
greeting (which was already received during the handshake).
Found by Sergey Kandaurov.
Previously, when cleaning up a QUIC stream in shutdown mode,
ngx_quic_shutdown_quic() was called, which could close the QUIC connection
right away. This could be a problem if the connection was referenced up the
stack. For example, this could happen in ngx_quic_init_streams(),
ngx_quic_close_streams(), ngx_quic_create_client_stream() etc.
With a typical HTTP/3 client the issue is unlikely because of HTTP/3 uni
streams which need a posted event to close. In this case QUIC connection
cannot be closed right away.
Now QUIC connection read event is posted and it will shut down the connection
asynchronously.
After receiving GOAWAY, client is not supposed to create new streams. However,
until client reads this frame, we allow it to create new streams, which are
gracefully rejected. To prevent client from abusing this algorithm, a new
limit is introduced. Upon reaching keepalive_requests * 2, server now closes
the entire QUIC connection claiming excessive load.
The "hq" mode is HTTP/0.9-1.1 over QUIC. The following limits are introduced:
- uni streams are not allowed
- keepalive_requests is enforced
- keepalive_time is enforced
In case of error, QUIC connection is finalized with 0x101 code. This code
corresponds to HTTP/3 General Protocol Error.
Previously, in-flight byte counter and congestion window were properly
maintained, but the limit was not properly implemented.
Now a new datagram is sent only if in-flight byte counter is less than window.
The limit is datagram-based, which means that a single datagram may lead to
exceeding the limit, but the next one will not be sent.
Previously, the error was ignored leading to unnecessary retransmits.
Now, unsent frames are returned into output queue, state is reset, and
timer is started for the next send attempt.
As per quic-http-34:
Endpoints SHOULD create the HTTP control stream as well as the
unidirectional streams required by mandatory extensions (such as the
QPACK encoder and decoder streams) first, and then create additional
streams as allowed by their peer.
Previously, client could create and destroy additional uni streams unlimited
number of times before creating mandatory streams.
OpenSSL is known to provide read keys for an encryption level before the
level is active in TLS, following the old BoringSSL API. In BoringSSL,
it was then fixed to defer releasing read keys until QUIC may use them.
The directive enables usage of UDP segmentation offloading by quic.
By default, gso is disabled since it is not always operational when
detected (depends on interface configuration).
To improve output performance, UDP segmentation offloading is used
if available. If there is a significant amount of data in an output
queue and path is verified, QUIC packets are not sent one-by-one,
but instead are collected in a buffer, which is then passed to kernel
in a single sendmsg call, using UDP GSO. Such method greatly decreases
number of system calls and thus system load.
Additionally, the ngx_init_srcaddr_cmsg() function is introduced which
initializes control message with connection local address.
The NGX_HAVE_ADDRINFO_CMSG macro is defined when at least one of methods
to deal with corresponding control message is available.
The ngx_wsasend_chain() and ngx_wsarecv_chain() functions were
modified to use only preallocated memory, and the number of
preallocated wsabufs was increased to 64.
Sometimes, QUIC packets need to be of certain (or minimal) size. This is
achieved by adding PADDING frames. It is possible, that adding padding will
affect header size, thus forcing us to recalculate padding size once more.
In d1bde5c3c5d2, the number of preallocated iovec's for ngx_readv_chain()
was increased. Still, in some setups, the function might allocate memory
for iovec's from a connection pool, which is only freed when closing the
connection.
The ngx_readv_chain() function was modified to use only preallocated
memory, similarly to the ngx_writev_chain() change in 8e903522c17a.
Renamed header -> field per quic-qpack naming convention, in particular:
- Header Field -> Field Line
- Header Block -> (Encoded) Field Section
- Without Name Reference -> With Literal Name
- Header Acknowledgement -> Section Acknowledgment
Control characters (0x00-0x1f, 0x7f) and space are not expected to appear
in the Host header. Requests with such characters in the Host header are
now unconditionally rejected.
In 71edd9192f24 logging of invalid headers which were rejected with the
NGX_HTTP_PARSE_INVALID_HEADER error was restricted to just the "client
sent invalid header line" message, without any attempts to log the header
itself.
This patch returns logging of the header up to the invalid character and
the character itself. The r->header_end pointer is now properly set
in all cases to make logging possible.
The same logging is also introduced when parsing headers from upstream
servers.
Control characters (0x00-0x1f, 0x7f), space, and colon were never allowed in
header names. The only somewhat valid use is header continuation which nginx
never supported and which is explicitly obsolete by RFC 7230.
Previously, such headers were considered invalid and were ignored by default
(as per ignore_invalid_headers directive). With this change, such headers
are unconditionally rejected.
It is expected to make nginx more resilient to various attacks, in particular,
with ignore_invalid_headers switched off (which is inherently unsecure, though
nevertheless sometimes used in the wild).
Control characters (0x00-0x1f, 0x7f) were never allowed in URIs, and must
be percent-encoded by clients. Further, these are not believed to appear
in practice. On the other hand, passing such characters might make various
attacks possible or easier, despite the fact that currently allowed control
characters are not significant for HTTP request parsing.
From now on, requests with spaces in URIs are immediately rejected rather
than allowed. Spaces were allowed in 31e9677b15a1 (0.8.41) to handle bad
clients. It is believed that now this behaviour causes more harm than
good.
Per RFC 3986 only the following characters are allowed in URIs unescaped:
unreserved = ALPHA / DIGIT / "-" / "." / "_" / "~"
gen-delims = ":" / "/" / "?" / "#" / "[" / "]" / "@"
sub-delims = "!" / "$" / "&" / "'" / "(" / ")"
/ "*" / "+" / "," / ";" / "="
And "%" can appear as a part of escaping itself. The following
characters are not allowed and need to be escaped: %00-%1F, %7F-%FF,
" ", """, "<", ">", "\", "^", "`", "{", "|", "}".
Not escaping ">" is known to cause problems at least with MS Exchange (see
http://nginx.org/pipermail/nginx-ru/2010-January/031261.html) and in
Tomcat (ticket #2191).
The patch adds escaping of the following chars in all URI parts: """, "<",
">", "\", "^", "`", "{", "|", "}". Note that comments are mostly preserved
to outline important characters being escaped.
HTTP clients are not allowed to generate such requests since Transfer-Encoding
introduction in RFC 2068, and they are not expected to appear in practice
except in attempts to perform a request smuggling attack. While handling of
such requests is strictly defined, the most secure approach seems to reject
them.
No valid CONNECT requests are expected to appear within nginx, since it
is not a forward proxy. Further, request line parsing will reject
proper CONNECT requests anyway, since we don't allow authority-form of
request-target. On the other hand, RFC 7230 specifies separate message
length rules for CONNECT which we don't support, so make sure to always
reject CONNECTs to avoid potential abuse.
Previously, TRACE requests were rejected before parsing Transfer-Encoding.
This is not important since keepalive is not enabled at this point anyway,
though rejecting such requests after properly parsing other headers is
less likely to cause issues in case of further code changes.
After 2096b21fcd10, a single RST_STREAM(NO_ERROR) may not result in an error.
This change removes several unnecessary ctx->type checks for such a case.
As per quic-http-34, these are the cases when this error should be generated:
If an endpoint receives a second SETTINGS frame
on the control stream, the endpoint MUST respond with a connection
error of type H3_FRAME_UNEXPECTED
SETTINGS frames MUST NOT be sent on any stream other than the control
stream. If an endpoint receives a SETTINGS frame on a different
stream, the endpoint MUST respond with a connection error of type
H3_FRAME_UNEXPECTED.
A client MUST NOT send a PUSH_PROMISE frame. A server MUST treat the
receipt of a PUSH_PROMISE frame as a connection error of type
H3_FRAME_UNEXPECTED; see Section 8.
The MAX_PUSH_ID frame is always sent on the control stream. Receipt
of a MAX_PUSH_ID frame on any other stream MUST be treated as a
connection error of type H3_FRAME_UNEXPECTED.
Receipt of an invalid sequence of frames MUST be treated as a
connection error of type H3_FRAME_UNEXPECTED; see Section 8. In
particular, a DATA frame before any HEADERS frame, or a HEADERS or
DATA frame after the trailing HEADERS frame, is considered invalid.
A CANCEL_PUSH frame is sent on the control stream. Receiving a
CANCEL_PUSH frame on a stream other than the control stream MUST be
treated as a connection error of type H3_FRAME_UNEXPECTED.
The GOAWAY frame is always sent on the control stream.
The quic-http-34 is ambiguous as to what error should be generated for the
first frame in control stream:
Each side MUST initiate a single control stream at the beginning of
the connection and send its SETTINGS frame as the first frame on this
stream. If the first frame of the control stream is any other frame
type, this MUST be treated as a connection error of type
H3_MISSING_SETTINGS.
If a DATA frame is received on a control stream, the recipient MUST
respond with a connection error of type H3_FRAME_UNEXPECTED.
If a HEADERS frame is received on a control stream, the recipient MUST
respond with a connection error of type H3_FRAME_UNEXPECTED.
Previously, H3_FRAME_UNEXPECTED had priority, but now H3_MISSING_SETTINGS has.
The arguments in the spec sound more compelling for H3_MISSING_SETTINGS.
- Function ngx_quic_control_flow() is introduced. This functions does
both MAX_DATA and MAX_STREAM_DATA flow controls. The function is called
from STREAM and RESET_STREAM frame handlers. Previously, flow control
was only accounted for STREAM. Also, MAX_DATA flow control was not accounted
at all.
- Function ngx_quic_update_flow() is introduced. This function advances flow
control windows and sends MAX_DATA/MAX_STREAM_DATA. The function is called
from RESET_STREAM frame handler, stream cleanup handler and stream recv()
handler.
Recent fixes to SSL shutdown with lingering close (554c6ae25ffc, 1.19.5)
broke logging of SSL variables. To make sure logging of SSL variables
works properly, avoid freeing c->ssl when doing an SSL shutdown before
lingering close.
Reported by Reinis Rozitis
(http://mailman.nginx.org/pipermail/nginx/2021-May/060670.html).
Instead of calling SSL_free() with each return point, introduced a single
place where cleanup happens. As a positive side effect, this fixes two
potential memory leaks on ngx_handle_read_event() and ngx_handle_write_event()
errors where there were no SSL_free() calls (though unlikely practical,
as errors there are only expected to happen due to bugs or kernel issues).
When starting processing a new encoder instruction, the header state is not
memzero'ed because generally it's burdensome. If the header value is empty,
this resulted in inserting a stale value left from the previous instruction.
Based on a patch by Zhiyong Sun.
On Linux, SO_REUSEADDR allows completely duplicate UDP sockets, so using
SO_REUSEADDR when testing configuration results in packets being dropped
if there is an existing traffic on the sockets being tested (ticket #2187).
While dropped packets are expected with UDP, it is better to avoid this
when possible.
With this change, SO_REUSEADDR is no longer set on datagram sockets when
testing configuration.
Since we anyway do not set SO_REUSEPORT when testing configuration
(see ecb5cd305b06), trying to open additional sockets does not make much
sense, as all these additional sockets are expected to result in EADDRINUSE
errors from bind(). On the other hand, there are reports that trying
to open these sockets takes significant time under load: total configuration
testing time greater than 15s was observed in ticket #2188, compared to less
than 1s without load.
With this change, no additional sockets are opened during testing
configuration.
As per quic-transport 34, FINAL_SIZE_ERROR is generated if an endpoint received
a STREAM frame or a RESET_STREAM frame containing a final size that was lower
than the size of stream data that was already received.
Since nginx always uses exactly one entry in the question section of
a DNS query, and never uses compression pointers in this entry, parsing
of a DNS response in ngx_resolver_process_response() does not expect
compression pointers to appear in the question section of the DNS
response. Indeed, compression pointers in the first name of a DNS response
hardly make sense, do not seem to be allowed by RFC 1035 (which says
"a pointer to a prior occurance of the same name", note "prior"), and
were never observed in practice.
Added an explicit check to ngx_resolver_process_response()'s parsing
of the question section to properly report an error if compression pointers
nevertheless appear in the question section.
Instead of checking on each label if we need to place a dot or not,
now it always adds a dot after a label, and reduces the resulting
length afterwards.
Previously, anything with any of the two high bits set were interpreted
as compression pointers. This is incorrect, as RFC 1035 clearly states
that "The 10 and 01 combinations are reserved for future use". Further,
the 01 combination is actually allocated for EDNS extended label type
(see RFC 2671 and RFC 6891), not really used though.
Fix is to reject unrecognized label types rather than misinterpreting
them as compression pointers.
It is believed to be harmless, and in the worst case it uses some
uninitialized memory as a part of the compression pointer length,
eventually leading to the "name is out of DNS response" error.
Generic function ngx_quic_order_bufs() is introduced. This function creates
and maintains a chain of buffers with holes. Holes are marked with b->sync
flag. Several buffers and holes in this chain may share the same underlying
memory buffer.
When processing STREAM frames with this function, frame data is copied only
once to the right place in the stream input chain. Previously data could
be copied twice. First when buffering an out-of-order frame data, and then
when filling stream buffer from ordered frame queue. Now there's only one
data chain for both tasks.
When variables are used in ssl_certificate or ssl_certificate_key, a request
is created in the certificate callback to evaluate the variables, and then
freed. Freeing it, however, updates c->log->action to "closing request",
resulting in confusing error messages like "client timed out ... while
closing request" when a client times out during the SSL handshake.
Fix is to restore c->log->action after calling ngx_http_free_request().
According to profiling, those two are among most frequently called,
so inlining is generally useful, and unrolling should help with it.
Further, this fixes undefined behaviour seen with invalid values.
Inspired by Yu Liu.
Similarly to smtpd_hard_error_limit in Postfix and smtp_max_unknown_commands
in Exim, specifies the number of errors after which the connection is closed.
The change is mostly the same as the SMTP one (04e43d03e153 and 3f5d0af4e40a),
and ensures that nginx is able to properly handle or reject multiple IMAP
commands. The s->cmd field is not really used and set for consistency.
Non-synchronizing literals handling in invalid/unknown commands is limited,
so when a non-synchronizing literal is detected at the end of a discarded
line, the connection is closed.
Only "A-Za-z0-9-._" characters now allowed (which is stricter than what
RFC 3501 requires, but expected to be enough for all known clients),
and tags shouldn't be longer than 32 characters.
Previously, s->backslash was set if any of the arguments was a quoted
string with a backslash character. After successful command parsing
this resulted in all arguments being filtered to remove backslashes.
This is, however, incorrect, as backslashes should not be removed from
IMAP literals. For example:
S: * OK IMAP4 ready
C: a01 login {9}
S: + OK
C: user\name "pass\"word"
S: * BAD internal server error
resulted in "Auth-User: username" instead of "Auth-User: user\name"
as it should.
Fix is to apply backslash filtering on per-argument basis during parsing.
As discussed in the previous change, s->arg_start handling in the "done"
labels of ngx_mail_pop3_parse_command(), ngx_mail_imap_parse_command(),
and ngx_mail_smtp_parse_command() is wrong: s->arg_start cannot be
set there, as it is handled and cleared on all code paths where the
"done" labels are reached. The relevant code is dead and now removed.
Previously, s->arg_start was left intact after invalid IMAP commands,
and this might result in an argument incorrectly added to the following
command. Similarly, s->backslash was left intact as well, leading
to unneeded backslash removal.
For example (LFs from the client are explicitly shown as "<LF>"):
S: * OK IMAP4 ready
C: a01 login "\<LF>
S: a01 BAD invalid command
C: a0000000000\2 authenticate <LF>
S: a00000000002 aBAD invalid command
The backslash followed by LF generates invalid command with s->arg_start
and s->backslash set, the following command incorrectly treats anything
from the old s->arg_start to the space after the command as an argument,
and removes the backslash from the tag. If there is no space, s->arg_end
will be NULL.
Both things seem to be harmless though. In particular:
- This can be used to provide an incorrect argument to a command without
arguments. The only command which seems to look at the single argument
is AUTHENTICATE, and it checks the argument length before trying to
access it.
- Backslash removal uses the "end" pointer, and stops due to "src < end"
condition instead of scanning all the process memory if s->arg_end is
NULL (and arg[0].len is huge).
- There should be no backslashes in unquoted strings.
An obvious fix is to clear s->arg_start and s->backslash on invalid commands,
similarly to how it is done in POP3 parsing (added in 810:e3aa8f305d21) and
SMTP parsing.
This, however, makes it clear that s->arg_start handling in the "done"
label is wrong: s->arg_start cannot be legitimately set there, as it
is expected to be cleared in all possible cases when the "done" label is
reached. The relevant code is dead and will be removed by the following
change.
The change is mostly the same as the SMTP one (04e43d03e153 and 3f5d0af4e40a),
and ensures that nginx is able to properly handle or reject multiple POP3
commands, as required by the PIPELINING capability (RFC 2449). The s->cmd
field is not really used and set for consistency.
There is no need to scan buffer from s->buffer->pos, as we already scanned
the buffer till "p" and wasn't able to find an LF.
There is no real need for this change in SMTP, since it is at most a
microoptimization of a non-common code path. Similar code in IMAP, however,
will have to start scanning from "p" to be correct, since there can be
newlines in IMAP literals.
Previously, if an invalid SMTP command was split between reads, nginx failed
to wait for LF before returning an error, and interpreted the rest of the
command received later as a separate command.
The sw_invalid state in ngx_mail_smtp_parse_command(), introduced in
04e43d03e153, did not work, since ngx_mail_smtp_auth_state() clears
s->state when returning an error due to NGX_MAIL_PARSE_INVALID_COMMAND.
And not clearing s->state will introduce another problem: the rest
of the command would trigger duplicate error when rest of the command is
received.
Fix is to return NGX_AGAIN from ngx_mail_smtp_parse_command() until full
command is received.
Previously, if there were some pipelined SMTP data in the buffer when
a proxied connection with the backend was established, nginx called
ngx_mail_proxy_handler() to send these data, and not tried to send the
response to the last command. In most cases, this response was later sent
along with the response to the pipelined command, but if for some reason
client decides to wait for the response before finishing the next command
this might result in a connection hang.
Fix is to always call ngx_mail_proxy_handler() to send the response, and
additionally post an event to send the pipelined data if needed.
When using server push, a segfault occured because
ngx_http_v3_create_push_request() accessed ngx_http_v3_session_t object the old
way. Prior to 9ec3e71f8a61, HTTP/3 session was stored directly in c->data.
Now it's referenced by the v3_session field of ngx_http_connection_t.
With this change, it is now possible to use ngx_conf_merge_ptr_value()
to merge complex values. This change follows much earlier changes in
ngx_conf_merge_ptr_value() and ngx_conf_set_str_array_slot()
in 1452:cd586e963db0 (0.6.10) and 1701:40d004d95d88 (0.6.22), and the
change in ngx_conf_set_keyval_slot() (7728:485dba3e2a01, 1.19.4).
To preserve compatibility with existing 3rd party modules, both NULL
and NGX_CONF_UNSET_PTR are accepted for now.
Client IDs cannot be reused on different paths. This change allows to reuse
client id previosly seen on the same path (but with different dcid) in case
when no unused client IDs are available.
According to quic-transport, 9.1:
PATH_CHALLENGE, PATH_RESPONSE, NEW_CONNECTION_ID, and PADDING frames
are "probing frames", and all other frames are "non-probing frames".
Previously it was parsed in ngx_http_v3_streams.c, while the streams were
parsed in ngx_http_v3_parse.c. Now all parsing is done in one file. This
simplifies parsing API and cleans up ngx_http_v3_streams.c.
Previously, an ngx_http_v3_connection_t object was created for HTTP/3 and
then assinged to c->data instead of the generic ngx_http_connection_t object.
Now a direct reference is added to ngx_http_connection_t, which is less
confusing and does not require a flag for http3.
It's used instead of accessing c->quic->parent->data directly. Apart from being
simpler, it allows to change the way session is stored in the future by changing
the macro.
Now ngx_http_v3_ack_header() and ngx_http_v3_inc_insert_count() always generate
decoder error. Our implementation does not use dynamic tables and does not
expect client to send Section Acknowledgement or Insert Count Increment.
Stream Cancellation, on the other hand, is allowed to be sent anyway. This is
why ngx_http_v3_cancel_stream() does not return an error.
The patch adds proper transitions between multiple networking addresses that
can be used by a single quic connection. New networking paths are validated
using PATH_CHALLENGE/PATH_RESPONSE frames.
Due to structure's alignment, some uninitialized memory contents may have
been passed between processes.
Zeroing was removed in 0215ec9aaa8a.
Reported by Johnny Wang.
7.2.1:
If a DATA frame is received on a control stream, the recipient MUST
respond with a connection error of type H3_FRAME_UNEXPECTED;
7.2.2:
If a HEADERS frame is received on a control stream, the recipient MUST
respond with a connection error (Section 8) of type H3_FRAME_UNEXPECTED.
With SMTP pipelining, ngx_mail_read_command() can be called with s->buffer
without any space available, to parse additional commands received to the
buffer on previous calls. Previously, this resulted in recv() being called
with zero length, resulting in zero being returned, which was interpreted
as a connection close by the client, so nginx silently closed connection.
Fix is to avoid calling c->recv() if there is no free space in the buffer,
but continue parsing of the already received commands.
Currently both names are used which is confusing. Historically these were
different objects, but now it's the same one. The name qs (quic stream) makes
more sense than sn (stream node).
PATH_RESPONSE was explicitly forbidden in 0-RTT since at least draft-22, but
the Frame Types table was not updated until recently while in IESG evaluation.
Stop including QUIC headers with no user-serviceable parts inside.
This allows to provide a much cleaner QUIC interface. To cope with that,
ngx_quic_derive_key() is now explicitly exported for v3 and quic modules.
Additionally, this completely hides the ngx_quic_keys_t internal type.
The "ngx_event_quic.h" header file now contains only public definitions,
used by modules. All internal definitions are moved into
the "ngx_event_quic_connection.h" header file.
The function correctly cleans up resources in case of failure to create
initial server id: it removes previously created udp node for odcid from
listening rbtree.
It turns out no browsers implement HTTP/2 GOAWAY handling properly, and
large enough number of resources on a page results in failures to load
some resources. In particular, Chrome seems to experience errors if
loading of all resources requires more than 1 connection (while it
is usually able to retry requests at least once, even with 2 connections
there are occasional failures for some reason), Safari if loading requires
more than 3 connections, and Firefox if loading requires more than 10
connections (can be configured with network.http.request.max-attempts,
defaults to 10).
It does not seem to be possible to resolve this on nginx side, even strict
limiting of maximum concurrency does not help, and loading issues seems to
be triggered by merely queueing of a request for a particular connection.
The only available mitigation seems to use higher keepalive_requests value.
The new default is 1000 and matches previously used default for
http2_max_requests. It is expected to be enough for 99.98% of the pages
(https://httparchive.org/reports/state-of-the-web?start=latest#reqTotal)
even in Chrome.
Similar to lingering_time, it limits total connection lifetime before
keepalive is switched off. The default is 1 hour, which is close to
the total maximum connection lifetime possible with default
keepalive_requests and keepalive_timeout.
Firefox uses several idle streams for PRIORITY frames[1], and
"http2_max_concurrent_streams 1;" results in "client sent too many
PRIORITY frames" errors when a connection is established by Firefox.
Fix is to relax the PRIORITY frames limit to use at least 100 as
the initial value (which is the recommended by the HTTP/2 protocol
minimum limit on the number of concurrent streams, so it is not
unreasonable for clients to assume that similar number of idle streams
can be used for prioritization).
[1] https://hg.mozilla.org/mozilla-central/file/32a9e6e145d6e3071c3993a20bb603a2f388722b/netwerk/protocol/http/Http2Stream.cpp#l1270
In FreeBSD 13, eventfd(2) was added, and this breaks build
with --test-build-epoll and without --with-file-aio. Fix is
to move eventfd(2) detection to auto/os/linux, as it is used
only on Linux as a notification mechanism for epoll().
In current versions (all versions based on zlib 1.2.11, at least
since 2018) it no longer uses 64K hash and does not force window
bits to 13 if it is less than 13. That is, it needs just 16 bytes
more memory than normal zlib, so these bytes are simply added to
the normal size calculation.
In limit_req, auth_delay, and upstream code to check for broken
connections, tests for possible connection close by the client
did not work if the connection was already closed when relevant
event handler was set. This happened because there were no additional
events in case of edge-triggered event methods, and read events
were disabled in case of level-triggered ones.
Fix is to explicitly post a read event if the c->read->ready flag
is set.
For new data to be reported with eventport on Solaris,
ngx_handle_read_event() needs to be called after reading response
headers. To do so, ngx_http_upstream_process_non_buffered_upstream()
now called unconditionally if there are no prepread data. This
won't cause any read() syscalls as long as upstream connection
is not ready for reading (c->read->ready is not set), but will result
in proper handling of all events.
If we need to be notified about further events, ngx_handle_read_event()
needs to be called after a read event is processed. Without this,
an event can be removed from the kernel and won't be reported again,
notably when using oneshot event methods, such as eventport on Solaris.
While here, error handling is also added, similar to one present in
ngx_resolver_tcp_read(). This is not expected to make a difference
and mostly added for consistency.
If an attempt is made to delete an event which was already reported,
port_dissociate() returns an error. Fix is avoid doing anything if
ev->active is not set.
Possible alternative approach would be to avoid calling ngx_del_event()
at all if ev->active is not set. This approach, however, will require
something else to re-add the other event of the connection, since both
read and write events are dissociated if an event is reported on a file
descriptor. Currently ngx_eventport_del_event() re-associates write
event if called to delete read event, and vice versa.
If, at the start of an event loop iteration, there are any timers
in the past (including timers expiring now), the ngx_process_events()
function is called with zero timeout, and returns immediately even
if there are no events. But the following code only calls
ngx_event_expire_timers() if time actually changed, so this results
in nginx spinning in the event loop till current time changes.
While such timers are not expected to appear under normal conditions,
as all such timers should be removed on previous event loop iterations,
they still can appear due to bugs, zero timeouts set in the configuration
(if this is not explicitly handled by the code), or due to external
time changes on systems without clock_gettime(CLOCK_MONOTONIC).
Fix is to call ngx_event_expire_timers() unconditionally. Calling
it on each event loop iteration is not expected to be significant from
performance point of view, especially compared to a syscall in
ngx_process_events().
Without explicit handling, a zero timer was actually added, leading to
multiple unneeded syscalls. Further, sending GOAWAY frame early might
be beneficial for clients.
Reported by Sergey Kandaurov.
Unlike in 75e908236701, which added the logic to ngx_http_finalize_request(),
this change moves it to a more generic routine ngx_http_finalize_connection()
to cover cases when a request is finalized with NGX_DONE.
In particular, this fixes unwanted connection transition into the keepalive
state after receiving EOF while discarding request body. With edge-triggered
event methods that means the connection will last for extra seconds as set in
the keepalive_timeout directive.
The response size check introduced in 39501ce97e29 did not take into
account possible padding on DATA frames, resulting in incorrect
"upstream sent response body larger than indicated content length" errors
if upstream server used padding in responses with known length.
Fix is to check the actual size of response buffers produced by the code,
similarly to how it is done in other protocols, instead of checking
the size of DATA frames.
Reported at:
http://mailman.nginx.org/pipermail/nginx-devel/2021-March/013907.html
Currently listener contains rbtree with multiple nodes for single QUIC
connection: each corresponding to specific server id. Each udp node points
to same ngx_connection_t, which points to QUIC connection via c->udp field.
Thus when an event handler is called, it only gets ngx_connection_t with
c->udp pointing to QUIC connection. This makes it hard to obtain actual
node which was used to dispatch packet (it requires to repeat DCID lookup).
Additionally, ngx_quic_connection_t->udp field is only needed to keep a
pointer in c->udp. The node is not added into the tree and does not carry
useful information.
Sometimes it is required to process datagram properties at higher level (i.e.
QUIC is interested in source address which may change and IP options). The
patch adds ngx_udp_dgram_t structure used to pass packet-related information
in c->udp.
Previously, when a new datagram arrived, data were copied from the UDP layer
to the QUIC layer via c->recv() interface. Now UDP buffer is accessed
directly.
OpenSSL 3.0 started to require HKDF-Extract output PRK length pointer
used to represent the amount of data written to contain the length of
the key buffer before the call. EVP_PKEY_derive() documents this.
See HKDF_Extract() internal implementation update in this change:
https://github.com/openssl/openssl/commit/5a285ad
The function ngx_quic_shutdown_connection() waits until all non-cancelable
streams are closed, and then closes the connection. In HTTP/3 cancelable
streams are all unidirectional streams except push streams.
The function is called from HTTP/3 when client reaches keepalive_requests.
In case of long header packets, dcid length was not read correctly.
While there, macros to parse uint64 was fixed as well as format specifiers
to print it in debug mode.
Thanks to Gao Yan <gaoyan09@baidu.com>.
Activated with the "proxy_protocol" directive. Can be combined with
"listen ... proxy_protocol;" and "set_real_ip_from ...;" to pass
client address provided to nginx in the PROXY protocol header.
Activated with the "proxy_protocol" parameter of the "listen" directive.
Obtained information is passed to the auth_http script in Proxy-Protocol-Addr,
Proxy-Protocol-Port, Proxy-Protocol-Server-Addr, and Proxy-Protocol-Server-Port
headers.
Similarly to 40e8ce405859 in the stream module, this reduces the time
accept mutex is held. This also simplifies following changes to
introduce PROXY protocol support.
If we need to be notified about further events, ngx_handle_read_event()
needs to be called after a read event is processed. Without this,
an event can be removed from the kernel and won't be reported again,
notably when using oneshot event methods, such as eventport on Solaris.
For consistency, existing ngx_handle_read_event() call removed from
ngx_mail_read_command(), as this call only covers one of the code paths
where ngx_mail_read_command() returns NGX_AGAIN. Instead, appropriate
processing added to the callers, covering all code paths where NGX_AGAIN
is returned.
As long as a read event is blocked (ignored), ngx_handle_read_event()
needs to be called to make sure no further notifications will be
triggered when using level-triggered event methods, such as select() or
poll().
The "!rev->ready" test seems to be a typo, introduced in the original
commit (719:f30b1a75fd3b). The ngx_handle_write_event() code properly
tests for "rev->ready" instead.
Due to this typo, read events might be unexpectedly removed during
proxying after an event on the other part of the proxied connection.
Catched by mail proxying tests.
The strerrordesc_np() function, introduced in glibc 2.32, provides an
async-signal-safe way to obtain error messages. This makes it possible
to avoid copying error messages.
Previously, systems without sys_nerr (or _sys_nerr) were handled with an
assumption that errors start at 0 and continuous. This is, however, not
something POSIX requires, and not true on some platforms.
Notably, on Linux, where sys_nerr is no longer available for newly linked
binaries starting with glibc 2.32, there are gaps in error list, which
used to stop us from properly detecting maximum errno. Further, on
GNU/Hurd errors start at 0x40000001.
With this change, maximum errno detection is moved to the runtime code,
now able to ignore gaps, and also detects the first error if needed.
This fixes observed "Unknown error" messages as seen on Linux with
glibc 2.32 and on GNU/Hurd.
With this change, behaviour of HTTP/2 becomes even closer to HTTP/1.x,
and client_header_timeout instead of keepalive_timeout is used before
the first request is received.
This fixes HTTP/2 connections being closed even before the first request
if "keepalive_timeout 0;" was used in the configuration; the problem
appeared in f790816a0e87 (1.19.7).
Using default highlighting makes it possible to easily overrule
highlighting specified in the syntax file, see ":highlight-default"
in vim help for details.
As per quic-transport-34:
An endpoint also restarts its idle timer when sending an ack-eliciting
packet if no other ack-eliciting packets have been sent since last receiving
and processing a packet.
Previously, the timer was set for any packet.
The structure is used to parse an HTTP/3 request. An object of this type is
added to ngx_http_request_t instead of h3_parse generic pointer.
Also, the new field is located outside of the request ephemeral zone to keep it
safe after request headers are parsed.
Previously, PINGs and other frames extended possible keepalive time,
making it possible to keep an open HTTP/2 connection for a long time.
Now the connection is always closed as long as keepalive_timeout expires,
similarly to how it happens in HTTP/1.x.
Note that as a part of this change, incomplete frames are no longer
trigger a separate timeout, so http2_recv_timeout (replaced by
client_header_timeout in previous patches) is essentially cancelled.
The client_header_timeout is, however, used for SSL handshake and
while reading HEADERS frames.
Instead, keepalive_timeout and keepalive_requests are now used. This
is expected to simplify HTTP/2 code and usage. This also matches
directives used by upstream module for all protocols.
In case of default settings, this effectively changes maximum number
of requests per connection from 1000 to 100. This looks acceptable,
especially given that HTTP/2 code now properly supports lingering close.
Further, this changes default keepalive timeout in HTTP/2 from 300 seconds
to 75 seconds. This also looks acceptable, and larger than PING interval
used by Firefox (network.http.spdy.ping-threshold defaults to 58s),
the only browser to use PINGs.
Instead, the client_header_timeout is now used for HTTP/2 reading.
Further, the timeout is changed to be set once till no further data
left to read, similarly to how client_header_timeout is used in other
places.
New connections are marked reusable by ngx_http_init_connection() if there
are no data available for reading. As a result, if SSL is not used,
ngx_http_v2_init() might be called when the connection is marked reusable.
If a HEADERS frame is immediately available for reading, this resulted
in connection being preserved in reusable state with an active request,
and possibly closed later as if during worker shutdown (that is, after
all active requests were finalized).
Fix is to explicitly mark connections non-reusable in ngx_http_v2_init()
instead of (incorrectly) assuming they are already non-reusable.
Found by Sergey Kandaurov.
If ngx_drain_connections() fails to immediately reuse any connections
and there are no free connections, it now additionally tries to reuse
a connection again. This helps to provide at least one free connection
in case of HTTP/2 with lingering close, where merely trying to reuse
a connection once does not free it, but makes it reusable again,
waiting for lingering close.
This is particularly important in HTTP/2, where keepalive connections
are closed with lingering. Before the patch, reusing a keepalive HTTP/2
connection resulted in the connection waiting for lingering close to
remain in the reusable connections queue, preventing ngx_drain_connections()
from closing additional connections.
The patch fixes it by marking the connection reusable again, and so
moving it in the reusable connections queue. Further, it makes actually
possible to reuse such connections if needed.
The "max_ack_delay", "ack_delay_exponent", and "max_udp_payload_size"
transport parameters were not communicated to client.
The "disable_active_migration" and "active_connection_id_limit"
parameters were not saved into zero-rtt context.
18.1. Reserved Transport Parameters
Transport parameters with an identifier of the form "31 * N + 27" for
integer values of N are reserved to exercise the requirement that
unknown transport parameters be ignored. These transport parameters
have no semantics, and can carry arbitrary values.
Setting the timer is brought into compliance with quic-recovery-34. Now it's
set from a single function ngx_quic_set_lost_timer() that takes into account
both loss detection and PTO. The following issues are fixed with this change:
- when in loss detection mode, discarding a context could turn off the
timer forever after switching to the PTO mode
- when in loss detection mode, sending a packet resulted in rescheduling the
timer as if it's always in the PTO mode
As per quic-transport-33:
An endpoint MUST acknowledge all ack-eliciting Initial and Handshake
packets immediately
If a packet carrying Initial or Handshake ACK was lost, a non-immediate ACK
should not be sent later. Instead, client is expected to send a new packet
to acknowledge.
Sending non-immediate ACKs for Initial packets can cause the client to
generate an inflated RTT sample.
The token generation in QUIC is reworked. Single host key is used to generate
all required keys of needed sizes using HKDF.
The "quic_stateless_reset_token_key" directive is removed. Instead, the
"quic_host_key" directive is used, which reads key from file, or sets it
to random bytes if not specified.
The flag was introduced to create type-aware CONNECTION_CLOSE frames,
and now is replaced with frame type information, directly accessible.
Notably, this fixes type logging for received frames in b3d9e57d0f62.
The flag is used in ngx_http_finalize_connection() to switch client connection
to the keepalive mode. Since eaea7dac3292 this code is not executed for HTTP/3
which allows us to revert the change and get back to the default branch code.
This part somehow slipped away from c5840ca2063d.
While it is not expected to be needed in case of lingering close,
it is good to keep it for correctness (see 2b5528023f6b).
The change reduces diff to the default branch for
src/http/ngx_http_request_body.c.
Also, client Content-Length, if present, is now checked against the real body
size sent by client.
- split ngx_quic_process_packet() in two functions with the second one called
ngx_quic_process_payload() in charge of decrypring and handling the payload
- renamed ngx_quic_payload_handler() to ngx_quic_handle_frames()
- moved error cleanup from ngx_quic_input() to ngx_quic_process_payload()
- moved handling closed connection from ngx_quic_handle_frames() to
ngx_quic_process_payload()
- minor fixes
Previously, quic connection object was created when Retry packet was sent.
This is neither necessary nor convenient, and contradicts the idea of retry:
protecting from bad clients and saving server resources.
Now, the connection is not created, token is verified cryptographically
instead of holding it in connection.
This function should be called at the end of an event handler to prepare the
event for the next handler call. Particularly, the "active" flag is set or
cleared depending on data availability.
With this call missing in one code path, read handler was not called again
after handling the initial part of the client request, if the request was too
big to fit into a single STREAM frame.
Now ngx_handle_read_event() is called in this code path. Also, read timer is
restarted.
Keeping post_accept_timeout in ngx_listening_t is no longer needed since
we've switched to 1 second timeout for deferred accept in 5541:fdb67cfc957d.
Further, using it in HTTP code can result in client_header_timeout being
used from an incorrect server block, notably if address-specific virtual
servers are used along with a wildcard listening socket, or if we've switched
to a different server block based on SNI in SSL handshake.
The stub status module and ngx_http_send_response() (used by the empty gif
module and the "return" directive) incorrectly assumed that responding
to HEAD requests always results in r->header_only being set. This is not
true, and results in incorrect behaviour, for example, in the following
configuration:
location / {
image_filter size;
return 200 test;
}
Fix is to remove this incorrect micro-optimization from both stub status
module and ngx_http_send_response().
Reported by Chris Newton.
After 7675:9afa45068b8f and 7678:bffcc5af1d72 (1.19.1), during non-buffered
simple proxying, responses with extra data might result in zero size buffers
being generated and "zero size buf" alerts in writer. This bug is similar
to the one with FastCGI proxying fixed in 7689:da8d758aabeb.
In non-buffered mode, normally the filter function is not called if
u->length is already 0, since u->length is checked after each call of
the filter function. There is a case when this can happen though: if
the response length is 0, and there are pre-read response body data left
after reading response headers. As such, a check for u->length is needed
at the start of non-buffered filter functions, similar to the one
for p->length present in buffered filter functions.
Appropriate checks added to the existing non-buffered copy filters
in the upstream (used by scgi and uwsgi proxying) and proxy modules.
A header with the name containing null, CR, LF, colon or uppercase characters,
is now considered an error. A header with the value containing null, CR or LF,
is also considered an error.
Also, header is considered invalid unless its name only contains lowercase
characters, digits, minus and optionally underscore. Such header can be
optionally ignored.
- :method, :path and :scheme are expected exactly once and not empty
- :method and :scheme character validation is added
- :authority cannot appear more than once
Notably, the version negotiation table is updated to reject draft-33/QUICv1
(which requires a new TLS codepoint) unless explicitly asked to built with.
The quic kernel bpf helper inspects packet payload for DCID, extracts key
and routes the packet into socket matching the key.
Due to reuseport feature, each worker owns a personal socket, which is
identified by the same key, used to create DCID.
BPF objects are locked in RAM and are subject to RLIMIT_MEMLOCK.
The "ulimit -l" command may be used to setup proper limits, if maps
cannot be created with EPERM or updated with ETOOLONG.
With introduction of open_file_cache in 1454:f497ed7682a7, opening a file
with ngx_open_cached_file() automatically adds a cleanup handler to close
the file. As such, calling ngx_close_file() directly for non-regular files
is no longer needed and will result in duplicate close() call.
In 1454:f497ed7682a7 ngx_close_file() call for non-regular files was removed
in the static module, but wasn't in the flv module. And the resulting
incorrect code was later copied to the mp4 module. Fix is to remove the
ngx_close_file() call from both modules.
Reported by Chris Newton.
The ngx_http_parse_complex_uri() function cannot make URI longer and does
not null-terminate URI, so there is no need to allocate an extra byte. This
allocation appears to be a leftover from changes in 461:a88a3e4e158f (0.1.5),
where null-termination of r->uri and many other strings was removed.
When the request line contains request-target in the absolute-URI form,
it can contain path-empty instead of a single slash (see RFC 7230, RFC 3986).
Previously, the ngx_http_parse_request_line() function only accepted empty
path when there was no query string.
With this change, non-empty query is also correctly handled. That is,
request line "GET http://example.com?foo HTTP/1.1" is accepted and results
in $uri "/" and $args "foo".
Note that $request_uri remains "?foo", similarly to how spaces in URIs
are handled. Providing "/?foo", similarly to how "/" is provided for
"GET http://example.com HTTP/1.1", requires allocation.
Previously, when processing client ACK, rtt could be calculated for a packet
different than the largest if it was missing in the sent chain. Even though
this is an unlikely situation, rtt based on a different packet could be larger
than needed leading to bigger pto timeout and performance degradation.
Previously, this only worked for Application level because before
quic-transport-30, there were the following constraints:
Because the receiver doesn't use the ACK Delay for Initial and Handshake
packets, a sender SHOULD send a value of 0.
When adjusting an RTT sample using peer-reported acknowledgement delays, an
endpoint ... MUST ignore the ACK Delay field of the ACK frame for packets
sent in the Initial and Handshake packet number space.
Now initial output packet is not padded anymore if followed by a handshake
packet. If the datagram is still not big enough to satisfy minimum size
requirements, handshake packet is padded.
The patch replaces c->send() occurences with c->send_chain(), because the
latter accounts for the local address, which may be different if the wildcard
listener is used.
Previously, server sent response to client using address different from
one client connected to.
Notably, this fixes an issue with Chrome that can emit a "certificate_unknown"
alert during the SSL handshake where c->ssl->no_wait_shutdown is not yet set.
The filter is responsible for creating HTTP/3 response header and body.
The change removes differences to the default branch for
ngx_http_chunked_filter_module and ngx_http_header_filter_module.
Instead, appropriate format specifier for hexadecimal is used
in ngx_log_debug().
The STREAM frame "data" debug is moved into ngx_quic_log_frame(), similar
to all other frame fields debug.
Previously, the number of next_upstream tries included servers marked
as "down", resulting in "no live upstreams" with the code 502 instead
of the code derived from an attempt to connect to the last tried "up"
server (ticket #2096).
Similarly to the problem fixed in 2096b21fcd10 (ticket #1792),
when a "trailer only" gRPC response (that is, a response with the
END_STREAM flag in the HEADERS frame) was immediately followed by
RST_STREAM(NO_ERROR) in the data preread along with the response
header, RST_STREAM wasn't properly skipped and caused "upstream
rejected request with error 0" errors.
Observed with "unknown service" gRPC errors returned by grpc-go.
Fix is to set ctx->done if we are going to parse additional data,
so the RST_STREAM(NO_ERROR) is properly skipped. Additionally, now
ngx_http_grpc_filter() will complain about frames sent for closed
stream if there are any.
When installing or running from a non-root user it is sometimes required to
override default, compiled in error log path. There was no way to do this
without rebuilding the binary (ticket #147).
This patch introduced "-e" command line option which allows one to override
compiled in error log path.
Header value returned from the HTTP parser is expected to be null-terminated or
have a spare byte after the value bytes. When an empty header value was passed
by client in a literal header representation, neither was true. This could
result in segfault. The fix is to assign a literal empty null-terminated
string in this case.
Thanks to Andrey Kolyshkin.
The largely duplicate type-specific functions ngx_quic_parse_initial_header(),
ngx_quic_parse_handshake_header(), and a missing one for 0-RTT, were merged.
The new order of functions listed in ngx_event_quic_transport.c reflects this.
|_ ngx_quic_parse_long_header - version-invariant long header fields
\_ ngx_quic_supported_version - a helper to decide we can go further
\_ ngx_quic_parse_long_header_v1 - QUICv1-specific long header fields
0-RTT packets previously appeared as Handshake are now logged as appropriate:
*1 quic packet rx long flags:db version:ff00001d
*1 quic packet rx early len:870
Logging SCID/DCID is no longer duplicated as were seen with Initial packets.
If trailers were missing and a chain carrying the last_buf flag had no data
in it, then last HTTP/1 chunk was broken. The problem was introduced while
implementing HTTP/3 response body generation.
The change fixes the issue and reduces diff to the mainline nginx.
Previously, if quic_stateless_reset_token_key was empty or unspecified,
initial stateless reset token was not generated. However subsequent tokens
were generated with empty key, which resulted in error with certain SSL
libraries, for example OpenSSL.
Now a random 32-byte stateless reset token key is generated if none is
specified in the configuration. As a result, stateless reset tokens are now
generated for all server ids.
Previously new dcid was generated in the same memory that was allocated for
qc->dcid when creating the QUIC connection. However this memory was also
referenced by initial_source_connection_id and retry_source_connection_id
transport parameters. As a result these parameters changed their values after
retry which broke the protocol.
Addon modules, both dynamic and static, can now use shared source files.
Shared sources result in only one make rule even if specified several
times in different modules.
Before introduction of request body filter in 42d9beeb22db, the only
possible return code from the ngx_http_request_body_filter() call
without actual buffers was NGX_HTTP_INTERNAL_SERVER_ERROR, and
the code in ngx_http_read_client_request_body() hardcoded the only
possible error to simplify the code of initial call to set rb->rest.
This is no longer true after introduction of request body filters though,
as a request body filter might need to return other errors, such as 403.
Fix is to preserve the error code actually returned by the call
instead of assuming 500.
Added logging before returning NGX_HTTP_INTERNAL_SERVER_ERROR if there
are busy buffers after a request body flush. This should never happen
with current code, though bugs can be introduced by 3rd party modules.
Make sure debugging will be easy enough.
A negotiated version is decoupled from NGX_QUIC_VERSION and, if supported,
now stored in c->quic->version after packets processing. It is then used
to create long header packets. Otherwise, the list of supported versions
(which may be many now) is sent in the Version Negotiation packet.
All packets in the connection are expected to have the same version.
Incoming packets with mismatched version are now rejected.
When doing lingering close, the socket was first shut down for writing,
so SSL shutdown initiated after lingering close was not able to send
the close_notify alerts (ticket #2056).
The fix is to call ngx_ssl_shutdown() before shutting down the socket.
A number of unsigned variables has a special value, usually -1 or some maximum,
which produces huge numeric value in logs and makes them hard to read.
In order to distinguish such values in log, they are casted to the signed type
and printed as literal '-1'.
The client address validation didn't complete with a valid token,
which was broken after packet processing refactoring in d0d3fc0697a0.
An invalid or expired token was treated as a connection error.
Now we proceed as outlined in draft-ietf-quic-transport-32,
section 8.1.3 "Address Validation for Future Connections" below,
which is unlike validating the client address using Retry packets.
When a server receives an Initial packet with an address validation
token, it MUST attempt to validate the token, unless it has already
completed address validation. If the token is invalid then the
server SHOULD proceed as if the client did not have a validated
address, including potentially sending a Retry.
The connection is now closed in this case on internal errors only.
All key handling functionality is moved into ngx_quic_protection.c.
Public structures from ngx_quic_protection.h are now private and new
methods are available to manipulate keys.
A negotiated cipher is cached in QUIC connection from the set secret callback
to avoid calling SSL_get_current_cipher() on each encrypt/decrypt operation.
This also reduces the number of unwanted c->ssl->connection occurrences.
Now "s", "V", and "v" format specifiers may be prefixed with "x" (lowercase)
or "X" (uppercase) to output corresponding data in hexadecimal format.
In collaboration with Maxim Dounin.
Previously, tx ACK frames held ranges in an array of ngx_quic_ack_range_t,
while rx ACK frames held ranges in the serialized format. Now serialized format
is used for both types of frames.
The patch resets ctx->frames queue, which may contain frames. It was possible
that congestion or amplification limits prevented all frames to be sent.
Retransmitted frames could be accounted twice as inflight: first time in
ngx_quic_congestion_lost() called from ngx_quic_resend_frames(), and later
from ngx_quic_discard_ctx().
This accounts for the following change:
* Require expansion of datagrams to ensure that a path supports at
least 1200 bytes:
- During the handshake ack-eliciting Initial packets from the
server need to be expanded
All values are prefixed with name and separated from it using colon.
Multiple values are listed without commas in between.
Rationale: this greatly simplifies log parsing for analysis.
For application level packets, only every second packet is now acknowledged,
respecting max ack delay.
13.2.1 Sending ACK Frames
In order to assist loss detection at the sender, an endpoint SHOULD
generate and send an ACK frame without delay when it receives an ack-
eliciting packet either:
* when the received packet has a packet number less than another
ack-eliciting packet that has been received, or
* when the packet has a packet number larger than the highest-
numbered ack-eliciting packet that has been received and there are
missing packets between that packet and this packet.
13.2.2. Acknowledgement Frequency
A receiver SHOULD send an ACK frame after receiving at least two
ack-eliciting packets.
In some cases it might be needed to reject SSL handshake based on SNI
server name provided, for example, to make sure an invalid certificate
is not returned to clients trying to contact a name-based virtual server
without SSL configured. Previously, a "ssl_ciphers aNULL;" was used for
this. This workaround, however, is not compatible with TLSv1.3, in
particular, when using BoringSSL, where it is not possible to configure
TLSv1.3 ciphers at all.
With this change, the ssl_reject_handshake directive is introduced,
which instructs nginx to reject SSL handshakes with an "unrecognized_name"
alert in a particular server block.
For example, to reject handshake with names other than example.com,
one can use the following configuration:
server {
listen 443 ssl;
ssl_reject_handshake on;
}
server {
listen 443 ssl;
server_name example.com;
ssl_certificate example.com.crt;
ssl_certificate_key example.com.key;
}
The following configuration can be used to reject all SSL handshakes
without SNI server name provided:
server {
listen 443 ssl;
ssl_reject_handshake on;
}
server {
listen 443 ssl;
server_name ~^;
ssl_certificate example.crt;
ssl_certificate_key example.key;
}
Additionally, the ssl_reject_handshake directive makes configuring
certificates for the default server block optional. If no certificates
are configured in the default server for a given listening socket,
certificates must be defined in all non-default server blocks with
the listening socket in question.
Similarly to ssl_conf_command, proxy_ssl_conf_command can be used to
set arbitrary OpenSSL configuration parameters as long as nginx is
compiled with OpenSSL 1.0.2 or later, when connecting to upstream
servers with SSL. Full list of available configuration commands
can be found in the SSL_CONF_cmd manual page
(https://www.openssl.org/docs/man1.1.1/man3/SSL_CONF_cmd.html).
Similarly to ssl_conf_command, proxy_ssl_conf_command (grpc_ssl_conf_command,
uwsgi_ssl_conf_command) can be used to set arbitrary OpenSSL configuration
parameters as long as nginx is compiled with OpenSSL 1.0.2 or later,
when connecting to upstream servers with SSL. Full list of available
configuration commands can be found in the SSL_CONF_cmd manual page
(https://www.openssl.org/docs/man1.1.1/man3/SSL_CONF_cmd.html).
With the ssl_conf_command directive it is now possible to set
arbitrary OpenSSL configuration parameters as long as nginx is compiled
with OpenSSL 1.0.2 or later. Full list of available configuration
commands can be found in the SSL_CONF_cmd manual page
(https://www.openssl.org/docs/man1.1.1/man3/SSL_CONF_cmd.html).
In particular, this allows configuring PrioritizeChaCha option
(ticket #1445):
ssl_conf_command Options PrioritizeChaCha;
It can be also used to configure TLSv1.3 ciphers in OpenSSL,
which fails to configure them via the SSL_CTX_set_cipher_list()
interface (ticket #1529):
ssl_conf_command Ciphersuites TLS_CHACHA20_POLY1305_SHA256;
Configuration commands are applied after nginx own configuration
for SSL, so they can be used to override anything set by nginx.
Note though that configuring OpenSSL directly with ssl_conf_command
might result in a behaviour nginx does not expect, and should be
done with care.
With this change, it is now possible to use ngx_conf_merge_ptr_value()
to merge keyval arrays. This change actually follows much earlier
changes in ngx_conf_merge_ptr_value() and ngx_conf_set_str_array_slot()
in 1452:cd586e963db0 (0.6.10) and 1701:40d004d95d88 (0.6.22).
To preserve compatibility with existing 3rd party modules, both NULL
and NGX_CONF_UNSET_PTR are accepted for now.
13.2.4. Limiting Ranges by Tracking ACK Frames
When a packet containing an ACK frame is sent, the largest
acknowledged in that frame may be saved. When a packet containing an
ACK frame is acknowledged, the receiver can stop acknowledging
packets less than or equal to the largest acknowledged in the sent
ACK frame.
The history of acknowledged packet is kept in send context as ranges.
Up to NGX_QUIC_MAX_RANGES ranges is stored.
As a result, instead of separate ack frames, single frame with ranges
is sent.
Per draft-ietf-quic-transport-32 on the topic:
: Similarly, a server MUST expand the payload of all UDP datagrams carrying
: ack-eliciting Initial packets to at least the smallest allowed maximum
: datagram size of 1200 bytes.
The history of acknowledged packet is kept in send context as ranges.
Up to NGX_QUIC_MAX_RANGES ranges is stored.
As a result, instead of separate ack frames, single frame with ranges
is sent.
Previously, if there were multiple limits configured, errors in
ngx_http_complex_value() during processing of a non-first limit
resulted in reference count leak in shared memory nodes of already
processed limits. Fix is to explicity unlock relevant nodes, much
like we do when rejecting requests.
The proxy_smtp_auth directive instructs nginx to authenticate users
on backend via the AUTH command (using the PLAIN SASL mechanism),
similar to what is normally done for IMAP and POP3.
If xclient is enabled along with proxy_smtp_auth, the XCLIENT command
won't try to send the LOGIN parameter.
In 7717:e3e8b8234f05, the 1st bit was incorrectly used. It shouldn't
be used for bitmask values, as it is used by NGX_CONF_BITMASK_SET.
Additionally, special value "off" added to make it possible to clear
inherited userid_flags value.
The "false" parameter of the proxy_redirect directive is deprecated.
Warning has been emitted since c2230102df6f (0.7.54).
The "off" parameter of the proxy_redirect, proxy_cookie_domain, and
proxy_cookie_path directives tells nginx not to inherit the
configuration from the previous configuration level.
Previously, after specifying the directive with the "off" parameter,
any other directives were ignored, and syntax checking was disabled.
The syntax was enforced to allow either one directive with the "off"
parameter, or several directives with other parameters.
Also, specifying "proxy_redirect default foo" no longer works like
"proxy_redirect default".
Previously, this field was not set while creating a QUIC stream connection.
As a result, calling ngx_connection_local_sockaddr() led to getsockname()
bad descriptor error.
If client acknowledged an Initial packet with CRYPTO frame and then
sent another Initial packet containing duplicate CRYPTO again, this
could result in resending frames off the empty send queue.
The new "quic_stateless_reset_token_key" directive is added. It sets the
endpoint key used to generate stateless reset tokens and enables feature.
If the endpoint receives short-header packet that can't be matched to
existing connection, a stateless reset packet is generated with
a proper token.
If a valid stateless reset token is found in the incoming packet,
the connection is closed.
Example configuration:
http {
quic_stateless_reset_token_key "foo";
...
}
The flag is tied to the initial secret creation. The presence of c->quic
pointer is sufficient to enable execution of ngx_quic_close_quic().
The ngx_quic_new_connection() function now returns the allocated quic
connection object and the c->quic pointer is set by the caller.
If an early error occurs before secrets initialization (i.e. in cases
of invalid retry token or nginx exiting), it is still possible to
generate an error response by trying to initialize secrets directly
in the ngx_quic_send_cc() function.
Before the change such early errors failed to send proper connection close
message and logged an error.
An auxilliary ngx_quic_init_secrets() function is introduced to avoid
verbose call to ngx_quic_set_initial_secret() requiring local variable.
All packet header parsing is now performed by ngx_quic_parse_packet()
function, located in the ngx_quic_transport.c file.
The packet processing is centralized in the ngx_quic_process_packet()
function which decides if the packet should be accepted, ignored or
connection should be closed, depending on the connection state.
As a result of refactoring, behavior has changed in some places:
- minimal size of Initial packet is now always tested
- connection IDs are always tested in existing connections
- old keys are discarded on encryption level switch
Now flags are processed in ngx_quic_input(), and raw->pos points to the first
byte after the flags. Redundant checks from ngx_quic_parse_short_header() and
ngx_quic_parse_long_header() are removed.
Previously, when a packet was declared lost, another packet was sent with the
same frames. Now lost frames are moved to the output frame queue and push
event is posted. This has the advantage of forming packets with more frames
than before.
Also, the start argument is removed from the ngx_quic_resend_frames()
function as excess information.
Previously the default server configuration context was used until the
:authority or host header was parsed. This led to using the configuration
parameters like client_header_buffer_size or request_pool_size from the default
server rather than from the server selected by SNI.
Also, the switch to the right server log is implemented. This issue manifested
itself as QUIC stream being logged to the default server log until :authority
or host is parsed.
Initially, client certificate verification didn't work due to the missing
hc->ssl on a QUIC stream, which is started to be set in 7738:7f0981be07c4.
Then it was lost in 7999:0d2b2664b41c introducing "quic" listen parameter.
This change re-adds hc->ssl back for all QUIC connections, similar to SSL.
As per HTTP/3 draft 30, section 7.2.8:
Frame types that were used in HTTP/2 where there is no corresponding
HTTP/3 frame have also been reserved (Section 11.2.1). These frame
types MUST NOT be sent, and their receipt MUST be treated as a
connection error of type H3_FRAME_UNEXPECTED.
In rare cases, such as memory allocation failure, SSL_set_SSL_CTX() returns
NULL, which could mean that a different SSL configuration has not been set.
Note that this new behaviour seemingly originated in OpenSSL-1.1.0 release.
HTTP/2 code failed to run posted requests after calling the request body
handler, and this resulted in connection hang if a subrequest was created
in the body handler and no other actions were made.
If 400 errors were redirected to an upstream server using the error_page
directive, DATA frames from the client might cause segmentation fault
due to null pointer dereference. The bug had appeared in 6989:2c4dbcd6f2e4
(1.13.0).
Fix is to skip such frames in ngx_http_v2_state_read_data() (similarly
to 7561:9f1f9d6e056a). With the fix, behaviour of 400 errors in HTTP/2
is now similar to one in HTTP/1.x, that is, nginx doesn't try to read the
request body.
Note that proxying 400 errors, as well as other early stage errors, to
upstream servers might not be a good idea anyway. These errors imply
that reading and processing of the request (and the request headers)
wasn't complete, and proxying of such incomplete request might lead to
various errors.
Reported by Chenglong Zhang.
This fixes "SSL_shutdown() failed (SSL: ... bad write retry)" errors
as observed on the second SSL_shutdown() call after SSL shutdown fixes in
09fb2135a589 (1.19.2), notably when HTTP/2 connections are closed due
to read timeouts while there are incomplete writes.
This fixes "SSL_shutdown() failed (SSL: ... bad write retry)" errors
as observed on the second SSL_shutdown() call after SSL shutdown fixes in
09fb2135a589 (1.19.2), notably when sending fails in ngx_http_test_expect(),
similarly to ticket #1194.
Note that there are some places where c->error is misused to prevent
further output, such as ngx_http_v2_finalize_connection() if there
are pending streams, or in filter finalization. These places seem
to be extreme enough to don't care about missing shutdown though.
For example, filter finalization currently prevents keepalive from
being used.
The c->read->ready and c->write->ready flags need to be cleared to ensure
that appropriate read or write events will be reported by kernel. Without
this, SSL shutdown might wait till the timeout after blocking on writing
or reading even if there is a socket activity.
OpenSSL 1.1.1 fails to return SSL_ERROR_SYSCALL if an error happens
during SSL_write() after close_notify alert from the peer, and returns
SSL_ERROR_ZERO_RETURN instead. Broken by this commit, which removes
the "i == 0" check around the SSL_RECEIVED_SHUTDOWN one:
https://git.openssl.org/?p=openssl.git;a=commitdiff;h=8051ab2
In particular, if a client closed the connection without reading
the response but with properly sent close_notify alert, this resulted in
unexpected "SSL_write() failed while ..." critical log message instead
of correct "SSL_write() failed (32: Broken pipe)" at the info level.
Since SSL_ERROR_ZERO_RETURN cannot be legitimately returned after
SSL_write(), the fix is to convert all SSL_ERROR_ZERO_RETURN errors
after SSL_write() to SSL_ERROR_SYSCALL.
If the variant hash doesn't match one we used as a secondary cache key,
we switch back to the original key. In this case, c->body_start was kept
updated from an existing cache node overwriting the new response value.
After file cache update, it led to discrepancy between a cache node and
cache file seen as critical errors "file cache .. has too long header".
As per HTTP/3 draft 29, section 4.1:
Frames of unknown types (Section 9), including reserved frames
(Section 7.2.8) MAY be sent on a request or push stream before,
after, or interleaved with other frames described in this section.
Also, trailers frame is now used as an indication of the request body end.
While for HTTP/1 unexpected eof always means an error, for HTTP/3 an eof right
after a DATA frame end means the end of the request body. For this reason,
since adding HTTP/3 support, eof no longer produced an error right after recv()
but was passed to filters which would make a decision. This decision was made
in ngx_http_parse_chunked() and ngx_http_v3_parse_request_body() based on the
b->last_buf flag.
Now that since 0f7f1a509113 (1.19.2) rb->chunked->length is a lower threshold
for the expected number of bytes, it can be set to zero to indicate that more
bytes may or may not follow. Now it's possible to move the check for eof from
parser functions to ngx_http_request_body_chunked_filter() and clean up the
parsing code.
Also, in the default branch, in case of eof, the following three things
happened, which were replaced with returning NGX_ERROR while implementing
HTTP/3:
- "client prematurely closed connection" message was logged
- c->error flag was set
- NGX_HTTP_BAD_REQUEST was returned
The change brings back this behavior for HTTP/1 as well as HTTP/3.
If a packet sent in response to an initial client packet was lost, then
successive client initial packets were dropped by nginx with the unexpected
dcid message logged. This was because the new DCID generated by the server was
not available to the client.
The check tested the total size of a packet header and unprotected packet
payload, which doesn't include the packet number length and expansion of
the packet protection AEAD. If the packet was corrupted, it could cause
false triggering of the condition due to unsigned type underflow leading
to a connection error.
Existing checks for the QUIC header and protected packet payload lengths
should be enough.
From quic-tls draft, section 5.4.2:
An endpoint MUST discard packets that are not long enough to contain
a complete sample.
The check includes the Packet Number field assumed to be 4 bytes long.
During long packet header parsing, pkt->len is updated with the Length
field value that is used to find next coalesced packets in a datagram.
For short packets it still contained the whole QUIC packet size.
This change uniforms packet length handling to always contain the total
length of the packet number and protected packet payload in pkt->len.
Previously STOP_SENDING was sent to client upon stream closure if rev->eof and
rev->error were not set. This was an indirect indication that no RESET_STREAM
or STREAM fin has arrived. But it is indeed possible that rev->eof is not set,
but STREAM fin has already been received, just not read out by the application.
In this case sending STOP_SENDING does not make sense and can be misleading for
some clients.
The peer may issue additional connection IDs up to the limit defined by
transport parameter "active_connection_id_limit", using NEW_CONNECTION_ID
frames, and retire such IDs using RETIRE_CONNECTION_ID frame.
It is required to distinguish internal errors from corrupted packets and
perform actions accordingly: drop the packet or close the connection.
While there, made processing of ngx_quic_decrypt() erorrs similar and
removed couple of protocol violation errors.
quic-transport
5.2:
Packets that are matched to an existing connection are discarded if
the packets are inconsistent with the state of that connection.
5.2.2:
Servers MUST drop incoming packets under all other circumstances.
The removal of QUIC packet protection depends on the largest packet number
received. When a garbage packet was received, the decoder still updated the
largest packet number from that packet. This could affect removing protection
from subsequent QUIC packets.
As per HTTP/3 draft 29, section 4.1:
When the server does not need to receive the remainder of the request,
it MAY abort reading the request stream, send a complete response, and
cleanly close the sending part of the stream.
On QUIC connections, SSL_shutdown() is used to call the send_alert callback
to send a CONNECTION_CLOSE frame. The reverse side is handled by other means.
At least BoringSSL doesn't differentiate whether this is a QUIC SSL method,
so waiting for the peer's close_notify alert should be explicitly disabled.
The logical quic connection state is tested by handler functions that
process corresponding types of packets (initial/handshake/application).
The packet is declined if state is incorrect.
No timeout is required for the input queue.
If a client attemtps to start a new connection with unsupported version,
a version negotiation packet is sent that contains a list of supported
versions (currently this is a single version, selected at compile time).
The function ngx_http_upstream_check_broken_connection() terminates the HTTP/1
request if client sends eof. For QUIC (including HTTP/3) the c->write->error
flag is now checked instead. This flag is set when the entire QUIC connection
is closed or STOP_SENDING was received from client.
Previously the request body DATA frame header was read by one byte because
filters were called only when the requested number of bytes were read. Now,
after 08ff2e10ae92 (1.19.2), filters are called after each read. More bytes
can be read at once, which simplifies and optimizes the code.
This also reduces diff with the default branch.
Previously, such packets weren't handled as the resulting zero remaining time
prevented setting the loss detection timer, which, instead, could be disarmed.
For implementation details, see quic-recovery draft 29, appendix A.10.
The PTO handler is split into separate PTO and loss detection handlers
that operate interchangeably depending on which timer should be set.
The present ngx_quic_lost_handler is now only used for packet loss detection.
It replaces ngx_quic_pto_handler if there are packets preceeding largest_ack.
Once there is no more such packets, ngx_quic_pto_handler is installed again.
Probes carry unacknowledged data previously sent in the oldest packet number,
one per each packet number space. That is, it could be up to two probes.
PTO backoff is now increased before scheduling next probes.
In particular, this prevents declaring packet number 0 as lost if
there aren't yet any acknowledgements in this packet number space.
For example, only Initial packets were acknowledged in handshake.
Previously a single STREAM frame was created for each buffer in stream output
chain which is wasteful with respect to memory. The following changes were
made in the stream send code:
- ngx_quic_stream_send_chain() no longer calls ngx_quic_stream_send() and got
a separate implementation that coalesces neighbouring buffers into a single
frame
- the new ngx_quic_stream_send_chain() respects the limit argument, which fixes
sendfile_max_chunk and limit_rate
- ngx_quic_stream_send() is reimplemented to call ngx_quic_stream_send_chain()
- stream frame size limit is moved out to a separate function
ngx_quic_max_stream_frame()
- flow control is moved out to a separate function ngx_quic_max_stream_flow()
- ngx_quic_stream_send_chain() is relocated next to ngx_quic_stream_send()
Reworked connections reuse, so closing connections is attempted in
advance, as long as number of free connections is less than 1/16 of
worker connections configured. This ensures that new connections can
be handled even if closing a reusable connection requires some time,
for example, for a lingering close (ticket #2017).
The 1/16 ratio is selected to be smaller than 1/8 used for disabling
accept when working with accept mutex, so nginx will try to balance
new connections to different workers first, and will start reusing
connections only if this won't help.
Previously, reusing connections happened silently and was only
visible in monitoring systems. This was shown to be not very user-friendly,
and administrators often didn't realize there were too few connections
available to withstand the load, and configured timeouts (keepalive_timeout
and http2_idle_timeout) were effectively reduced to keep things running.
To provide at least some information about this, a warning is now logged
(at most once per second, to avoid flooding the logs).
Sending shutdown when ngx_http_test_reading() detects the connection is
closed can result in "SSL_shutdown() failed (SSL: ... bad write retry)"
critical log messages if there are blocked writes.
Fix is to avoid sending shutdown via the c->ssl->no_send_shutdown flag,
similarly to how it is done in ngx_http_keepalive_handler() for kqueue
when pending EOF is detected.
Reported by Jan Prachař
(http://mailman.nginx.org/pipermail/nginx-devel/2018-December/011702.html).
Without the flag, SSL shutdown is attempted on such connections,
resulting in useless work and/or bogus "SSL_shutdown() failed
(SSL: ... bad write retry)" critical log messages if there are
blocked writes.
Previously, bidirectional shutdown never worked, due to two issues
in the code:
1. The code only tested SSL_ERROR_WANT_READ and SSL_ERROR_WANT_WRITE
when there was an error in the error queue, which cannot happen.
The bug was introduced in an attempt to fix unexpected error logging
as reported with OpenSSL 0.9.8g
(http://mailman.nginx.org/pipermail/nginx/2008-January/003084.html).
2. The code never called SSL_shutdown() for the second time to wait for
the peer's close_notify alert.
This change fixes both issues.
Note that after this change bidirectional shutdown is expected to work for
the first time, so c->ssl->no_wait_shutdown now makes a difference. This
is not a problem for HTTP code which always uses c->ssl->no_wait_shutdown,
but might be a problem for stream and mail code, as well as 3rd party
modules.
To minimize the effect of the change, the timeout, which was used to be 30
seconds and not configurable, though never actually used, is now set to
3 seconds. It is also expanded to apply to both SSL_ERROR_WANT_READ and
SSL_ERROR_WANT_WRITE, so timeout is properly set if writing to the socket
buffer is not possible.
If some additional data from a pipelined request happens to be
read into the body buffer, we copy it to r->header_in or allocate
an additional large client header buffer for it.
This ensures that copying won't write more than the buffer size
even if the buffer comes from hc->free and it is smaller than the large
client header buffer size in the virtual host configuration. This might
happen if size of large client header buffers is different in name-based
virtual hosts, similarly to the problem with number of buffers fixed
in 6926:e662cbf1b932.
Creating client-initiated streams is moved from ngx_quic_handle_stream_frame()
to a separate function ngx_quic_create_client_stream(). This function is
responsible for creating streams with lower ids as well.
Also, simplified and fixed initial data buffering in
ngx_quic_handle_stream_frame(). It is now done before calling the initial
handler as the handler can destroy the stream.
Previously this function generated an error trying to figure out if client shut
down the write end of the connection. The reason for this error was that a
QUIC stream has no socket descriptor. However checking for eof is not the
right thing to do for an HTTP/3 QUIC stream since HTTP/3 clients are expected
to shut down the write end of the stream after sending the request.
Now the function handles QUIC streams separately. It checks if c->read->error
is set. The error flags for c->read and c->write are now set for all streams
when closing the QUIC connection instead of setting the pending_eof flag.
According to quic-transport draft 29, section 19.3.1:
The value of the Gap field establishes the largest packet number
value for the subsequent ACK Range using the following formula:
largest = previous_smallest - gap - 2
Thus, given a largest packet number for the range, the smallest value
is determined by the formula:
smallest = largest - ack_range
While here, changed min/max to uint64_t for consistency.
A QUIC stream could be destroyed by handler while in ngx_quic_stream_input().
To detect this, ngx_quic_find_stream() is used to check that it still exists.
Previously, a stream id was passed to this routine off the frame structure.
In case of stream cleanup, it is freed along with other frames belonging to
the stream on cleanup. Then, a cleanup handler reuses last frames to update
MAX_STREAMS and serve other purpose. Thus, ngx_quic_find_stream() is passed
a reused frame with zeroed out part pointed by stream_id. If a stream with
id 0x0 still exists, this leads to use-after-free.
After 05e42236e95b (1.19.1) responses with extra data might result in
zero size buffers being generated and "zero size buf" alerts in writer
(if f->rest happened to be 0 when processing additional stdout data).
The limits on active bidi and uni client streams are maintained at their
initial values initial_max_streams_bidi and initial_max_streams_uni by sending
a MAX_STREAMS frame upon each client stream closure.
Also, the following is changed for data arriving to non-existing streams:
- if a stream was already closed, such data is ignored
- when creating a new stream, all streams of the same type with lower ids are
created too
Previously, the document generated by the xslt filter was always fully sent
to client even if a range was requested and response status was 206 with
appropriate Content-Range.
The xslt module is unable to serve a range because of suspending the header
filter chain. By the moment full response xml is buffered by the xslt filter,
range header filter is not called yet, but the range body filter has already
been called and did nothing.
The fix is to disable ranges by resetting the r->allow_ranges flag much like
the image filter that employs a similar technique.
The ngx_http_perl_module module doesn't have a notion of including additional
search paths through --with-cc-opt, which results in compile error incomplete
type 'enum ssl_encryption_level_t' when building nginx without QUIC support.
The enum is visible from quic event headers and eventually pollutes ngx_core.h.
The fix is to limit including headers to compile units that are real consumers.
According to quic-transport draft 29, section 21.12.1.1:
Prior to validation, endpoints are limited in what they are able to
send. During the handshake, a server cannot send more than three
times the data it receives; clients that initiate new connections or
migrate to a new network path are limited.
The ngx_quic_queue_frame() functions puts a frame into send queue and
schedules a push timer to actually send data.
The patch adds tracking for data amount in the queue and sends data
immediately if amount of data exceeds limit.
Instead of timer-based retransmissions with constant packet lifetime,
this patch implements ack-based loss detection and probe timeout
for the cases, when no ack is received, according to the quic-recovery
draft 29.
The c->quic->retransmit timer is now called "pto".
The ngx_quic_retransmit() function is renamed to "ngx_quic_detect_lost()".
This is a preparation for the following patches.
According to the quic-recovery 29, Section 5: Estimating the Round-Trip Time.
Currently, integer arithmetics is used, which loses sub-millisecond accuracy.
The slice filter allows ranges for the response by setting the r->allow_ranges
flag, which enables the range filter. If the range was not requested, the
range filter adds an Accept-Ranges header to the response to signal the
support for ranges.
Previously, if an Accept-Ranges header was already present in the first slice
response, client received two copies of this header. Now, the slice filter
removes the Accept-Ranges header from the response prior to setting the
r->allow_ranges flag.
As long as the "Content-Length" header is given, we now make sure
it exactly matches the size of the response. If it doesn't,
the response is considered malformed and must not be forwarded
(https://tools.ietf.org/html/rfc7540#section-8.1.2.6). While it
is not really possible to "not forward" the response which is already
being forwarded, we generate an error instead, which is the closest
equivalent.
Previous behaviour was to pass everything to the client, but this
seems to be suboptimal and causes issues (ticket #1695). Also this
directly contradicts HTTP/2 specification requirements.
Note that the new behaviour for the gRPC proxy is more strict than that
applied in other variants of proxying. This is intentional, as HTTP/2
specification requires us to do so, while in other types of proxying
malformed responses from backends are well known and historically
tolerated.
Previous behaviour was to pass everything to the client, but this
seems to be suboptimal and causes issues (ticket #1695). Fix is to
drop extra data instead, as it naturally happens in most clients.
Additionally, we now also issue a warning if the response is too
short, and make sure the fact it is truncated is propagated to the
client. The u->error flag is introduced to make it possible to
propagate the error to the client in case of unbuffered proxying.
For responses to HEAD requests there is an exception: we do allow
both responses without body and responses with body matching the
Content-Length header.
Previous behaviour was to pass everything to the client, but this
seems to be suboptimal and causes issues (ticket #1695). Fix is to
drop extra data instead, as it naturally happens in most clients.
This change covers generic buffered and unbuffered filters as used
in the scgi and uwsgi modules. Appropriate input filter init
handlers are provided by the scgi and uwsgi modules to set corresponding
lengths.
Note that for responses to HEAD requests there is an exception:
we do allow any response length. This is because responses to HEAD
requests might be actual full responses, and it is up to nginx
to remove the response body. If caching is enabled, only full
responses matching the Content-Length header will be cached
(see b779728b180c).
Previously, additional data after final chunk was either ignored
(in the same buffer, or during unbuffered proxying) or sent to the
client (in the next buffer already if it was already read from the
socket). Now additional data are properly detected and ignored
in all cases. Additionally, a warning is now logged and keepalive
is disabled in the connection.
Previous behaviour was to pass everything to the client, but this
seems to be suboptimal and causes issues (ticket #1695). Fix is to
drop extra data instead, as it naturally happens in most clients.
If a memcached response was followed by a correct trailer, and then
the NUL character followed by some extra data - this was accepted by
the trailer checking code. This in turn resulted in ctx->rest underflow
and caused negative size buffer on the next reading from the upstream,
followed by the "negative size buf in writer" alert.
Fix is to always check for too long responses, so a correct trailer cannot
be followed by extra data.
After sending the GOAWAY frame, a connection is now closed using
the lingering close mechanism.
This allows for the reliable delivery of the GOAWAY frames, while
also fixing connection resets observed when http2_max_requests is
reached (ticket #1250), or with graceful shutdown (ticket #1544),
when some additional data from the client is received on a fully
closed connection.
For HTTP/2, the settings lingering_close, lingering_timeout, and
lingering_time are taken from the "server" level.
Previously, the expression (ch & 0x7f) was promoted to a signed integer.
Depending on the platform, the size of this integer could be less than 8 bytes,
leading to overflow when handling the higher bits of the result. Also, sign
bit of this integer could be replicated when adding to the 64-bit st->value.
Previously errors led only to closing streams.
To simplify closing QUIC connection from a QUIC stream context, new macro
ngx_http_v3_finalize_connection() is introduced. It calls
ngx_quic_finalize_connection() for the parent connection.
The function finalizes QUIC connection with an application protocol error
code and sends a CONNECTION_CLOSE frame with type=0x1d.
Also, renamed NGX_QUIC_FT_CONNECTION_CLOSE2 to NGX_QUIC_FT_CONNECTION_CLOSE_APP.
Previously dynamic table was not functional because of zero limit on its size
set by default. Now the following changes enable it:
- new directives to set SETTINGS_QPACK_MAX_TABLE_CAPACITY and
SETTINGS_QPACK_BLOCKED_STREAMS
- send settings with SETTINGS_QPACK_MAX_TABLE_CAPACITY and
SETTINGS_QPACK_BLOCKED_STREAMS to the client
- send Insert Count Increment to the client
- send Header Acknowledgement to the client
- evict old dynamic table entries on overflow
- decode Required Insert Count from client
- block stream if Required Insert Count is not reached
Using SSL_CTX_set_verify(SSL_VERIFY_PEER) implies that OpenSSL will
send a certificate request during an SSL handshake, leading to unexpected
certificate requests from browsers as long as there are any client
certificates installed. Given that ngx_ssl_trusted_certificate()
is called unconditionally by the ngx_http_ssl_module, this affected
all HTTPS servers. Broken by 699f6e55bbb4 (not released yet).
Fix is to set verify callback in the ngx_ssl_trusted_certificate() function
without changing the verify mode.
Client streams may send literal strings which are now limited in size by the
new directive. The default value is 4096.
The directive is similar to HTTP/2 directive http2_max_field_size.
So that connections are protected from failing from on-path attacks.
Decryption failure of long packets used during handshake still leads
to connection close since it barely makes sense to handle them there.
A previously used undefined error code is now replaced with the generic one.
Note that quic-transport prescribes keeping connection intact, discarding such
QUIC packets individually, in the sense that coalesced packets could be there.
This is selectively handled in the next change.
The patch removes remnants of the old state tracking mechanism, which did
not take into account assimetry of read/write states and was not very
useful.
The encryption state now is entirely tracked using SSL_quic_read/write_level().
quic-transport draft 29:
section 7:
* authenticated negotiation of an application protocol (TLS uses
ALPN [RFC7301] for this purpose)
...
Endpoints MUST explicitly negotiate an application protocol. This
avoids situations where there is a disagreement about the protocol
that is in use.
section 8.1:
When using ALPN, endpoints MUST immediately close a connection (see
Section 10.3 of [QUIC-TRANSPORT]) with a no_application_protocol TLS
alert (QUIC error code 0x178; see Section 4.10) if an application
protocol is not negotiated.
Changes in ngx_quic_close_quic() function are required to avoid attempts
to generated and send packets without proper keys, what happens in case
of failed ALPN check.
quic-transport draft 29, section 14:
QUIC depends upon a minimum IP packet size of at least 1280 bytes.
This is the IPv6 minimum size [RFC8200] and is also supported by most
modern IPv4 networks. Assuming the minimum IP header size, this
results in a QUIC maximum packet size of 1232 bytes for IPv6 and 1252
bytes for IPv4.
Since the packet size can change during connection lifetime, the
ngx_quic_max_udp_payload() function is introduced that currently
returns minimal allowed size, depending on address family.
quic-tls, 8.2:
The quic_transport_parameters extension is carried in the ClientHello
and the EncryptedExtensions messages during the handshake. Endpoints
MUST send the quic_transport_parameters extension; endpoints that
receive ClientHello or EncryptedExtensions messages without the
quic_transport_parameters extension MUST close the connection with an
error of type 0x16d (equivalent to a fatal TLS missing_extension
alert, see Section 4.10).
Clearing cache based on free space left on a file system is
expected to allow better disk utilization in some cases, notably
when disk space might be also used for something other than nginx
cache (including nginx own temporary files) and while loading
cache (when cache size might be inaccurate for a while, effectively
disabling max_size cache clearing).
Based on a patch by Adam Bambuch.
With XFS, using "allocsize=64m" mount option results in large preallocation
being reported in the st_blocks as returned by fstat() till the file is
closed. This in turn results in incorrect cache size calculations and
wrong clearing based on max_size.
To avoid too aggressive cache clearing on such volumes, st_blocks values
which result in sizes larger than st_size and eight blocks (an arbitrary
limit) are no longer trusted, and we use st_size instead.
The ngx_de_fs_size() counterpart is intentionally not modified, as
it is used on closed files and hence not affected by this problem.
NFS on Linux is known to report wsize as a block size (in both f_bsize
and f_frsize, both in statfs() and statvfs()). On the other hand,
typical file system block sizes on Linux (ext2/ext3/ext4, XFS) are limited
to pagesize. (With FAT, block sizes can be at least up to 512k in
extreme cases, but this doesn't really matter, see below.)
To avoid too aggressive cache clearing on NFS volumes on Linux, block
sizes larger than pagesize are now ignored.
Note that it is safe to ignore large block sizes. Since 3899:e7cd13b7f759
(1.0.1) cache size is calculated based on fstat() st_blocks, and rounding
to file system block size is preserved mostly for Windows.
Note well that on other OSes valid block sizes seen are at least up
to 65536. In particular, UFS on FreeBSD is known to work well with block
and fragment sizes set to 65536.
When validating second and further certificates, ssl callback could be called
twice to report the error. After the first call client connection is
terminated and its memory is released. Prior to the second call and in it
released connection memory is accessed.
Errors triggering this behavior:
- failure to create the request
- failure to start resolving OCSP responder name
- failure to start connecting to the OCSP responder
The fix is to rearrange the code to eliminate the second call.
The flush flag was not set when forwarding the request body to the uwsgi
server. When using uwsgi_pass suwsgi://..., this causes the uwsgi server
to wait indefinitely for the request body and eventually time out due to
SSL buffering.
This is essentially the same change as 4009:3183165283cc, which was made
to ngx_http_proxy_module.c.
This will fix the uwsgi bug https://github.com/unbit/uwsgi/issues/1490.
This is a temporary workaround, proper retransmission mechanism based on
quic-recovery rfc draft is yet to be implemented.
Currently hardcoded value is too small for real networks. The patch
sets static PTO, considering rtt of ~333ms, what gives about 1s.
This ensures that certificate verification is properly logged to debug
log during upstream server certificate verification. This should help
with debugging various certificate issues.
Listening UNIX sockets were not removed on graceful shutdown, preventing
the next runs. The fix is to replace the custom socket closing code in
ngx_master_process_cycle() by the ngx_close_listening_sockets() call.
When changing binary, sending a SIGTERM to the new binary's master process
should not remove inherited UNIX sockets unless the old binary's master
process has exited.
Also, if both are present, require that they have the same value. These
requirements are specified in HTTP/3 draft 28.
Current implementation of HTTP/2 treats ":authority" and "Host"
interchangeably. New checks only make sure at least one of these values is
present in the request. A similar check existed earlier and was limited only
to HTTP/1.1 in 38c0898b6df7.
The flags was originally added by 8f038068f4bc, and is propagated correctly
in the stream module. With QUIC introduction, http module now uses datagram
sockets as well, thus the fix.
Previously, invalid connection preface errors were only logged at debug
level, providing no visible feedback, in particular, when a plain text
HTTP/2 listening socket is erroneously used for HTTP/1.x connections.
Now these are explicitly logged at the info level, much like other
client-related errors.
When enabled, certificate status is stored in cache and is used to validate
the certificate in future requests.
New directive ssl_ocsp_cache is added to configure the cache.
OCSP validation for client certificates is enabled by the "ssl_ocsp" directive.
OCSP responder can be optionally specified by "ssl_ocsp_responder".
When session is reused, peer chain is not available for validation.
If the verified chain contains certificates from the peer chain not available
at the server, validation will fail.
Previously only the first responder address was used per each stapling update.
Now, in case of a network or parsing error, next address is used.
This also fixes the issue with unsupported responder address families
(ticket #1330).
Preserving pointers within the client buffer is not needed for HTTP/3 because
all data is either allocated from pool or static. Unlike with HTTP/1, data
typically cannot be referenced directly within the client buffer. Trying to
preserve NULLs or external pointers lead to broken pointers.
Also, reverted changes in ngx_http_alloc_large_header_buffer() not relevant
for HTTP/3 to minimize diff to mainstream.
New field r->parse_start is introduced to substitute r->request_start and
r->header_name_start for request length accounting. These fields only work for
this purpose in HTTP/1 because HTTP/1 request line and header line start with
these values.
Also, error logging is now fixed to output the right part of the request.
As per HTTP/3 draft 27, a request or response containing uppercase header
field names MUST be treated as malformed. Also, existing rules applied
when parsing HTTP/1 header names are also applied to HTTP/3 header names:
- null character is not allowed
- underscore character may or may not be treated as invalid depending on the
value of "underscores_in_headers"
- all non-alphanumeric characters with the exception of '-' are treated as
invalid
Also, the r->locase_header field is now filled while parsing an HTTP/3
header.
Error logging for invalid headers is fixed as well.
The first one parses pseudo-headers and is analagous to the request line
parser in HTTP/1. The second one parses regular headers and is analogous to
the header parser in HTTP/1.
Additionally, error handling of client passing malformed uri is now fixed.
The function ngx_http_parse_chunked() is also called from the proxy module to
parse the upstream response. It should always parse HTTP/1 body in this case.
According to quic-transport draft 28 section 10.3.1:
When sending CONNECTION_CLOSE, the goal is to ensure that the peer
will process the frame. Generally, this means sending the frame in a
packet with the highest level of packet protection to avoid the
packet being discarded. After the handshake is confirmed (see
Section 4.1.2 of [QUIC-TLS]), an endpoint MUST send any
CONNECTION_CLOSE frames in a 1-RTT packet. However, prior to
confirming the handshake, it is possible that more advanced packet
protection keys are not available to the peer, so another
CONNECTION_CLOSE frame MAY be sent in a packet that uses a lower
packet protection level.
There is no need in a separate type for the QUIC connection state.
The only state not found in the SSL library is NGX_QUIC_ST_UNAVAILABLE,
which is actually a flag used by the ngx_quic_close_quic() function
to prevent cleanup of uninitialized connection.
Sections 4.10.1 and 4.10.2 of quic transport describe discarding of initial
and handshake keys. Since the keys are discarded, we no longer need
to retransmit packets and corresponding queues should be emptied.
This patch removes previously added workaround that did not require
acknowledgement for initial packets, resulting in avoiding retransmission,
which is wrong because a packet could be lost and we have to retransmit it.
It was possible that retransmit timer was not set after the first
retransmission attempt, due to ngx_quic_retransmit() did not set
wait time properly, and the condition in retransmit handler was incorrect.
Section 17.2 and 17.3 of QUIC transport:
Fixed bit: Packets containing a zero value for this bit are not
valid packets in this version and MUST be discarded.
Reserved bit: An endpoint MUST treat receipt of a packet that has
a non-zero value for these bits, after removing both packet and
header protection, as a connection error of type PROTOCOL_VIOLATION.
When an error occurs, then c->quic->error field may be populated
with an appropriate error code, and the CONNECTION CLOSE frame will be
sent to the peer before the connection is closed. Otherwise, the error
treated as internal and INTERNAL_ERROR code is sent.
The pkt->error field is populated by functions processing packets to
indicate an error when it does not fit into pass/fail return status.
As per QUIC transport, the first flight of 0-RTT packets obviously uses same
Destination and Source Connection ID values as the client's first Initial.
The fix is to match 0-RTT against original DCID after it has been switched.
The ordered frame handler is always called for the existing stream, as it is
allocated from this stream. Instead of searching stream by id, pointer to the
stream node is passed.
The idea is to skip any zeroes that follow valid QUIC packet. Currently such
behavior can be only observed with Firefox which sends zero-padded initial
packets.
Now there's no need to annotate every frame in ACK-eliciting packet.
Sending ACK was moved to the first place, so that queueing ACK frame
no longer postponed up to the next packet after pushing STREAM frames.
+ added "quic" prefix to all error messages
+ rephrased some messages
+ removed excessive error logging from frame parser
+ added ngx_quic_check_peer() function to check proper source/destination
match and do it one place
- the ngx_quic_hexdump0() macro is renamed to ngx_quic_hexdump();
the original ngx_quic_hexdump() macro with variable argument is
removed, extra information is logged normally, with ngx_log_debug()
- all labels in hex dumps are prefixed with "quic"
- the hexdump format is simplified, length is moved forward to avoid
situations when the dump is truncated, and length is not shown
- ngx_quic_flush_flight() function contents is debug-only, placed under
NGX_DEBUG macro to avoid "unused variable" warnings from compiler
- frame names in labels are capitalized, similar to other places
+ all dumps are moved under one of the following macros (undefined by default):
NGX_QUIC_DEBUG_PACKETS
NGX_QUIC_DEBUG_FRAMES
NGX_QUIC_DEBUG_FRAMES_ALLOC
NGX_QUIC_DEBUG_CRYPTO
+ all QUIC debug messages got "quic " prefix
+ all input frames are reported as "quic frame in FOO_FRAME bar:1 baz:2"
+ all outgoing frames re reported as "quic frame out foo bar baz"
+ all stream operations are prefixed with id, like: "quic stream id 0x33 recv"
+ all transport parameters are prefixed with "quic tp"
(hex dump is moved to caller, to avoid using ngx_cycle->log)
+ packet flags and some other debug messages are updated to
include packet type
As per https://tools.ietf.org/html/rfc7540#section-8.1,
: A server can send a complete response prior to the client
: sending an entire request if the response does not depend on
: any portion of the request that has not been sent and
: received. When this is true, a server MAY request that the
: client abort transmission of a request without error by
: sending a RST_STREAM with an error code of NO_ERROR after
: sending a complete response (i.e., a frame with the
: END_STREAM flag). Clients MUST NOT discard responses as a
: result of receiving such a RST_STREAM, though clients can
: always discard responses at their discretion for other
: reasons.
Previously, RST_STREAM(NO_ERROR) received from upstream after
a frame with the END_STREAM flag was incorrectly treated as an
error. Now, a single RST_STREAM(NO_ERROR) is properly handled.
This fixes problems observed with modern grpc-c [1], as well
as with the Go gRPC module.
[1] https://github.com/grpc/grpc/pull/1661
We always generate stream frames that have length. The 'len' member is used
during parsing incoming frames and can be safely ignored when generating
output.
There are following flags in quic connection:
closing - true, when a connection close is initiated, for whatever reason
draining - true, when a CC frame is received from peer
The following state machine is used for closing:
+------------------+
| I/HS/AD |
+------------------+
| | |
| | V
| | immediate close initiated:
| | reasons: close by top-level protocol, fatal error
| | + sends CC (probably with app-level message)
| | + starts close_timer: 3 * PTO (current probe timeout)
| | |
| | V
| | +---------+ - Reply to input with CC (rate-limited)
| | | CLOSING | - Close/Reset all streams
| | +---------+
| | | |
| V V |
| receives CC |
| | |
idle | |
timer | |
| V |
| +----------+ | - MUST NOT send anything (MAY send a single CC)
| | DRAINING | | - if not already started, starts close_timer: 3 * PTO
| +----------+ | - if not already done, close all streams
| | |
| | |
| close_timer fires
| |
V V
+------------------------+
| CLOSED | - clean up all the resources, drop connection
+------------------------+ state completely
The ngx_quic_close_connection() function gets an "rc" argument, that signals
reason of connection closing:
NGX_OK - initiated by application (i.e. http/3), follow state machine
NGX_DONE - timedout (while idle or draining)
NGX_ERROR - fatal error, destroy connection immediately
The PTO calculations are not yet implemented, hardcoded value of 5s is used.
The function is split into three:
ngx_quic_close_connection() itself cleans up all core nginx things
ngx_quic_close_quic() deals with everything inside c->quic
ngx_quic_close_streams() deals with streams cleanup
The quic and streams cleanup functions may return NGX_AGAIN, thus signalling
that cleanup is not ready yet, and the close cannot continue to next step.
The header size macros for long and short packets were fixed to provide
correct values in bytes.
Currently the sending code limits frames so they don't exceed max_packet_size.
But it does not account the case when a single frame can exceed the limit.
As a result of this patch, big payload (CRYPTO and STREAM) will be split
into a number of smaller frames that fit into advertised max_packet_size
(which specifies final packet size, after encryption).
chrome-unstable 83.0.4103.7 starts with Initial packet number 1.
I couldn't find a proper explanation besides this text in quic-transport:
An endpoint MAY skip packet numbers when sending
packets to detect this (Optimistic ACK Attack) behavior.
+ MAX_STREAM_DATA frame is sent when recv() is performed on stream
The new value is a sum of total bytes received by stream + free
space in a buffer;
The sending of MAX_STREM_DATA frame in response to STREAM_DATA_BLOCKED
frame is adjusted to follow the same logic as above.
+ MAX_DATA frame is sent when total amount of received data is 2x
of current limit. The limit is doubled.
+ Default values of transport parameters are adjusted to more meaningful
values:
initial stream limits are set to quic buffer size instead of
unrealistically small 255.
initial max data is decreased to 16 buffer sizes, in an assumption that
this is enough for a relatively short connection, instead of randomly
chosen big number.
All this allows to initiate a stable flow of streams that does not block
on stream/connection limits (tested with FF 77.0a1 and 100K requests)
Before the patch, full STREAM frame handling was delayed until the frame with
zero offset is received. Only node in the streams tree was created.
This lead to problems when such stream was deleted, in particular, it had no
handlers set for read events.
This patch creates new stream immediately, but delays data delivery until
the proper offset will arrive. This is somewhat similar to how accept()
operation works.
The ngx_quic_add_stream() function is no longer needed and merged into stream
handler. The ngx_quic_stream_input() now only handles frames for existing
streams and does not deal with stream creation.
Frames can still float in the following queues:
- crypto frames reordering queues (one per encryption level)
- moved crypto frames cleanup to the moment where all streams are closed
- stream frames reordering queues (one per packet number namespace)
- frames retransmit queues (one per packet number namespace)
Each stream node now includes incoming frames queue and sent/received counters
for tracking offset. The sent counter is not used, c->sent is used, not like
in crypto buffers, which have no connections.
If offset in CRYPTO frame doesn't match expected, following actions are taken:
a) Duplicate frames or frames within [0...current offset] are ignored
b) New data from intersecting ranges (starts before current_offset, ends
after) is consumed
c) "Future" frames are stored in a sorted queue (min offset .. max offset)
Once a frame is consumed, current offset is updated and the queue is inspected:
we iterate the queue until the gap is found and act as described
above for each frame.
The amount of data in buffered frames is limited by corresponding macro.
The CRYPTO and STREAM frame structures are now compatible: they share
the same set of initial fields. This allows to have code that deals with
both of this frames.
The ordering layer now processes the frame with offset and invokes the
handler when it can organise an ordered stream of data.
Quote: Conceptually, a packet number space is the context in which a packet
can be processed and acknowledged.
ngx_quic_namespace_t => ngx_quic_send_ctx_t
qc->ns => qc->send_ctx
ns->largest => send_ctx->largest_ack
The ngx_quic_ns(level) macro now returns pointer, not just index:
ngx_quic_get_send_ctx(c->quic, level)
ngx_quic_retransmit_ns() => ngx_quic_retransmit()
ngx_quic_output_ns() => ngx_quic_output_frames()
The request processing is delayed by a timer. Since nginx updates
internal time once at the start of each event loop iteration, this
normally ensures constant time delay, adding a mitigation from
time-based attacks.
A notable exception to this is the case when there are no additional
events before the timer expires. To ensure constant-time processing
in this case as well, we trigger an additional event loop iteration
by posting a dummy event for the next event loop iteration.
The offset in client CRYPTO frames is tracked in c->quic->crypto_offset_in.
This means that CRYPTO frames with non-zero offset are now accepted making
possible to finish handshake with client certificates that exceed max packet
size (if no reordering happens).
The c->quic->crypto_offset field is renamed to crypto_offset_out to avoid
confusion with tracking of incoming CRYPTO stream.
+ since number of ranges in unknown, provide a function to parse them once
again in handler to avoid memory allocation
+ ack handler now processes all ranges, not only the first
+ ECN counters are parsed and saved into frame if present
Such frames are grouped together in a switch and just ignored, instead of
closing the connection This may improve test coverage. All such frames
require acknowledgment.
The qc->closing flag is set when a connection close is initiated for the first
time.
No timers will be set if the flag is active.
TODO: this is a temporary solution to avoid running timer handlers after
connection (and it's pool) was destroyed. It looks like currently we have
no clear policy of connection closing in regard to timers.
Found with a previously received Initial packet with ACK only, which
instantiates a new connection but do not produce the handshake keys.
This can be triggered by a fairly well behaving client, if the server
stands behind a load balancer that stripped Initial packets exchange.
Found by F5 test suite.
This makes sending large number of bidirectional stream work within ngtcp2,
which doesn't bother sending optional STREAMS_BLOCKED when exhausted.
This also introduces tracking currently opened and maximum allowed streams.
Currently, the output is called periodically, each 200 ms to invoke
ngx_quic_output() that will push all pending frames into packets.
TODO: implement flags a-là Nagle & co (NO_DELAY/NO_PUSH...)
All frames collected to packet are moved into a per-namespace send queue.
QUIC connection has a timer which fires on the closest max_ack_delay time.
The frame is deleted from the queue when a corresponding packet is acknowledged.
The NGX_QUIC_MAX_RETRANSMISSION is a timeout that defines maximum length
of retransmission of a frame.
The quic->keys[4] array now contains secrets related to the corresponding
encryption level. All protection-level functions get proper keys and do
not need to switch manually between levels.
If early data is accepted, SSL_do_handshake() completes as soon as ClientHello
is processed. SSL_in_init() will report the handshake is still in progress.
Static buffers are used instead in functions where decryption takes place.
The pkt->plaintext points to the beginning of a static buffer.
The pkt->payload.data points to decrypted data actual start.
+ ngx_quic_encrypt():
- no longer accepts pool as argument
- pkt is 1st arg
- payload is passed as pkt->payload
- performs encryption to the specified static buffer
+ ngx_quic_create_long/short_packet() functions:
- single buffer for everything, allocated by caller
- buffer layout is: [ ad | payload | TAG ]
the result is in the beginning of buffer with proper length
- nonce is calculated on stack
- log is passed explicitly, pkt is 1st arg
- no more allocations inside
+ ngx_quic_create_long_header():
- args changed: no need to pass str_t
+ added ngx_quic_create_short_header()
+ Client-related errors (i.e. parsing) are done at INFO level
+ c->log->action is updated through the process of receiving, parsing.
handling packet/payload and generating frames/output.
For ngx_http_process_request() part to work, this required to set both
r->http_connection->ssl and c->ssl on a QUIC stream. To avoid damaging
global SSL object, ngx_ssl_shutdown() is managed to ignore QUIC streams.
+ ngx_quic_init_ssl_methods() is no longer there, we setup methods on SSL
connection directly.
+ the handshake_handler is actually a generic quic input handler
+ updated c->log->action and debug to reflect changes and be more informative
+ c->quic is always set in ngx_quic_input()
+ the quic connection state is set by the results of SSL_do_handshake();
note:
+ parameters are available in SSL connection since they are obtained by ssl
stack
quote:
During connection establishment, both endpoints make authenticated
declarations of their transport parameters. These declarations are
made unilaterally by each endpoint.
and really, we send our parameters before we read client's.
no handling of incoming parameters is made by this patch.
- integer parameters can be configured using the following directives:
quic_max_idle_timeout
quic_max_ack_delay
quic_max_packet_size
quic_initial_max_data
quic_initial_max_stream_data_bidi_local
quic_initial_max_stream_data_bidi_remote
quic_initial_max_stream_data_uni
quic_initial_max_streams_bidi
quic_initial_max_streams_uni
quic_ack_delay_exponent
quic_active_migration
quic_active_connection_id_limit
- only following parameters are actually sent:
active_connection_id_limit
initial_max_streams_uni
initial_max_streams_bidi
initial_max_stream_data_bidi_local
initial_max_stream_data_bidi_remote
initial_max_stream_data_uni
(other parameters are to be added into ngx_quic_create_transport_params()
function as needed, should be easy now)
- draft 24 and draft 27 are now supported
(at compile-time using quic_version macro)
The ngx_quic_parse_frame() functions now has new 'pkt' argument: the packet
header of a currently processed frame. This allows to log errors/debug
closer to reasons and perform additional checks regarding possible frame
types. The handler only performs processing of good frames.
A number of functions like read_uint32(), parse_int[_multi] probably should
be implemented as a macro, but currently it is better to have them as
functions for simpler debugging.
Cleanup in ngx_event_quic.c:
+ reorderded functions, structures
+ added missing prototypes
+ added separate handlers for each frame type
+ numerous indentation/comments/TODO fixes
+ removed non-implemented qc->state and corresponding enum;
this requires deep thinking, stub was unused.
+ streams inside quic connection are now in own structure
All code dealing with serializing/deserializing
is moved int srv/event/ngx_event_quic_transport.c/h file.
All macros for dealing with data are internal to source file.
The header file exposes frame types and error codes.
The exported functions are currently packet header parsers and writers
and frames parser/writer.
The ngx_quic_header_t structure is updated with 'log' member. This avoids
passing extra argument to parsing functions that need to report errors.
+ support for more than one initial packet
+ workaround for trailing zeroes in packet
+ ignore application data packet if no keys yet (issue in draft 27/ff nightly)
+ fixed PING frame parser
+ STREAM frames need to be acknowledged
The following HTTP configuration is used for firefox (v74):
http {
ssl_certificate_key localhost.key;
ssl_certificate localhost.crt;
ssl_protocols TLSv1.2 TLSv1.3;
server {
listen 127.0.0.1:10368 reuseport http3;
ssl_quic on;
server_name localhost;
location / {
return 200 "This-is-QUICK\n";
}
}
server {
listen 127.0.0.1:5555 ssl; # point the browser here
server_name localhost;
location / {
add_header Alt-Svc 'h3-24=":10368";ma=100';
return 200 "ALT-SVC";
}
}
}
New files:
src/event/ngx_event_quic_protection.h
src/event/ngx_event_quic_protection.c
The protection.h header provides interface to the crypto part of the QUIC:
2 functions to initialize corresponding secrets:
ngx_quic_set_initial_secret()
ngx_quic_set_encryption_secret()
and 2 functions to deal with packet processing:
ngx_quic_encrypt()
ngx_quic_decrypt()
Also, structures representing secrets are defined there.
All functions require SSL connection and a pool, only crypto operations
inside, no access to nginx connections or events.
Currently pool->log is used for the logging (instead of original c->log).
- events handling moved into src/event/ngx_event_quic.c
- http invokes once ngx_quic_run() and passes stream callback
(diff to original http_request.c is now minimal)
- streams are stored in rbtree using ID as a key
- when a new stream is registered, appropriate callback is called
- ngx_quic_stream_t type represents STREAM and stored in c->qs
- now NEW_CONNECTION_ID frames can be received and parsed
The packet structure is created in ngx_quic_input() and passed
to all handlers (initial, handshake and application data).
The UDP datagram buffer is saved as pkt->raw;
The QUIC packet is stored as pkt->data and pkt->len (instead of pkt->buf)
(pkt->len is adjusted after parsing headers to actual length)
The pkt->pos is removed, pkt->raw->pos is used instead.
- added basic parsing of ACK, PING and PADDING frames on input
- added preliminary parsing of SHORT headers
The ngx_quic_output() is now called after processing of each input packet.
Frames are added into output queue according to their level: inital packets
go ahead of handshake and application data, so they can be merged properly.
The payload handler is called from both new, handshake and applicataion data
handlers (latter is a stub).
As was objserved with ngtcp2 client, Finished CRYPTO frame within Handshake
packet may not be sent for some reason if there's nothing to append on 1-RTT.
This results in unnecessary retransmit. To avoid this edge case, a non-zero
active_connection_id_limit transport parameter is now used to append datagram
with NEW_CONNECTION_ID 1-RTT frames.
Now handshake generates frames, and they are queued in c->quic->frames.
The ngx_quic_output() is called from ngx_quic_flush_flight() or manually,
processes the queue and encrypts all frames according to required encryption
level.
ngx_quic_hexdump0(log, format, buffer, buffer_size);
- logs hexdump of buffer to specified error log
ngx_quic_hexdump0(c->log, "this is foo:", foo.data, foo.len);
ngx_quic_hexdump(log, format, buffer, buffer_size, ...)
- same as hexdump0, but more format/args possible:
ngx_quic_hexdump(c->log, "a=%d b=%d, foo is:", foo.data, foo.len, a, b);
When "aio" or "aio threads" is used while processing the response body of an
in-memory background subrequest, the subrequest could be finalized with an aio
operation still in progress. Upon aio completion either parent request is
woken or the old r->write_event_handler is called again. The latter may result
in request errors. In either case post_subrequest handler is never called with
the full response body, which is typically expected when using in-memory
subrequests.
Currently in nginx background subrequests are created by the upstream module
and the mirror module. The issue does not manifest itself with these
subrequests because they are header-only. But it can manifest itself with
third-party modules which create in-memory background subrequests.
We used to have default error_page overwrite for 495, 496, and 497, so
a configuration like
error_page 495 /error;
will result in error 400, much like without any error_page configured.
The 494 status code was introduced later (in 3848:de59ad6bf557, nginx 0.9.4),
and relevant changes to ngx_http_core_error_page() were missed, resulting
in inconsistent behaviour of "error_page 494" - with error_page configured
it results in 494 being returned instead of 400.
Reported by Frank Liu,
http://mailman.nginx.org/pipermail/nginx/2020-February/058957.html.
Introduced ngx_quic_input() and ngx_quic_output() as interface between
nginx and protocol. They are the only functions that are exported.
While there, added copyrights.
In "co64" atom chunk start offset is a 64-bit unsigned integer. When trimming
the "mdat" atom, chunk offsets are casted to off_t values which are typically
64-bit signed integers. A specially crafted mp4 file with huge chunk offsets
may lead to off_t overflow and result in negative trim boundaries.
The consequences of the overflow are:
- Incorrect Content-Length header value in the response.
- Negative left boundary of the response file buffer holding the trimmed "mdat".
This leads to pread()/sendfile() errors followed by closing the client
connection.
On rare systems where off_t is a 32-bit integer, this scenario is also feasible
with the "stco" atom.
The fix is to add checks which make sure data chunks referenced by each track
are within the mp4 file boundaries. Additionally a few more checks are added to
ensure mp4 file consistency and log errors.
Duplicate "Host" headers were allowed in nginx 0.7.0 (revision b9de93d804ea)
as a workaround for some broken Motorola phones which used to generate
requests with two "Host" headers[1]. It is believed that this workaround
is no longer relevant.
[1] http://mailman.nginx.org/pipermail/nginx-ru/2008-May/017845.html
The "identity" transfer coding has been removed in RFC 7230. It is
believed that it is not used in real life, and at the same time it
provides a potential attack vector.
We anyway do not support more than one transfer encoding, so accepting
requests with multiple Transfer-Encoding headers doesn't make sense.
Further, we do not handle multiple headers, and ignore anything but
the first header.
Reported by Filippo Valsorda.
A connection could get stuck without timers if a client has partially sent
the HEADERS frame such that it was split on the individual header boundary.
In this case, it cannot be processed without the rest of the HEADERS frame.
The fix is to call ngx_http_v2_state_headers_save() in this case. Normally,
it would be called from the ngx_http_v2_state_header_block() handler on the
next iteration, when there is not enough data to continue processing. This
isn't the case if recv_buffer became empty and there's no more data to read.
With the recent change to prevent frames flood in d4448892a294,
nginx will finalize the connection with NGX_HTTP_V2_INTERNAL_ERROR
whenever flood is detected, causing nginx aborting or stopping if
the debug_points directive is used in nginx config.
Previous change 1ce3f01a4355 incorrectly introduced processing of the
ngx_posted_next_events queue at the end of operation, effectively making
posted next events a nop, since at the end of an event loop iteration
the queue is always empty. Correct approach is to move events to the
ngx_posted_events queue at an iteration start, as it was done previously.
Further, in some cases the c->read event might be already in the
ngx_posted_events queue, and calling ngx_post_event() with the
ngx_posted_next_events queue won't do anything. To make sure the event
will be correctly placed into the ngx_posted_next_events queue
we now check if it is already posted.
Introduced in 9d2ad2fb4423 available bytes handling in SSL relied
on connection read handler being overwritten to set the ready flag
and the amount of available bytes. This approach is, however, does
not work properly when connection read handler is changed, for example,
when switching to a next pipelined request, and can result in unexpected
connection timeouts, see here:
http://mailman.nginx.org/pipermail/nginx-devel/2019-December/012825.html
Fix is to introduce ngx_event_process_posted_next() instead, which
will set ready and available regardless of how event handler is set.
When ngx_http_v2_close_stream_handler() is used to retry stream close
after queued frames are sent, client timeouts on the stream can be
logged multiple times and/or in addition to already happened errors.
To resolve this, separate ngx_http_v2_retry_close_stream_handler()
was introduced, which does not try to log timeouts.
If a stream is closed with queued frames, it is possible that no further
write events will occur on the stream, leading to the socket leak.
To fix this, the stream's fake connection read handler is set to
ngx_http_v2_close_stream_handler(), to make sure that finalizing the
connection with ngx_http_v2_finalize_connection() will be able to
close the stream regardless of the current number of queued frames.
Additionally, the stream's fake connection fc->error flag is explicitly
set, so ngx_http_v2_handle_stream() will post a write event when queued
frames are finally sent even if stream flow control window is exhausted.
These checks were missed when chunked support was introduced. And also
added an explicit error message to ngx_http_dav_copy_move_handler()
(it was missed for some reason, in contrast to DELETE and MKCOL handlers).
While empty replacements were caught at run-time, parsing code
of the "rewrite" directive expects that a minimum length of the
"replacement" argument is 1.
If a rewritten URI has the null character, only a part of URI was
copied to a memory buffer allocated for path. In some setups this
could be exploited to expose uninitialized memory via the Location
header.
The "alias" directive cannot be used in the same location where URI
was rewritten. This has been detected in the "rewrite ... break"
case, but not when the standalone "break" directive was used.
This change also fixes proxy_pass with URI component in a similar
case:
location /aaa/ {
rewrite ^ /xxx/yyy;
break;
proxy_pass http://localhost:8080/bbb/;
}
Previously, the "/bbb/yyy" would be sent to a backend instead of
"/xxx/yyy". And if location's prefix was longer than the rewritten
URI, a segmentation fault might occur.
Previously, connections returned from keepalive cache had c->data
pointing to the keepalive cache item. While this shouldn't be a problem
for correct code, as c->data is not expected to be used before it is set,
explicitly clearing it might help to avoid confusion.
Previously only an rbtree was associated with a limit_conn. To make it
possible to associate more data with a limit_conn, shared context is introduced
similar to limit_req. Also, shared pool pointer is kept in a way similar to
limit_req.
Now a new structure ngx_proxy_protocol_t holds these fields. This allows
to add more PROXY protocol fields in the future without modifying the
connection structure.
With MinGW-w64, building 64-bit nginx binary with GCC 8 and above
results in warning due to cast of GetProcAddress() result to ngx_wsapoll_pt,
which GCC thinks is incorrect. Added intermediate cast to "void *" to
silence the warning.
FormatMessage() seems to return many errors which essentially indicate that
the language in question is not available. At least the following were
observed in the wild and during testing: ERROR_MUI_FILE_NOT_FOUND (15100)
(ticket #1868), ERROR_RESOURCE_TYPE_NOT_FOUND (1813). While documentation
says it should be ERROR_RESOURCE_LANG_NOT_FOUND (1815), this doesn't seem
to be the case.
As such, checking error code was removed, and as long as FormatMessage()
returns an error, we now always try the default language.
Added code to track number of bytes available in the socket.
This makes it possible to avoid looping for a long time while
working with fast enough peer when data are added to the socket buffer
faster than we are able to read and process data.
When kernel does not provide number of bytes available, it is
retrieved using ioctl(FIONREAD) as long as a buffer is filled by
SSL_read().
It is assumed that number of bytes returned by SSL_read() is close
to the number of bytes read from the socket, as we do not use
SSL compression. But even if it is not true for some reason, this
is not important, as we post an additional reading event anyway.
Note that data can be buffered at SSL layer, and it is not possible
to simply stop reading at some point and wait till the event will
be reported by the kernel again. This can be only done when there
are no data in SSL buffers, and there is no good way to find out if
it's the case.
Instead of trying to figure out if SSL buffers are empty, this patch
introduces events posted for the next event loop iteration - such
events will be processed only on the next event loop iteration,
after going into the kernel and retrieving additional events. This
seems to be simple and reliable approach.
This makes it possible to avoid looping for a long time while working
with a fast enough peer when data are added to the socket buffer faster
than we are able to read and process them (ticket #1431). This is
basically what we already do on FreeBSD with kqueue, where information
about the number of bytes in the socket buffer is returned by
the kevent() call.
With other event methods rev->available is now set to -1 when the socket
is ready for reading. Later in ngx_recv() and ngx_recv_chain(), if
full buffer is received, real number of bytes in the socket buffer is
retrieved using ioctl(FIONREAD). Reading more than this number of bytes
ensures that even with edge-triggered event methods the event will be
triggered again, so it is safe to stop processing of the socket and
switch to other connections.
Using ioctl(FIONREAD) only after reading a full buffer is an optimization.
With this approach we only call ioctl(FIONREAD) when there are at least
two recv()/readv() calls.
As long as there are data to read in the socket, yet the amount of data
is less than total size of the buffers in the chain, this saves one
unneeded read() syscall. Before this change, reading only stopped if
ngx_ssl_recv() returned no data, that is, two read() syscalls in a row
returned EAGAIN.
In SSL connections, data can be buffered by the SSL layer, and it is
wrong to avoid doing c->recv_chain() if c->read->available is 0 and
c->read->pending_eof is set. And tests show that the optimization in
question indeed can result in incorrect detection of premature connection
close if upstream closes the connection without sending a close notify
alert at the same time. Fix is to disable c->read->available optimization
for SSL connections.
This could happen when graceful shutdown configured by worker_shutdown_timeout
times out and is then followed by another timeout such as proxy_read_timeout.
In this case, the HEADERS frame is added to the output queue, but attempt to
send it fails (due to c->error forcibly set during graceful shutdown timeout).
This triggers request finalization which attempts to close the stream. But the
stream cannot be closed because there is a frame in the output queue, and the
connection cannot be finalized. This leaves the connection open without any
timer events leading to alert.
The fix is to post write event when sending output queue fails on c->error.
That will finalize the connection.
With this patch, all traffic over an HTTP/2 connection is counted in
the h2c->total_bytes field, and payload traffic is counted in
the h2c->payload_bytes field. As long as total traffic is many times
larger than payload traffic, we consider this to be a flood.
In 8df664ebe037, we've switched to maximizing stream window instead
of sending RST_STREAM. Since then handling of RST_STREAM with NO_ERROR
was fixed at least in Chrome, hence we switch back to using RST_STREAM.
This allows more effective rejecting of large bodies, and also minimizes
non-payload traffic to be accounted in the next patch.
Previously, if a response to the PTR request was cached, and ngx_resolver_dup()
failed to allocate memory for the resulting name, then the original node was
freed but left in expire_queue. A subsequent address resolving would end up
in a use-after-free memory access of the node either in ngx_resolver_expire()
or ngx_resolver_process_ptr(), when accessing it through expire_queue.
The fix is to leave the resolver node intact.
Don't waste server resources by sending RST_STREAM frames. Instead,
reject WINDOW_UPDATE frames with invalid zero increment by closing
connection with PROTOCOL_ERROR.
Don't waste server resources by sending RST_STREAM frames. Instead,
reject HEADERS and PRIORITY frames with self-dependency by closing
connection with PROTOCOL_ERROR.
When ngx_http_discard_request_body() call was added to ngx_http_send_response(),
there were no return codes other than NGX_OK and NGX_HTTP_INTERNAL_SERVER_ERROR.
Now it can also return NGX_HTTP_BAD_REQUEST, but ngx_http_send_response() still
incorrectly transforms it to NGX_HTTP_INTERNAL_SERVER_ERROR.
The fix is to propagate ngx_http_discard_request_body() errors.
As defined in HTTP/1.1, body chunks have the following ABNF:
chunk = chunk-size [ chunk-ext ] CRLF chunk-data CRLF
where chunk-data is a sequence of chunk-size octets.
With this change, chunk-data that doesn't end up with CRLF at chunk-size
offset will be treated as invalid, such as in the example provided below:
4
SEE-THIS-AND-
4
THAT
0
Previously, if unbuffered request body reading wasn't finished before
the request was redirected to a different location using error_page
or X-Accel-Redirect, and the request body is read again, this could
lead to disastrous effects, such as a duplicate post_handler call or
"http request count is zero" alert followed by a segmentation fault.
This happened in the following configuration (ticket #1819):
location / {
proxy_request_buffering off;
proxy_pass http://bad;
proxy_intercept_errors on;
error_page 502 = /error;
}
location /error {
proxy_pass http://backend;
}
Fixed excessive memory growth and CPU usage if stream windows are
manipulated in a way that results in generating many small DATA frames.
Fix is to limit the number of simultaneously allocated DATA frames.
Fixed uncontrolled memory growth if peer sends a stream of
headers with a 0-length header name and 0-length header value.
Fix is to reject headers with zero name length.
When using SMTP with SSL and resolver, read events might be enabled
during address resolving, leading to duplicate ngx_mail_ssl_handshake_handler()
calls if something arrives from the client, and duplicate session
initialization - including starting another resolving. This can lead
to a segmentation fault if the session is closed after first resolving
finished. Fix is to block read events while resolving.
Reported by Robert Norris,
http://mailman.nginx.org/pipermail/nginx/2019-July/058204.html.
After ac5a741d39cf it is now possible that after zstream.avail_out
reaches 0 and we allocate additional buffer, there will be no more data
to put into this buffer, triggering "zero size buf" alert. Fix is to
reset b->temporary flag in this case.
Additionally, an optimization added to avoid allocating additional buffer
in this case, by checking if last deflate() call returned Z_STREAM_END.
Note that checking for Z_STREAM_END by itself is not enough to fix alerts,
as deflate() can return Z_STREAM_END without producing any output if the
buffer is smaller than gzip trailer.
Reported by Witold Filipczyk,
http://mailman.nginx.org/pipermail/nginx-devel/2019-July/012469.html.
Due to shortcomings of the ccv->zero flag implementation in complex value
interface, length of the resulting string from ngx_http_complex_value()
might either not include terminating null character or include it,
so the only safe way to work with the result is to use it as a
null-terminated string.
Reported by Patrick Wollgast.
When "-" follows a parameter of maximum length, a single byte buffer
overflow happens, since the error branch does not check parameter length.
Fix is to avoid saving "-" to the parameter key, and instead use an error
message with "-" explicitly written. The message is mostly identical to
one used in similar cases in the preequal state.
Reported by Patrick Wollgast.
With level-triggered event methods it is important to specify
the NGX_CLOSE_EVENT flag to ngx_handle_read_event(), otherwise
the event won't be removed, resulting in CPU hog.
Reported by Patrick Wollgast.
To save memory hash code uses u_short to store resulting bucket sizes,
so maximum bucket size is limited to 65536 minus ngx_cacheline_size (larger
values will be aligned to 65536 which will overflow u_short). However,
there were no checks to enforce this, and using larger bucket sizes
resulted in overflows and segmentation faults.
Appropriate safety checks to enforce this added to ngx_hash_init().
When nginx is used with zlib patched with [1], which provides
integration with the future IBM Z hardware deflate acceleration, it ends
up computing CRC32 twice: one time in hardware, which always does this,
and one time in software by explicitly calling crc32().
crc32() calls were added in changesets 133:b27548f540ad ("nginx-0.0.1-
2003-09-24-23:51:12 import") and 134:d57c6835225c ("nginx-0.0.1-
2003-09-26-09:45:21 import") as part of gzip wrapping feature - back
then zlib did not support it.
However, since then gzip wrapping was implemented in zlib v1.2.0.4,
and it's already being used by nginx for log compression.
This patch replaces hand-written gzip wrapping with the one provided by
zlib. It simplifies the code, and makes it avoid computing CRC32 twice
when using hardware acceleration.
[1] https://github.com/madler/zlib/pull/410
Similarly to the change in 5491:74bfa803a5aa (1.5.9), we should accept
properly escaped URIs and unescape them as needed, else it is not possible
to handle URIs with question marks.
As we now have ctx->header_sent flag, it is further used to prevent
duplicate $r->send_http_header() calls, prevent output before sending
header, and $r->internal_redirect() after sending header.
Further, $r->send_http_header() protected from calls after
$r->internal_redirect().
Returning NGX_HTTP_INTERNAL_SERVER_ERROR if a perl code died after
sending header will lead to a "header already sent" alert. To avoid
it, we now check if header was already sent, and return NGX_ERROR
instead if it was.
Previously, redirects scheduled with $r->internal_redirect() were followed
even if the code then died. Now these are ignored and nginx will return
an error instead.
Variable handlers are not expected to send anything to the client, cannot
sleep or read body, and are not expected to modify the request. Added
appropriate protection to prevent accidental foot shooting.
Duplicate $r->sleep() and/or $r->has_request_body() calls result
in undefined behaviour (in practice, connection leaks were observed).
To prevent this, croak() added in appropriate places.
Previously, allocation errors in nginx.xs were more or less ignored,
potentially resulting in incorrect code execution in specific low-memory
conditions. This is changed to use ctx->error bit and croak(), similarly
to how output errors are now handled.
Note that this is mostly a cosmetic change, as Perl itself exits on memory
allocation errors, and hence nginx with Perl is hardly usable in low-memory
conditions.
When an error happens, the ctx->error bit is now set, and croak()
is called to terminate further processing. The ctx->error bit is
checked in ngx_http_perl_call_handler() to cancel further processing,
and is also checked in various output functions - to make sure these won't
be called if croak() was handled by an eval{} in perl code.
In particular, this ensures that output chain won't be called after
errors, as filters might not expect this to happen. This fixes some
segmentation faults under low memory conditions. Also this stops
request processing after filter finalization or request body reading
errors.
For cases where an HTTP error status can be additionally returned (for
example, 416 (Requested Range Not Satisfiable) from the range filter),
the ctx->status field is also added.
This ensures that correct ctx is always available, including after
filter finalization. In particular, this fixes a segmentation fault
with the following configuration:
location / {
image_filter test;
perl 'sub {
my $r = shift;
$r->send_http_header();
$r->print("foo\n");
$r->print("bar\n");
}';
}
This also seems to be the only way to correctly handle filter finalization
in various complex cases, for example, when embedded perl is used both
in the original handler and in an error page called after filter
finalization.
The NGX_DONE test in ngx_http_perl_handle_request() was introduced
in 1702:86bb52e28ce0, which also modified ngx_http_perl_call_handler()
to return NGX_DONE with c->destroyed. The latter part was then
removed in 3050:f54b02dbb12b, so NGX_DONE test is no longer needed.
Embedded perl does not set any request fields needed for conditional
requests processing. Further, filter finalization in the not_modified
filter can cause segmentation faults due to cleared ctx as in
ticket #1786.
Before 5fb1e57c758a (1.7.3) the not_modified filter was implicitly disabled
for perl responses, as r->headers_out.last_modified_time was -1. This
change restores this behaviour by using the explicit r->disable_not_modified
flag.
Note that this patch doesn't try to address perl module robustness against
filter finalization and other errors returned from filter chains. It should
be eventually reworked to handle errors instead of ignoring them.
A new directive limit_req_dry_run allows enabling the dry run mode. In this
mode requests are neither rejected nor delayed, but reject/delay status is
logged as usual.
In case of filter finalization, essential request fields like r->uri,
r->args etc could be changed, which affected the cache update subrequest.
Also, after filter finalization r->cache could be set to NULL, leading to
null pointer dereference in ngx_http_upstream_cache_background_update().
The fix is to create background cache update subrequest before sending the
cached response.
Since initial introduction in 1aeaae6e9446 (1.11.10) background cache update
subrequest was created after sending the cached response because otherwise it
blocked the parent request output. In 9552758a786e (1.13.1) background
subrequests were introduced to eliminate the delay before sending the final
part of the cached response. This also made it possible to create the
background cache update subrequest before sending the response.
Note that creating the subrequest earlier does not change the fact that in case
of filter finalization the background cache update subrequest will likely not
have enough time to successfully update the cache entry. Filter finalization
leads to the main request termination as soon the current iteration of request
processing is complete.
Previously, a variant not present in shared memory and stored on disk using a
secondary key was read using c->body_start from a variant stored with a main
key. This could result in critical errors "cache file .. has too long header".
Previously the stale-if-error extension of the Cache-Control upstream header
triggered the return of a stale response for all error conditions that can be
specified in the proxy_cache_use_stale directive. The list of these errors
includes both network/timeout/format errors, as well as some HTTP codes like
503, 504, 403, 429 etc. The latter prevented a cache entry from being updated
by a response with any of these HTTP codes during the stale-if-error period.
Now stale-if-error only works for network/timeout/format errors and ignores
the upstream HTTP code. The return of a stale response for certain HTTP codes
is still possible using the proxy_cache_use_stale directive.
This change also applies to the stale-while-revalidate extension of the
Cache-Control header, which triggers stale-if-error if it is missing.
Reported at
http://mailman.nginx.org/pipermail/nginx/2020-July/059723.html.
In ngx_http_range_singlepart_body() special buffers where passed
unmodified, including ones after the end of the range. As such,
if the last buffer of a response was sent separately as a special
buffer, two buffers with b->last_buf set were present in the response.
In particular, this might result in a duplicate final chunk when using
chunked transfer encoding (normally range filter and chunked transfer
encoding are not used together, but this may happen if there are trailers
in the response). This also likely to cause problems in HTTP/2.
Fix is to skip all special buffers after we've sent the last part of
the range requested. These special buffers are not meaningful anyway,
since we set b->last_buf in the buffer with the last part of the range,
and everything is expected to be flushed due to it.
Additionally, ngx_http_next_body_filter() is now called even
if no buffers are to be passed to it. This ensures that various
write events are properly propagated through the filter chain. In
particular, this fixes test failures observed with the above change
and aio enabled.
Filters are not allowed to change incoming chain links, and should allocate
their own links if any modifications are needed. Nevertheless
ngx_http_range_singlepart_body() modified incoming chain links in some
cases, notably at the end of the requested range.
No problems caused by this are currently known, mostly because of
limited number of possible modifications and the position of the range
body filter in the filter chain. Though this behaviour is clearly incorrect
and tests demonstrate that it can at least cause some proxy buffers being
lost when using proxy_force_ranges, leading to less effective handling
of responses.
Fix is to always allocate new chain links in ngx_http_range_singlepart_body().
Links are explicitly freed to ensure constant memory usage with long-lived
requests.
Postpone filter is an essential part of subrequest functionality. In absence
of it a subrequest response body is sent to the client out of order with
respect to the main request header and body, as well as other subrequests.
For in-memory subrequests the response is also sent to the client instead of
being stored in memory.
Currently the postpone filter is automatically enabled if one of the following
standard modules which are known to create subrequests is enabled: ssi, slice,
addition. However a third-party module that creates subrequests can still be
built without the postpone filter or be dynamically loaded in nginx built
without it.
If a complex value is expected to be of type size_t, and the compiled
value is constant, the constant size_t value is remembered at compile
time.
The value is accessed through ngx_http_complex_value_size() which
either returns the remembered constant or evaluates the expression
and parses it as size_t.
Previously, ngx_utf8_decode() was called from ngx_utf8_length() with
incorrect length, potentially resulting in out-of-bounds read when
handling invalid UTF-8 strings.
In practice out-of-bounds reads are not possible though, as autoindex, the
only user of ngx_utf8_length(), provides null-terminated strings, and
ngx_utf8_decode() anyway returns an errors when it sees a null in the
middle of an UTF-8 sequence.
Reported by Yunbin Liu.
If OCSP stapling was enabled with dynamic certificate loading, with some
OpenSSL versions (1.0.2o and older, 1.1.0h and older; fixed in 1.0.2p,
1.1.0i, 1.1.1) a segmentation fault might happen.
The reason is that during an abbreviated handshake the certificate
callback is not called, but the certificate status callback was called
(https://github.com/openssl/openssl/issues/1662), leading to NULL being
returned from SSL_get_certificate().
Fix is to explicitly check SSL_get_certificate() result.
With CFLAGS set as in 7da71a7b141a, OpenSSL compilation drops various
non-important compiler options. To avoid this, a define is added
instead - OpenSSL is smart enough to recognize -D... in Configure
arguments.
OpenSSL 1.1.0 and above uses BCrypt if available (Windows 7 or higher).
This results in an unusable binary on older Windows versions, when building
with newer Windows SDK (such as 7.0A). Using CFLAGS to define _WIN32_WINNT
allows to set a desired ABI and make sure the binary works with Windows XP.
To not mix with other potential CFLAGS uses, it is set in GNUmakefile.
When building OpenSSL 1.1.1b, as used for win32 builds, with tests
it takes about twice as long and near ~1GB of additional disk space.
Using "no-tests" OpenSSL configuration option allows to skip them.
Since such an option is supported since OpenSSL 1.1.1 only, it is
residing here and not in configure.
If X509_get_issuer_name() or X509_get_subject_name() returned NULL,
this could lead to a certificate reference leak. It cannot happen
in practice though, since each function returns an internal pointer
to a mandatory subfield of the certificate successfully decoded by
d2i_X509() during certificate message processing (closes#1751).
Previously the ngx_inet_resolve_host() function sorted addresses in a way that
IPv4 addresses came before IPv6 addresses. This was implemented in eaf95350d75c
(1.3.10) along with the introduction of getaddrinfo() which could resolve host
names to IPv6 addresses. Since the "listen" directive only used the first
address, sorting allowed to preserve "listen" compatibility with the previous
behavior and with the behavior of nginx built without IPv6 support. Now
"listen" uses all resolved addresses which makes sorting pointless.
Previously only one address was used by the listen directive handler even if
host name resolved to multiple addresses. Now a separate listening socket is
created for each address.
This makes it possible to provide certificates directly via variables
in ssl_certificate / ssl_certificate_key directives, without using
intermediate files.
It was accidentally introduced in 77436d9951a1 (1.15.9). In MSVC 2015
and more recent MSVC versions it triggers warning C4456 (declaration of
'pkey' hides previous local declaration). Previously, all such warnings
were resolved in 2a621245f4cf.
Reported by Steve Stevenson.
Server name callback is always called by OpenSSL, even
if server_name extension is not present in ClientHello. As such,
checking c->ssl->handshaked before the SSL_get_servername() result
should help to more effectively prevent renegotiation in
OpenSSL 1.1.0 - 1.1.0g, where neither SSL3_FLAGS_NO_RENEGOTIATE_CIPHERS
nor SSL_OP_NO_RENEGOTIATION is available.
The SSL_OP_NO_CLIENT_RENEGOTIATION option was introduced in LibreSSL 2.5.1.
Unlike OpenSSL's SSL_OP_NO_RENEGOTIATION, it only disables client-initiated
renegotiation, and hence can be safely used on all SSL contexts.
If ngx_pool_cleanup_add() fails, we have to clean just created SSL context
manually, thus appropriate call added.
Additionally, ngx_pool_cleanup_add() moved closer to ngx_ssl_create() in
the ngx_http_ssl_module, to make sure there are no leaks due to intermediate
code.
Notably this affects various allocation errors, and should generally
improve things if an allocation error actually happens during a callback.
Depending on the OpenSSL version, returning an error can result in
either SSL_R_CALLBACK_FAILED or SSL_R_CLIENTHELLO_TLSEXT error from
SSL_do_handshake(), so both errors were switched to the "info" level.
OpenSSL 1.1.1 does not save server name to the session if server name
callback returns anything but SSL_TLSEXT_ERR_OK, thus breaking
the $ssl_server_name variable in resumed sessions.
Since $ssl_server_name can be used even if we've selected the default
server and there are no other servers, it looks like the only viable
solution is to always return SSL_TLSEXT_ERR_OK regardless of the actual
result.
To fix things in the stream module as well, added a dummy server name
callback which always returns SSL_TLSEXT_ERR_OK.
A virtual server may have no SSL context if it does not have certificates
defined, so we have to use config of the ngx_http_ssl_module from the
SSL context in the certificate callback. To do so, it is now passed as
the argument of the callback.
The stream module doesn't really need any changes, but was modified as
well to match http code.
Dynamic certificates re-introduce problem with incorrect session
reuse (AKA "virtual host confusion", CVE-2014-3616), since there are
no server certificates to generate session id context from.
To prevent this, session id context is now generated from ssl_certificate
directives as specified in the configuration. This approach prevents
incorrect session reuse in most cases, while still allowing sharing
sessions across multiple machines with ssl_session_ticket_key set as
long as configurations are identical.
Passwords have to be copied to the configuration pool to be used
at runtime. Also, to prevent blocking on stdin (with "daemon off;")
an empty password list is provided.
To make things simpler, password handling was modified to allow
an empty array (with 0 elements and elts set to NULL) as an equivalent
of an array with 1 empty password.
To evaluate variables, a request is created in the certificate callback,
and then freed. To do this without side effects on the stub_status
counters and connection state, an additional function was introduced,
ngx_http_alloc_request().
Only works with OpenSSL 1.0.2+, since there is no SSL_CTX_set_cert_cb()
in older versions.
This makes it possible to reuse certificate loading at runtime,
as introduced in the following patches.
Additionally, this improves error logging, so nginx will now log
human-friendly messages "cannot load certificate" instead of only
referring to sometimes cryptic names of OpenSSL functions.
The "(SSL:)" snippet currently appears in logs when nginx code uses
ngx_ssl_error() to log an error, but OpenSSL's error queue is empty.
This can happen either because the error wasn't in fact from OpenSSL,
or because OpenSSL did not indicate the error in the error queue
for some reason.
In particular, currently "(SSL:)" can be seen in errors at least in
the following cases:
- When SSL_write() fails due to a syscall error,
"[info] ... SSL_write() failed (SSL:) (32: Broken pipe)...".
- When loading a certificate with no data in it,
"[emerg] PEM_read_bio_X509_AUX(...) failed (SSL:)".
This can easily happen due to an additional empty line before
the end line, so all lines of the certificate are interpreted
as header lines.
- When trying to configure an unknown curve,
"[emerg] SSL_CTX_set1_curves_list("foo") failed (SSL:)".
Likely there are other cases as well.
With this change, "(SSL:...)" will be only added to the error message
if there is something in the error queue. This is expected to make
logs more readable in the above cases. Additionally, with this change
it is now possible to use ngx_ssl_error() to log errors when some
of the possible errors are not from OpenSSL and not expected to have
anything in the error queue.
Checking multiple errors at once is a bad practice, as in general
it is not guaranteed that an object can be used after the error.
In this particular case, checking errors after multiple allocations
can result in excessive errors being logged when there is no memory
available.
On Windows, connect() errors are only reported via exceptfds descriptor set
from select(). Previously exceptfds was set to NULL, and connect() errors
were not detected at all, so connects to closed ports were waiting till
a timeout occurred.
Since ongoing connect() means that there will be a write event active,
except descriptor set is copied from the write one. While it is possible
to construct except descriptor set as a concatenation of both read and write
descriptor sets, this looks unneeded.
With this change, connect() errors are properly detected now when using
select(). Note well that it is not possible to detect connect() errors with
WSAPoll() (see https://daniel.haxx.se/blog/2012/10/10/wsapoll-is-broken/).
WSAPoll() is only available with Windows Vista and newer (and only
available during compilation if _WIN32_WINNT >= 0x0600). To make
sure the code works with Windows XP, we do not redefine _WIN32_WINNT,
but instead load WSAPoll() dynamically if it is not available during
compilation.
Also, sockets are not guaranteed to be small integers on Windows.
So an index array is used instead of NGX_USE_FD_EVENT to map
events to connections.
Previously, select was compiled in by default, but the NGX_HAVE_SELECT
macro was not set, resulting in iocp being used by default unless
the "--with-select_module" configure option was explicitly specified.
Since the iocp module is not finished and does not work properly, this
effectively meant that the "--with-select_module" option was mandatory.
With the change NGX_HAVE_SELECT is properly set, making "--with-select_module"
optional. Accordingly, it is removed from misc/GNUmakefile win32 target.
Previously, the code incorrectly assumed "ngx_event_t *" elements
instead of "struct pollfd".
This is mostly cosmetic change, as this code is never called now.
Previously, when using proxy_upload_rate and proxy_download_rate, the buffer
size for reading from a socket could be reduced as a result of rate limiting.
For connection-oriented protocols this behavior is normal since unread data will
normally be read at the next iteration. But for datagram-oriented protocols
this is not the case, and unread part of the datagram is lost.
Now buffer size is not limited for datagrams. Rate limiting still works in this
case by delaying the next reading event.
A shared connection does not own its file descriptor, which means that
ngx_handle_read_event/ngx_handle_write_event calls should do nothing for it.
Currently the c->shared flag is checked in several places in the stream proxy
module prior to calling these functions. However it was not done everywhere.
Missing checks could lead to calling
ngx_handle_read_event/ngx_handle_write_event on shared connections.
The problem manifested itself when using proxy_upload_rate and resulted in
either duplicate file descriptor error (e.g. with epoll) or incorrect further
udp packet processing (e.g. with kqueue).
The fix is to set and reset the event active flag in a way that prevents
ngx_handle_read_event/ngx_handle_write_event from scheduling socket events.
Previous interface of ngx_open_dir() assumed that passed directory name
has a room for NGX_DIR_MASK at the end (NGX_DIR_MASK_LEN bytes). While all
direct users of ngx_dir_open() followed this interface, this also implied
similar requirements for indirect uses - in particular, via ngx_walk_tree().
Currently none of ngx_walk_tree() uses provides appropriate space, and
fixing this does not look like a right way to go. Instead, ngx_dir_open()
interface was changed to not require any additional space and use
appropriate allocations instead.
If SSL_write_early_data() returned SSL_ERROR_WANT_WRITE, stop further reading
using a newly introduced c->ssl->write_blocked flag, as otherwise this would
result in SSL error "ssl3_write_bytes:bad length". Eventually, normal reading
will be restored by read event posted from successful SSL_write_early_data().
While here, place "SSL_write_early_data: want write" debug on the path.
Previously, if an SRV record was successfully resolved, but all of its A
records failed to resolve, NXDOMAIN was returned to the caller, which is
considered a successful resolve rather than an error. This could result in
losing the result of a previous successful resolve by the caller.
Now NXDOMAIN is only returned if at least one A resolve completed with this
code. Otherwise the error state of the first A resolve is returned.
Previously, unnamed regex captures matched in the parent request, were not
available in a cloned subrequest. Now 3 fields related to unnamed captures
are copied to a cloned subrequest: r->ncaptures, r->captures and
r->captures_data. Since r->captures cannot be changed by either request after
creating a clone, a new flag r->realloc_captures is introduced to force
reallocation of r->captures.
The issue was reported as a proxy_cache_background_update misbehavior in
http://mailman.nginx.org/pipermail/nginx/2018-December/057251.html.
In the past, there were several security issues which resulted in
worker process memory disclosure due to buffers with negative size.
It looks reasonable to check for such buffers in various places,
much like we already check for zero size buffers.
While here, removed "#if 1 / #endif" around zero size buffer checks.
It looks highly unlikely that we'll disable these checks anytime soon.
On 32-bit platforms mp4->buffer_pos might overflow when a large
enough (close to 4 gigabytes) atom is being skipped, resulting in
incorrect memory addesses being read further in the code. In most
cases this results in harmless errors being logged, though may also
result in a segmentation fault if hitting unmapped pages.
To address this, ngx_mp4_atom_next() now only increments mp4->buffer_pos
up to mp4->buffer_end. This ensures that overflow cannot happen.
Variables now do not depend on presence of the HTTP status code in response.
If the corresponding event occurred, variables contain time between request
creation and the event, and "-" otherwise.
Previously, intermediate value of the $upstream_response_time variable held
unix timestamp.
The directive allows to drop binding between a client and existing UDP stream
session after receiving a specified number of packets. First packet from the
same client address and port will start a new session. Old session continues
to exist and will terminate at moment defined by configuration: either after
receiving the expected number of responses, or after timeout, as specified by
the "proxy_responses" and/or "proxy_timeout" directives.
By default, proxy_requests is zero (disabled).
An attack that continuously switches HTTP/2 connection between
idle and active states can result in excessive CPU usage.
This is because when a connection switches to the idle state,
all of its memory pool caches are freed.
This change limits the maximum allowed number of idle state
switches to 10 * http2_max_requests (i.e., 10000 by default).
This limits possible CPU usage in one connection, and also
imposes a limit on the maximum lifetime of a connection.
Initially reported by Gal Goldshtein from F5 Networks.
Fixed uncontrolled memory growth in case peer is flooding us with
some frames (e.g., SETTINGS and PING) and doesn't read data. Fix
is to limit the number of allocated control frames.
Previously there was no validation for the size of a 64-bit atom
in an mp4 file. This could lead to a CPU hog when the size is 0,
or various other problems due to integer underflow when calculating
atom data size, including segmentation fault or worker process
memory disclosure.
Size of a shared memory zones must be at least two pages - one page
for slab allocator internal data, and another page for actual allocations.
Using 8192 instead is wrong, as there are systems with page sizes other
than 4096.
Note well that two pages is usually too low as well. In particular, cache
is likely to use two allocations of different sizes for global structures,
and at least four pages will be needed to properly allocate cache nodes.
Except in a few very special cases, with keys zone of just two pages nginx
won't be able to start. Other uses of shared memory impose a limit
of 8 pages, which provides some room for global allocations. This patch
doesn't try to address this though.
Inspired by ticket #1665.
With maximum version explicitly set, TLSv1.3 will not be unexpectedly
enabled if nginx compiled with OpenSSL 1.1.0 (without TLSv1.3 support)
will be run with OpenSSL 1.1.1 (with TLSv1.3 support).
In e3ba4026c02d (1.15.4) nginx own renegotiation checks were disabled
if SSL_OP_NO_RENEGOTIATION is available. But since SSL_OP_NO_RENEGOTIATION
is only set on a connection, not in an SSL context, SSL_clear_option()
removed it as long as a matching virtual server was found. This resulted
in a segmentation fault similar to the one fixed in a6902a941279 (1.9.8),
affecting nginx built with OpenSSL 1.1.0h or higher.
To fix this, SSL_OP_NO_RENEGOTIATION is now explicitly set in
ngx_http_ssl_servername() after adjusting options. Additionally, instead
of c->ssl->renegotiation we now check c->ssl->handshaked, which seems
to be a more correct flag to test, and will prevent the segmentation fault
from happening even if SSL_OP_NO_RENEGOTIATION is not working.
The "no suitable signature algorithm" errors are reported by OpenSSL 1.1.1
when using TLSv1.3 if there are no shared signature algorithms. In
particular, this can happen if the client limits available signature
algorithms to something we don't have a certificate for, or to an empty
list. For example, the following command:
openssl s_client -connect 127.0.0.1:8443 -sigalgs rsa_pkcs1_sha1
will always result in the "no suitable signature algorithm" error
as the "rsa_pkcs1_sha1" algorithm refers solely to signatures which
appear in certificates and not defined for use in TLS 1.3 handshake
messages.
The SSL_R_NO_COMMON_SIGNATURE_ALGORITHMS error is what BoringSSL returns
in the same situation.
The "no suitable key share" errors are reported by OpenSSL 1.1.1 when
using TLSv1.3 if there are no shared groups (that is, elliptic curves).
In particular, it is easy enough to trigger by using only a single
curve in ssl_ecdh_curve:
ssl_ecdh_curve secp384r1;
and using a different curve in the client:
openssl s_client -connect 127.0.0.1:443 -curves prime256v1
On the client side it is seen as "sslv3 alert handshake failure",
"SSL alert number 40":
0:error:14094410:SSL routines:ssl3_read_bytes:sslv3 alert handshake failure:ssl/record/rec_layer_s3.c:1528:SSL alert number 40
It can be also triggered with default ssl_ecdh_curve by using a curve
which is not in the default list (X25519, prime256v1, X448, secp521r1,
secp384r1):
openssl s_client -connect 127.0.0.1:8443 -curves brainpoolP512r1
Given that many clients hardcode prime256v1, these errors might become
a common problem with TLSv1.3 if ssl_ecdh_curve is redefined. Previously
this resulted in not using ECDH with such clients, but with TLSv1.3 it
is no longer possible and will result in a handshake failure.
The SSL_R_NO_SHARED_GROUP error is what BoringSSL returns in the same
situation.
Seen at:
https://serverfault.com/questions/932102/nginx-ssl-handshake-error-no-suitable-key-share
Previously, configurations with typo, for example
fastcgi_cache_valid 200301 302 5m;
successfully pass configuration test. Adding check for status
codes > 599, and such configurations are now properly rejected.
The bgcolor attribute overrides compatibility settings in browsers
and leads to undesirable behavior when the default font color is set
to white in the browser, since font-color is not also overridden.
Following 7319:dcab86115261, as long as SSL_OP_NO_RENEGOTIATION is
defined, it is OpenSSL library responsibility to prevent renegotiation,
so the checks are meaningless.
Additionally, with TLSv1.3 OpenSSL tends to report SSL_CB_HANDSHAKE_START
at various unexpected moments - notably, on KeyUpdate messages and
when sending tickets. This change prevents unexpected connection
close on KeyUpdate messages and when finishing handshake with upcoming
early data changes.
Trying to look into r->err_status in the "return" directive
makes it behave differently than real errors generated in other
parts of the code, and is an endless source of various problems.
This behaviour was introduced in 726:7b71936d5299 (0.4.4) with
the comment "fix: "return" always overrode "error_page" response code".
It is not clear if there were any real cases this was expected to fix,
but there are several cases which are broken due to this change, some
previously fixed (4147:7f64de1cc2c0).
In ticket #1634, the problem is that when r->err_status is set to
a non-special status code, it is not possible to return a response
by simply returning r->err_status. If this is the case, the only
option is to return script's e->status instead. An example
configuration:
location / {
error_page 404 =200 /err502;
return 404;
}
location = /err502 {
return 502;
}
After the change, such a configuration will properly return
standard 502 error, much like it happens when a 502 error is
generated by proxy_pass.
This also fixes the following configuration to properly close
connection as clearly requested by "return 444":
location / {
error_page 404 /close;
return 404;
}
location = /close {
return 444;
}
Previously, this required "error_page 404 = /close;" to work
as intended.
Socket leak was observed in the following configuration:
error_page 400 = /close;
location = /close {
return 444;
}
The problem is that "return 444" triggers termination of the request,
and due to error_page termination thinks that it needs to use a posted
request to clear stack. But at the early request processing where 400
errors are generated there are no ngx_http_run_posted_requests() calls,
so the request is only terminated after an external event.
Variants of the problem include "error_page 497" instead (ticket #695)
and various other errors generated during early request processing
(405, 414, 421, 494, 495, 496, 501, 505).
The same problem can be also triggered with "return 499" and "return 408"
as both codes trigger ngx_http_terminate_request(), much like "return 444".
To fix this, the patch adds ngx_http_run_posted_requests() calls to
ngx_http_process_request_line() and ngx_http_process_request_headers()
functions, and to ngx_http_v2_run_request() and ngx_http_v2_push_stream()
functions in HTTP/2.
Since the ngx_http_process_request() function is now only called via
other functions which call ngx_http_run_posted_requests(), the call
there is no longer needed and was removed.
It is possible that after SSL_read() will return SSL_ERROR_WANT_WRITE,
further calls will return SSL_ERROR_WANT_READ without reading any
application data. We have to call ngx_handle_write_event() and
switch back to normal write handling much like we do if there are some
application data, or the write there will be reported again and again.
Similarly, we have to switch back to normal read handling if there
is saved read handler and SSL_write() returns SSL_ERROR_WANT_WRITE.
While SSL_read() most likely to return SSL_ERROR_WANT_WRITE (and SSL_write()
accordingly SSL_ERROR_WANT_READ) during an SSL renegotiation, it is
not necessary mean that a renegotiation was started. In particular,
it can never happen during a renegotiation or can happen multiple times
during a renegotiation.
Because of the above, misleading "peer started SSL renegotiation" info
messages were replaced with "SSL_read: want write" and "SSL_write: want read"
debug ones.
Additionally, "SSL write handler" and "SSL read handler" are now logged
by the SSL write and read handlers, to make it easier to understand that
temporary SSL handlers are called instead of normal handlers.
The "do { c->recv() } while (c->read->ready)" form used in the
ngx_http_lingering_close_handler() is not really correct, as for
example with SSL c->read->ready may be still set when returning NGX_AGAIN
due to SSL_ERROR_WANT_WRITE. Therefore the above might be an infinite loop.
This doesn't really matter in lingering close, as we shutdown write side
of the socket anyway and also disable renegotiation (and even without shutdown
and with renegotiation it requires using very large certificate chain and
tuning socket buffers to trigger SSL_ERROR_WANT_WRITE). But for the sake of
correctness added an NGX_AGAIN check.
If sending request body was not completed (u->request_body_sent is not set),
the upstream keepalive module won't save such a connection. However, it
is theoretically possible (though highly unlikely) that sending of some
control frames can be blocked after the request body was sent. The
ctx->output_blocked flag introduced to disable keepalive in such cases.
The code is now able to parse additional control frames after
the response is received, and can send control frames as well.
This fixes keepalive problems as observed with grpc-c, which can
send window update and ping frames after the response, see
http://mailman.nginx.org/pipermail/nginx/2018-August/056620.html.
Previously the preread phase code ignored NGX_AGAIN value returned from
c->recv() and relied only on c->read->ready. But this flag is not reliable and
should only be checked for optimization purposes. For example, when using
SSL, c->read->ready may be set when no input is available. This can lead to
calling preread handler infinitely in a loop.
The problem does not manifest itself currently, because in case of
non-buffered reading, chain link created by u->create_request method
consists of a single element.
Found by PVS-Studio.
The directive configures maximum number of requests allowed on
a connection kept in the cache. Once a connection reaches the number
of requests configured, it is no longer saved to the cache.
The default is 100.
Much like keepalive_requests for client connections, this is mostly
a safeguard to make sure connections are closed periodically and the
memory allocated from the connection pool is freed.
The directive configures maximum time a connection can be kept in the
cache. By configuring a time which is smaller than the corresponding
timeout on the backend side one can avoid the race between closing
a connection by the backend and nginx trying to use the same connection
to send a request at the same time.
LibreSSL 2.8.0 "added const annotations to many existing APIs from OpenSSL,
making interoperability easier for downstream applications". This includes
the const change in the SSL_CTX_sess_set_get_cb() callback function (see
9dd43f4ef67e), which breaks compilation.
To fix this, added a condition on how we redefine OPENSSL_VERSION_NUMBER
when working with LibreSSL (see 382fc7069e3a). With LibreSSL 2.8.0,
we now set OPENSSL_VERSION_NUMBER to 0x1010000fL (OpenSSL 1.1.0), so the
appropriate conditions in the code will use "const" as it happens with
OpenSSL 1.1.0 and later versions.
There are clients which cannot handle HPACK's dynamic table size updates
as added in 12cadc4669a7 (1.13.6). Notably, old versions of OkHttp library
are known to fail on it (ticket #1397).
This change makes it possible to work with such clients by only sending
dynamic table size updates in response to SETTINGS_HEADER_TABLE_SIZE. As
a downside, clients which do not use SETTINGS_HEADER_TABLE_SIZE will
continue to maintain default 4k table.
Previously, a chunk of spaces larger than NGX_CONF_BUFFER (4096 bytes)
resulted in the "too long parameter" error during parsing such a
configuration. This was because the code only set start and start_line
on non-whitespace characters, and hence adjacent whitespace characters
were preserved when reading additional data from the configuration file.
Fix is to always move start and start_line if the last character was
a space.
Early data AKA 0-RTT mode is enabled as long as "ssl_early_data on" is
specified in the configuration (default is off).
The $ssl_early_data variable evaluates to "1" if the SSL handshake
isn't yet completed, and can be used to set the Early-Data header as
per draft-ietf-httpbis-replay-04.
BoringSSL currently requires SSL_CTX_set_max_proto_version(TLS1_3_VERSION)
to be able to enable TLS 1.3. This is because by default max protocol
version is set to TLS 1.2, and the SSL_OP_NO_* options are merely used
as a blacklist within the version range specified using the
SSL_CTX_set_min_proto_version() and SSL_CTX_set_max_proto_version()
functions.
With this change, we now call SSL_CTX_set_max_proto_version() with an
explicit maximum version set. This enables TLS 1.3 with BoringSSL.
As a side effect, this change also limits maximum protocol version to
the newest protocol we know about, TLS 1.3. This seems to be a good
change, as enabling unknown protocols might have unexpected results.
Additionally, we now explicitly call SSL_CTX_set_min_proto_version()
with 0. This is expected to help with Debian system-wide default
of MinProtocol set to TLSv1.2, see
http://mailman.nginx.org/pipermail/nginx-ru/2017-October/060411.html.
Note that there is no SSL_CTX_set_min_proto_version macro in BoringSSL,
so we call SSL_CTX_set_min_proto_version() and SSL_CTX_set_max_proto_version()
as long as the TLS1_3_VERSION macro is defined.
The behaviour is now in line with COPY of a directory with contents,
which preserves access masks on individual files, as well as the "cp"
command.
Requested by Roman Arutyunyan.
This fixes wrong permissions and file time after cross-device MOVE
in the DAV module (ticket #1577). Broken in 8101d9101ed8 (0.8.9) when
cross-device copying was introduced in ngx_ext_rename_file().
With this change, ngx_copy_file() always calls ngx_set_file_time(),
either with the time provided, or with the time from the original file.
This is considered acceptable given that copying the file is costly anyway,
and optimizing cases when we do not need to preserve time will require
interface changes.
Previously, ngx_open_file(NGX_FILE_CREATE_OR_OPEN) was used, resulting
in destination file being partially rewritten if exists. Notably,
this affected WebDAV COPY command (ticket #1576).
Previously, "%uA" was used, which corresponds to ngx_atomic_uint_t.
Size of ngx_atomic_uint_t can be easily different from uint64_t,
leading to undefined results.
In TLSv1.3, NewSessionTicket messages arrive after the handshake and
can come at any time. Therefore we use a callback to save the session
when we know about it. This approach works for < TLSv1.3 as well.
The callback function is set once per location on merge phase.
Since SSL_get_session() in BoringSSL returns an unresumable session for
TLSv1.3, peer save_session() methods have been updated as well to use a
session supplied within the callback. To preserve API, the session is
cached in c->ssl->session. It is preferably accessed in save_session()
methods by ngx_ssl_get_session() and ngx_ssl_get0_session() wrappers.
In OpenSSL 1.1.0 the SSL_CTRL_CLEAR_OPTIONS macro was removed, so
conditional compilation test on it results in SSL_clear_options()
and SSL_CTX_clear_options() not being used. Notably, this caused
"ssl_prefer_server_ciphers off" to not work in SNI-based virtual
servers if server preference was switched on in the default server.
It looks like the only possible fix is to test OPENSSL_VERSION_NUMBER
explicitly.
Starting with OpenSSL 1.1.0, SSL_R_UNSUPPORTED_PROTOCOL instead of
SSL_R_UNKNOWN_PROTOCOL is reported when a protocol is disabled via
an SSL_OP_NO_* option.
Additionally, SSL_R_VERSION_TOO_LOW is reported when using MinProtocol
or when seclevel checks (as set by @SECLEVEL=n in the cipher string)
rejects a protocol, and this is what happens with SSLv3 and @SECLEVEL=1,
which is the default.
There is also the SSL_R_VERSION_TOO_HIGH error code, but it looks like
it is not possible to trigger it.
There should be at least one worker connection for each listening socket,
plus an additional connection for channel between worker and master,
or starting worker processes will fail.
Previously, listenings sockets were not cloned if the worker_processes
directive was specified after "listen ... reuseport".
This also simplifies upcoming configuration check on the number
of worker connections, as it needs to know the number of listening
sockets before cloning.
The variable keeps the latest SSL protocol version supported by the client.
The variable has the same format as $ssl_protocol.
The version is read from the client_version field of ClientHello. If the
supported_versions extension is present in the ClientHello, then the version
is set to TLSv1.3.
Errors when sending UDP datagrams can happen, e.g., when local IP address
changes (see fa0e093b64d7), or an unavailable DNS server on the LAN can cause
send() to fail with EHOSTDOWN on BSD systems. If this happens during
initial query, retry sending immediately, to a different DNS server when
possible. If this is not enough, allow normal resend to happen by ignoring
the return code of the second ngx_resolver_send_query() call, much like we
do in ngx_resolver_resend().
The "http request" and "https proxy request" errors cannot happen
with HTTP due to pre-handshake checks in ngx_http_ssl_handshake(),
but can happen when SSL is used in stream and mail modules.
With gRPC it is possible that a request sending is blocked due to flow
control. Moreover, further sending might be only allowed once the
backend sees all the data we've already sent. With such a backend
it is required to clear the TCP_NOPUSH socket option to make sure all
the data we've sent are actually delivered to the backend.
As such, we now clear TCP_NOPUSH in ngx_http_upstream_send_request()
also on NGX_AGAIN if c->write->ready is set. This fixes a test (which
waits for all the 64k bytes as per initial window before allowing more
bytes) with sendfile enabled when the body was written to a file
in a different context.
Now tcp_nopush on peer connections is disabled if it is disabled on
the client connection, similar to how we handle c->sendfile. Previously,
tcp_nopush was always used on upstream connections, regardless of
the "tcp_nopush" directive.
We copy input buffers to our buffers, so various flags might be
unexpectedly set in buffers returned by ngx_chain_get_free_buf().
In particular, the b->in_file flag might be set when the body was
written to a file in a different context. With sendfile enabled this
in turn might result in protocol corruption if such a buffer was reused
for a control frame.
Make sure to clear buffers and set only fields we really need to be set.
The module implements random load-balancing algorithm with optional second
choice. In the latter case, the best of two servers is chosen, accounting
number of connections and server weight.
Example:
upstream u {
random [two [least_conn]];
server 127.0.0.1:8080;
server 127.0.0.1:8081;
server 127.0.0.1:8082;
server 127.0.0.1:8083;
}
Before 4a8c9139e579, ngx_resolver_create() didn't use configuration
pool, and allocations were done using malloc().
In 016352c19049, when resolver gained support of several servers,
new allocations were done from the pool.
With u->conf->preserve_output set the request body file might be used
after the response header is sent, so avoid cleaning it. (Normally
this is not a problem as u->conf->preserve_output is only set with
r->request_body_no_buffering, but the request body might be already
written to a file in a different context.)
Previously, only one client packet could be processed in a udp stream session
even though multiple response packets were supported. Now multiple packets
coming from the same client address and port are delivered to the same stream
session.
If it's required to maintain a single stream of data, nginx should be
configured in a way that all packets from a client are delivered to the same
worker. On Linux and DragonFly BSD the "reuseport" parameter should be
specified for this. Other systems do not currently provide appropriate
mechanisms. For these systems a single stream of udp packets is only
guaranteed in single-worker configurations.
The proxy_response directive now specifies how many packets are expected in
response to a single client packet.
Previously, ngx_event_recvmsg() got remote socket addresses after creating
the connection object. In preparation to handling multiple UDP packets in a
single session, this code was moved up.
On Linux recvmsg() syscall may return a zero-length client address when
receiving a datagram from an unbound unix datagram socket. It is usually
assumed that socket address has at least the sa_family member. Zero-length
socket address caused buffer over-read in functions which receive socket
address, for example ngx_sock_ntop(). Typically the over-read resulted in
unexpected socket family followed by session close. Now a fake socket address
is allocated instead of a zero-length client address.
Negative times can appear since workers only update time on an event
loop iteration start. If a worker was blocked for a long time during
an event loop iteration, it is possible that another worker already
updated the time stored in the node. As such, time since last update
of the node (ms) will be negative.
Previous code used ngx_abs(ms) in the calculations. That is, negative
times were effectively treated as positive ones. As a result, it was
not possible to maintain high request rates, where the same node can be
updated multiple times from during an event loop iteration.
In particular, this affected setups with many SSL handshakes, see
http://mailman.nginx.org/pipermail/nginx/2018-May/056291.html.
Fix is to only update the last update time stored in the node if the
new time is larger than previously stored one. If a future time is
stored in the node, we preserve this time as is.
To prevent breaking things on platforms without monotonic time available
if system time is updated backwards, a safety limit of 60 seconds is
used. If the time stored in the node is more than 60 seconds in the future,
we assume that the time was changed backwards and update lr->last
to the current time.
The bug in question was fixed in glibc 2.3.2 and is no longer expected
to manifest itself on real servers. On the other hand, the workaround
causes compilation problems on various systems. Previously, we've
already fixed the code to compile with musl libc (fd6fd02f6a4d), and
now it is broken on Fedora 28 where glibc's crypt library was replaced
by libxcrypt. So the workaround was removed.
FreeBSD returns EINVAL when getsockopt(TCP_FASTOPEN) is called on a unix
domain socket, resulting in "getsockopt(TCP_FASTOPEN) ... failed" messages
during binary upgrade when unix domain listen sockets are present in
the configuration. Added EINVAL to the list of ignored error codes.
While 325b3042edd6 fixed it on MINIX, it broke it on systems
that output the word "version" on several lines with "cc -v".
The fix is to only consider "clang version" or "LLVM version"
as clang version, but this time only using sed(1).
Previously, only unix domain sockets were reopened to tolerate cases when
local syslog server was restarted. It makes sense to treat other cases
(for example, local IP address changes) similarly.
Cast to intermediate "void *" to lose compiler knowledge about the original
type and pass the warning. This is not a real fix but rather a workaround.
Found by gcc8.
In mail and stream modules, no certificate provided is a fatal condition,
much like with the "ssl" and "starttls" directives.
In http, "listen ... ssl" can be used in a non-default server without
certificates as long as there is a certificate in the default one, so
missing certificate is only fatal for default servers.
In 51e1f047d15d, the "ssl" directive name was incorrectly hardcoded
in the error message shown when there are some SSL keys defined, but
not for all certificates. Right approach is to use the "mode" variable,
which can be either "ssl" or "starttls".
Previously, result of ngx_atoi() was assigned to an ngx_uint_t variable,
and errors reported by ngx_atoi() became positive, so the following check
in "status < 100" failed to catch them. This resulted in the configurations
like "proxy_cache_valid 2xx 30s" being accepted as correct, while they
in fact do nothing. Changing type to ngx_int_t fixes this, and such
configurations are now properly rejected.
Previously, ngx_http_upstream_process_header() might be called after
we've finished reading response headers and switched to a different read
event handler, leading to errors with gRPC proxying. Additionally,
the u->conf->read_timeout timer might be re-armed during reading response
headers (while this is expected to be a single timeout on reading
the whole response header).
Previously, ngx_http_upstream_test_next() used an outdated condition on
whether it will be possible to switch to a different server or not. It
did not take into account restrictions on non-idempotent requests, requests
with non-buffered request body, and the next upstream timeout.
For such requests, switching to the next upstream server was rejected
later in ngx_http_upstream_next(), resulting in nginx own error page
being returned instead of the original upstream response.
- use normal prefixes for types and macros
- removed some macros and types
- revised debug messages
- removed useless check of ngx_sock_ntop() returning 0
- removed special processing of AF_UNSPEC
The protocol used on inbound connection is auto-detected and corresponding
parser is used to extract passed addresses. TLV parameters are ignored.
The maximum supported size of PROXY protocol header is 107 bytes
(similar to version 1).
All cases are harmless and should not happen on valid values, though can
result in bad values being shown incorrectly in logs.
Found by Coverity (CID 1430311, 1430312, 1430313).
This was previously used, but was incorrectly removed in 83d54192e97b
while removing old threads remnants. Instead of using it conditionally
when threads are not used, we now set in unconditionally, as even with
thread pools enabled we never call OpenSSL functions in threads.
This fixes resulting binary when using --with-openssl with OpenSSL 1.1.0+
and without -lpthread linked (notably on FreeBSD without PCRE).
OpenSSL now uses pthread_atfork(), and this requires -lpthread on Linux
to compile. Introduced NGX_LIBPTHREAD to add it as appropriate, similar
to existing NGX_LIBDL.
The fields "uri", "location", and "url" from ngx_http_upstream_conf_t
moved to ngx_http_proxy_loc_conf_t and ngx_http_proxy_vars_t, reflect
this change in create_loc_conf comments.
The gRPC protocol makes a distinction between HEADERS frame with
the END_STREAM flag set, and a HEADERS frame followed by an empty
DATA frame with the END_STREAM flag. The latter is not permitted,
and results in errors not being propagated through nginx. Instead,
gRPC clients complain that "server closed the stream without sending
trailers" (seen in grpc-go) or "13: Received RST_STREAM with error
code 2" (seen in grpc-c).
To fix this, nginx now returns HEADERS with the END_STREAM flag if
the response length is known to be 0, and we are not expecting
any trailer headers to be added. And the response length is
explicitly set to 0 in the gRPC proxy if we see initial HEADERS frame
with the END_STREAM flag set.
According to the gRPC protocol specification, the "TE" header is used
to detect incompatible proxies, and at least grpc-c server rejects
requests without "TE: trailers".
To preserve the logic, we have to pass "TE: trailers" to the backend if
and only if the original request contains "trailers" in the "TE" header.
Note that no other TE values are allowed in HTTP/2, so we have to remove
anything else.
The module allows passing requests to upstream gRPC servers.
The module is built by default as long as HTTP/2 support is compiled in.
Example configuration:
grpc_pass 127.0.0.1:9000;
Alternatively, the "grpc://" scheme can be used:
grpc_pass grpc://127.0.0.1:9000;
Keepalive support is available via the upstream keepalive module. Note
that keepalive connections won't currently work with grpc-go as it fails
to handle SETTINGS_HEADER_TABLE_SIZE.
To use with SSL:
grpc_pass grpcs://127.0.0.1:9000;
SSL connections use ALPN "h2" when available. At least grpc-go works fine
without ALPN, so if ALPN is not available we just establish a connection
without it.
Tested with grpc-c++ and grpc-go.
The flag can be used to continue sending request body even after we've
got a response from the backend. In particular, this is needed for gRPC
proxying of bidirectional streaming RPCs, and also to send control frames
in other forms of RPCs.
The flag indicates whether last ngx_output_chain() returned NGX_AGAIN
or not. If the flag is set, we arm the u->conf->send_timeout timer.
The flag complements c->write->ready test, and allows to stop sending
the request body in an output filter due to protocol-specific flow
control.
Basic trailer headers support allows one to access response trailers
via the $upstream_trailer_* variables.
Additionally, the u->conf->pass_trailers flag was introduced. When the
flag is set, trailer headers from the upstream response are passed to
the client. Like normal headers, trailer headers will be hidden
if present in u->conf->hide_headers_hash.
When clock_gettime(CLOCK_MONOTONIC) (or faster variants, _FAST on FreeBSD,
and _COARSE on Linux) is available, we now use it for ngx_current_msec.
This should improve handling of timers if system time changes (ticket #189).
The r->out chain link could be left uninitialized in case of error.
A segfault could happen if the subrequest handler accessed it.
The issue was introduced in commit 20f139e9ffa8.
Previously, only the upstream response body could be accessed with the
NGX_HTTP_SUBREQUEST_IN_MEMORY feature. Now any response body from a subrequest
can be saved in a memory buffer. It is available as a single buffer in r->out
and the buffer size is configured by the subrequest_output_buffer_size
directive.
Upstream, proxy and fastcgi code used to handle the old-style feature is
removed.
On some platforms (for example, Linux with glibc 2.12-2.25) IPv4 transparent
proxying is available, but IPv6 transparent proxying is not. The entire feature
is enabled in this case and NGX_HAVE_TRANSPARENT_PROXY macro is set to 1.
Previously, an attempt to enable transparency for an IPv6 socket was silently
ignored in this case and was usually followed by a bind(2) EADDRNOTAVAIL error
(ticket #1487). Now the error is generated for unavailable IPv6 transparent
proxy.
If during configuration parsing of the geo directive the memory
allocation has failed, pool used to parse configuration inside
the block, and sometimes the temporary pool were not destroyed.
There is no need to calculate hashes of static strings at runtime. The
ngx_hash() macro can be used to do it during compilation instead, similarly
to how it is done in ngx_http_proxy_module.c for "Server" and "Date" headers.
In particular, if a stream object allocation failed, and a client sent
the PRIORITY frame for this stream, ngx_http_v2_set_dependency() could
dereference a null pointer while trying to re-parent a dependency node.
r->headers_in.host can be NULL in ngx_http_v2_push_resource().
This happens when a request is terminated with 400 before the :authority
or Host header is parsed, and either pushing is enabled on the server{}
level or error_page 400 redirects to a location with pushes configured.
Found by Coverity (CID 1429156).
Resources to be pushed are configured with the "http2_push" directive.
Also, preload links from the Link response headers, as described in
https://www.w3.org/TR/preload/#server-push-http-2, can be pushed, if
enabled with the "http2_push_preload" directive.
Only relative URIs with absolute paths can be pushed.
The number of concurrent pushes is normally limited by a client, but
cannot exceed a hard limit set by the "http2_max_concurrent_pushes"
directive.
Previously, when request body was not available or was previously read in
memory rather than a file, client received HTTP 500 error, but no explanation
was logged in error log. This could happen, for example, if request body was
read or discarded prior to error_page redirect, or if mirroring was enabled
along with dav.
This fixes segfault in configurations with multiple virtual servers sharing
the same port, where a non-default virtual server block misses certificate.
Following ad3f342f14ba046c (1.9.13), it is possible that a request where
header was already sent will be finalized with NGX_HTTP_BAD_GATEWAY,
triggering an attempt to return additional error response and the
"header already sent" alert as a result.
In particular, it is trivial to reproduce the problem with a HEAD request
and caching enabled. With caching enabled nginx will change HEAD to GET
and will set u->pipe->downstream_error to suppress sending the response
body to the client. When a backend-related error occurs (for example,
proxy_read_timeout expires), ngx_http_finalize_upstream_request() will
be called with NGX_HTTP_BAD_GATEWAY. After ad3f342f14ba046c this will
result in ngx_http_finalize_request(NGX_HTTP_BAD_GATEWAY).
Fix is to move u->pipe->downstream_error handling to a later point,
where all special response codes are changed to NGX_ERROR.
Reported by Jan Prachar,
http://mailman.nginx.org/pipermail/nginx-devel/2018-January/010737.html.
Specifically, it is now allowed to start with a variable expression with braces:
${name}. The opening curly bracket in such a token was previously considered
the start of a new block. Variables located anywhere else in a token worked
fine: foo${name}.
Previously, capset(2) was called with the 64-bit capabilities version
_LINUX_CAPABILITY_VERSION_3. With this version Linux kernel expected two
copies of struct __user_cap_data_struct, while only one was submitted. As a
result, random stack memory was accessed and random capabilities were requested
by the worker. This sometimes caused capset() errors. Now the 32-bit version
_LINUX_CAPABILITY_VERSION_1 is used instead. This is OK since CAP_NET_RAW is
a 32-bit capability (CAP_NET_RAW = 13).
Previously included file sys/capability.h mentioned in capset(2) man page,
belongs to the libcap-dev package, which may not be installed on some Linux
systems when compiling nginx. This prevented the capabilities feature from
being detected and compiled on that systems.
Now linux/capability.h system header is included instead. Since capset()
declaration is located in sys/capability.h, now capset() syscall is defined
explicitly in code using the SYS_capset constant, similarly to other
Linux-specific features in nginx.
The capability is retained automatically in unprivileged worker processes after
changing UID if transparent proxying is enabled at least once in nginx
configuration.
The feature is only available in Linux.
In 2c7b488a61fb, IP_BIND_ADDRESS_NO_PORT test was accidentally placed
between SO_BINDANY, IP_TRANSPARENT, and IP_BINDANY tests. Moved it after
these tests.
If the flag space_in_uri is set, the URI in HTTP upstream request is escaped to
convert space to %20. However this flag is not checked while creating the
default cache key. This leads to different cache keys for requests
'/foo bar' and '/foo%20bar', while the upstream requests are identical.
Additionally, the change fixes background cache updates when the client URI
contains unescaped space. Default cache key in a subrequest is always based on
escaped URI, while the main request may not escape it. As a result, background
cache update subrequest may update a different cache entry.
Inheriting this flag will make the cloned subrequest behave consistently with
the parent. Specifically, the upstream HTTP request and cache key created by
the proxy module may depend directly on unparsed_uri if valid_unparsed_uri flag
is set. Previously, the flag was zero for cloned requests, which could make
background update proxy a request different than its parent and cache the result
with a different key. For example, if client URI contained the escaped slash
character %2F, it was used as is by the proxy module in the main request, but
was unescaped in the subrequests.
Similar problems exist in the slice module.
Previously, the unparsed uri was explicitly allowed to be used only by the main
request. However the valid_unparsed_uri flag is nonzero only in the main
request, which makes the main request check pointless.
If the data to write is bigger than what the socket can send, and the
reminder is smaller than NGX_SSL_BUFSIZE, then SSL_write() fails with
SSL_ERROR_WANT_WRITE. The reminder of payload however is successfully
copied to the low-level buffer and all the output chain buffers are
flushed. This means that retry logic doesn't work because
ngx_http_upstream_process_non_buffered_request() checks only if there's
anything in the output chain buffers and ignores the fact that something
may be buffered in low-level parts of the stack.
Signed-off-by: Patryk Lesiewicz <patryk@google.com>
If a connection with the read delayed flag set was stored in the keepalive
cache, and after picking it from the cache a read timer was set on that
connection, this timer was considered a delay timer rather than a socket read
event timer as expected. The latter timeout is usually much longer than the
former, which caused a significant delay in request processing.
The issue manifested itself with proxy_limit_rate and upstream keepalive
enabled and exists since 973ee2276300 (1.7.7) when proxy_limit_rate was
introduced.
On some systems, it's possible that reaper of orphaned processes is
set to something other than "init" process. On such systems, the
changing binary procedure did not work.
The fix is to check if PPID has changed, instead of assuming it's
always 1 for orphaned processes.
As per POSIX, basic regular expressions have no alternations, and the
interpretation of the "\|" construct is undefined. At least on MINIX
and Solaris grep interprets "\|" as literal "|", and not as an alternation
as GNU grep does. Removed such constructs introduced in f1daa0356a1d.
This fixes clang detection on MINIX.
The ngx_http_upstream_process_upgraded() did not handle c->close request,
and upgraded connections do not use the write filter. As a result,
worker_shutdown_timeout did not affect upgraded connections (ticket #1419).
Fix is to handle c->close in the ngx_http_request_handler() function, thus
covering most of the possible cases in http handling.
Additionally, mail proxying did not handle neither c->close nor c->error,
and thus worker_shutdown_timeout did not work for mail connections. Fix is
to add c->close handling to ngx_mail_proxy_handler().
Also, added explicit handling of c->close to stream proxy,
ngx_stream_proxy_process_connection(). This improves worker_shutdown_timeout
handling in stream, it will no longer wait for some data being transferred
in a connection before closing it, and will also provide appropriate
logging at the "info" level.
A zlib variant from Intel as available from https://github.com/jtkukunas/zlib
uses 64K hash instead of scaling it from the specified memory level, and
also uses 16-byte padding in one of the window-sized memory buffers, and can
force window bits to 13 if compression level is set to 1 and appropriate
compile options are used. As a result, nginx complained with "gzip filter
failed to use preallocated memory" alerts.
This change improves deflate_state allocation detection by testing that
items is 1 (deflate_state is the only allocation where items is 1).
Additionally, on first failure to use preallocated memory we now assume
that we are working with the Intel's modified zlib, and switch to using
appropriate preallocations. If this does not help, we complain with the
usual alerts.
Previous version of this patch was published at
http://mailman.nginx.org/pipermail/nginx/2014-July/044568.html.
The zlib variant in question is used by default in ClearLinux from Intel,
see http://mailman.nginx.org/pipermail/nginx-ru/2017-October/060421.html,
http://mailman.nginx.org/pipermail/nginx-ru/2017-November/060544.html.
Previously, nginx failed to move buffer position when parsing an incomplete
record header, and due to this wasn't be able to continue parsing once
remaining bytes of the record header were received.
This can affect response header parsing, potentially generating spurious errors
like "upstream sent unexpected FastCGI request id high byte: 1 while reading
response header from upstream". While this is very unlikely, since usually
record headers are written in a single buffer, this still can happen in real
life, for example, if a record header will be split across two TCP packets
and the second packet will be delayed.
This does not affect non-buffered response body proxying, due to "buf->pos =
buf->last;" at the start of the ngx_http_fastcgi_non_buffered_filter()
function. Also this does not affect buffered response body proxying, as
each input buffer is only passed to the filter once.
This is what usually happens for zones no longer used in the new
configuration, but zones where size or tag were changed were freed
when creating new memory zones. If reconfiguration failed (for
example, due to a conflicting listening socket), this resulted in a
segmentation fault in the master process.
Reported by Zhihua Cao,
http://mailman.nginx.org/pipermail/nginx-devel/2017-October/010536.html.
In particular, if ngx_http_postpone_filter_add() fails in ngx_chain_add_copy(),
the output chain of the postponed request was left in an invalid state.
This header carries the definition of HMAC_Init_ex(). In OpenSSL this
header is included by <openssl/ssl.h>, but it's not so in BoringSSL.
It's probably a good idea to explicitly include this header anyway,
regardless of whether it's included by other headers or not.
Upgrading an upstream connection is usually followed by reading from the client
which a subrequest is not allowed to do. Moreover, accessing the header_in
request field while processing upgraded connection ends up with a null pointer
dereference since the header_in buffer is only created for the the main request.
If proxy_next_upstream includes http_503/http_504, and upstream
returns 503/504, $upstream_status converted this to 502 for any
values except the last one.
The NGX_DONE value returned from ngx_http_upstream_cache_send() indicates
that upstream was already finalized in ngx_http_upstream_process_headers().
It was treated as a generic error which resulted in duplicate finalization.
Handled NGX_HTTP_UPSTREAM_INVALID_HEADER from ngx_http_upstream_cache_send().
Previously, it could return within ngx_http_upstream_finalize_request(), and
since it's below NGX_HTTP_SPECIAL_RESPONSE, a client connection could stuck.
When parsing of headers in a cache file fails, already parsed headers
need to be cleared, and protocol state needs to be reinitialized. To do
so, u->request_sent is now set to ensure ngx_http_upstream_reinit() will
be called.
This change complements improvements in 46ddff109e72.
This slightly reduces cost of selecting a peer if all or almost all peers
failed, see ticket #1030. There should be no measureable difference with
other workloads.
While this may result in non-ideal distribution of requests if nginx
won't be able to select a server in a reasonable number of attempts,
this still looks better than severe performance degradation observed
if there is no limit and there are many points configured (ticket #1030).
This is also in line with what we do for other hash balancing methods.
Previously, unix sockets were treated as AF_INET ones, and this may
result in buffer overread on Linux, where unbound unix sockets have
2-byte addresses.
Note that it is not correct to use just sun_path as a binary representation
for unix sockets. This will result in an empty string for unbound unix
sockets, and thus behaviour of limit_req and limit_conn will change when
switching from $remote_addr to $binary_remote_addr. As such, normal text
representation is used.
Reported by Stephan Dollberg.
At least FreeBSD, macOS, NetBSD, and OpenBSD can return unix sockets
with non-null-terminated sun_path. Additionally, the address may become
non-null-terminated if it does not fit into the buffer provided and was
truncated (may happen on macOS, NetBSD, and Solaris, which allow unix socket
addresess larger than struct sockaddr_un). As such, ngx_sock_ntop() might
overread the sockaddr provided, as it used "%s" format and thus assumed
null-terminated string.
To fix this, the ngx_strnlen() function was introduced, and it is now used
to calculate correct length of sun_path.
Some OSes (notably macOS, NetBSD, and Solaris) allow unix socket addresses
larger than struct sockaddr_un. Moreover, some of them (macOS, Solaris)
return socklen of the socket address before it was truncated to fit the
buffer provided. As such, on these systems socklen must not be used without
additional check that it is within the buffer provided.
Appropriate checks added to ngx_event_accept() (after accept()),
ngx_event_recvmsg() (after recvmsg()), and ngx_set_inherited_sockets()
(after getsockname()).
We also obtain socket addresses via getsockname() in
ngx_connection_local_sockaddr(), but it does not need any checks as
it is only used for INET and INET6 sockets (as there can be no
wildcard unix sockets).
The sync flag of HTTP/2 request body buffer is used when the size of request
body is unknown or bigger than configured "client_body_buffer_size". In this
case the buffer points to body data inside the global receive buffer that is
used for reading all HTTP/2 connections in the worker process. Thus, when the
sync flag is set, the buffer must be flushed to a temporary file, otherwise
the request body data can be overwritten.
Previously, the sync buffer wasn't flushed to a temporary file if the whole
body was received in one DATA frame with the END_STREAM flag and wasn't
copied into the HTTP/2 body preread buffer. As a result, the request body
might be corrupted (ticket #1384).
Now, setting r->request_body_in_file_only enforces writing the sync buffer
to a temporary file in all cases.
When caching intercepted errors, previous behaviour was to use
proxy_cache_valid times specified, regardless of various cache control
headers present in the response. Fix is to check u->cacheable and
use u->cache->valid_sec as set by various cache control response headers,
similar to how we do this in the normal caching code path.
If cache file is truncated, it is possible that u->process_header()
will return NGX_AGAIN. Added appropriate handling of this case by
changing the error to NGX_HTTP_UPSTREAM_INVALID_HEADER.
Also, added appropriate logging of this and NGX_HTTP_UPSTREAM_INVALID_HEADER
cases at the "crit" level. Note that this will result in duplicate logging
in case of NGX_HTTP_UPSTREAM_INVALID_HEADER. While this is something better
to avoid, it is considered to be an overkill to implement cache-specific
error logging in u->process_header().
Additionally, u->buffer.start is now reset to be able to receive a new
response, and u->cache_status set to MISS to provide the value in the
$upstream_cache_status variable, much like it happens on other cache file
errors detected by ngx_http_file_cache_read(), instead of HIT, which is
believed to be misleading.
It is to be used as a bitmask with various bits set/reset when appropriate.
63b8b157b776 made a similar change to ngx_http_upstream_rr_peer_t.down and
ngx_stream_upstream_rr_peer_t.down.
Previously, "get indexed header" message was logged when in fact only
header name was obtained using an index, and "get indexed header name"
was logged when full header representation (name and value) was obtained
using an index. Fixed version logs "get indexed name" and "get indexed
header" respectively.
Previously, when the first UDP response packet was not received from the
proxied server within proxy_timeout, no error message was logged before
switching to the next upstream. Additionally, when one of succeeding response
packets was not received within the timeout, the timeout error had low severity
because it was logged as a client connection error as opposed to upstream
connection error.
Various buffers are allocated in an assumption that there would be
no more than 4 year digits. This might not be true on platforms
with 64-bit time_t, as 64-bit time_t is able to represent more than that.
Such dates with more than 4 year digits hardly make sense though, as
various date formats in use do not allow them anyway.
As such, all dates are now truncated by ngx_gmtime() to December 31, 9999.
This should have no effect on valid dates, though will prevent potential
buffer overflows on invalid ones.
In ngx_gmtime(), instead of casting to ngx_uint_t we now work with
time_t directly. This allows using dates after 2038 on 32-bit platforms
which use 64-bit time_t, notably NetBSD and OpenBSD.
As the code is not able to work with negative time_t values, argument
is now set to 0 for negative values. As a positive side effect, this
results in Epoch being used for such values instead of a date in distant
future.
This change lets NGINX talk to clients with SETTINGS_HEADER_TABLE_SIZE
smaller than the default 4KB. Previously, NGINX would ACK the SETTINGS
frame with a small dynamic table size, but it would never send dynamic
table size update, leading to a connection-level COMPRESSION_ERROR.
Also, it allows clients to release 4KB of memory per connection, since
NGINX doesn't use HPACK's dynamic table when encoding headers, however
clients had to maintain it, since NGINX never signaled that it doesn't
use it.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
When switching to a next upstream, some buffers could be stuck in the middle
of the filter chain. A condition existed that raised an error when this
happened. As it turned out, this condition prevented switching to a next
upstream if ssl preread was used with the TCP protocol (see the ticket).
In fact, the condition does not make sense for TCP, since after successful
connection to an upstream switching to another upstream never happens. As for
UDP, the issue with stuck buffers is unlikely to happen, but is still possible.
Specifically, if a filter delays sending data to upstream.
The condition can be relaxed to only check the "buffered" bitmask of the
upstream connection. The new condition is simpler and fixes the ticket issue
as well. Additionally, the upstream_out chain is now reset for UDP prior to
connecting to a new upstream to prevent repeating the client data twice.
When secure link checksum has length of 23 or 24 bytes, decoded base64 value
could occupy 17 or 18 bytes which is more than 16 bytes previously allocated
for it on stack. The buffer overflow does not have any security implications
since only one local variable was corrupted and this variable was not used in
this case.
The fix is to increase buffer size up to 18 bytes. Useless buffer size
initialization is removed as well.
This fixes at least the following cases, where no last_modified_time
(assuming caching is not enabled) resulted in incorrect behaviour:
- slice filter and If-Range requests (ticket #1357);
- If-Range requests with proxy_force_ranges;
- expires modified.
The $ssl_server_name variable used SSL_get_servername() result directly,
but this is not safe: it references a memory allocation in an SSL
session, and this memory might be freed at any time due to renegotiation.
Instead, copy the name to memory allocated from the pool.
This variable contains URL-encoded client SSL certificate. In contrast
to $ssl_client_cert, it doesn't depend on deprecated header continuation.
The NGX_ESCAPE_URI_COMPONENT variant of encoding is used, so the resulting
variable can be safely used not only in headers, but also as a request
argument.
The $ssl_client_cert variable should be considered deprecated now.
The $ssl_client_raw_cert variable will be eventually renambed back
to $ssl_client_cert.
Total length of a response with multiple ranges can be larger than a size_t
variable can hold, so type changed to off_t. Previously, an incorrect
Content-Length was returned when requesting more than 4G of ranges from
a large enough file on a 32-bit system.
An additional size_t variable introduced to calculate size of the boundary
header buffer, as off_t is not needed here and will require type casts on
win32.
Reported by Shuxin Yang,
http://mailman.nginx.org/pipermail/nginx/2017-July/054384.html.
The "fd" field should be after 3 pointers for ngx_event_ident() to use it.
This was broken by ccad84a174e0. While it does not seem to be currently used
for aio-related events, it should be a good idea to preserve the correct
layout nevertheless.
Pass NGX_FILE_OPEN to ngx_open_file() to fix "The parameter is incorrect"
error on win32 when using the ssl_session_ticket_key directive or loading
a binary geo base. On UNIX, this change is a no-op.
On Windows, a worker process does not call ngx_slab_init() from
ngx_init_zone_pool(), so ngx_slab_max_size, ngx_slab_exact_size,
and ngx_slab_exact_shift were left uninitialized.
The variable was considered non-existent in the absence of any
valid_referers directives.
Given the following config snippet,
location / {
return 200 $invalid_referer;
}
location /referer {
valid_referers server_names;
}
"location /" should work identically and independently on other
"location /referer".
The fix is to always add the $invalid_referer variable as long
as the module is compiled in, as is done by other modules.
The shared objects should generally be allocated from shared memory.
While peers->name and the data it points to allocated from cf->pool
happened to work on UNIX, it broke on Windows. On UNIX this worked
only because the shared memory zone for upstreams is re-created for
every new configuration.
But on Windows, a worker process does not inherit the address space
of the master process, so the peers->name pointed to data allocated
from cf->pool by the master process, and was invalid.
The phase is added instead of the try_files phase. Unlike the old phase, the
new one supports registering multiple handlers. The try_files implementation is
moved to a separate ngx_http_try_files_module, which now registers a precontent
phase handler.
The new request flag "preserve_body" indicates that the request body file should
not be removed by the upstream module because it may be used later by a
subrequest. The flag is set by the SSI (ticket #585), addition and slice
modules. Additionally, it is also set by the upstream module when a background
cache update subrequest is started to prevent the request body file removal
after an internal redirect. Only the main request is now allowed to remove the
file.
The http_rewrite module cannot be selected when http is disabled.
Fixed the PCRE check condition to avoid irrelevant check failure.
This is a regression from 4d874b4d82ed.
Signed-off-by: Samuel Martin <s.martin49@gmail.com>
When closing a socket with SO_REUSEPORT, Linux drops all connections waiting
in this socket's listen queue. Previously, it was believed to only result
in connection resets when reconfiguring nginx to use smaller number of worker
processes. It also results in connection resets during configuration
testing though.
Workaround is to avoid using SO_REUSEPORT when testing configuration. It
should prevent listening sockets from being created if a conflicting socket
already exists, while still preserving detection of other possible errors.
It should also cover UDP sockets.
The only downside of this approach seems to be that a configuration testing
won't be able to properly report the case when nginx was compiled with
SO_REUSEPORT, but the kernel is not able to set it. Such errors will be
reported on a real start instead.
Suffix ranges no longer allowed to set negative start values, to prevent
ranges with negative start from appearing even if total size protection
will be removed.
The overflow can be used to circumvent the restriction on total size of
ranges introduced in c2a91088b0c0 (1.1.2). Additionally, overflow
allows producing ranges with negative start (such ranges can be created
by using a suffix, "bytes=-100"; normally this results in 200 due to
the total size check). These can result in the following errors in logs:
[crit] ... pread() ... failed (22: Invalid argument)
[alert] ... sendfile() failed (22: Invalid argument)
When using cache, it can be also used to reveal cache file header.
It is believed that there are no other negative effects, at least with
standard nginx modules.
In theory, this can also result in memory disclosure and/or segmentation
faults if multiple ranges are allowed, and the response is returned in a
single in-memory buffer. This never happens with standard nginx modules
though, as well as known 3rd party modules.
Fix is to properly protect from possible overflow when incrementing size.
It is safe because re-sending still works during graceful shutdown as
long as resolving takes place (and resolve tasks set their own timeouts
that are not cancelable).
Also, the new ctx->cancelable flag can be set to make resolve task's
timeout event cancelable.
Notably, on ppc64 with 64k pagesize, slab 0 (of size 8) requires
128 64-bit elements for bitmasks. The code bogusly assumed that
one uintptr_t is enough for bitmasks plus at least one free slot.
Resolving an SRV record includes resolving its host names in subrequests.
Previously, if memory allocation failed while reporting a subrequest result
after receiving a response from a DNS server, the SRV resolve handler was
called immediately with the NGX_ERROR state. However, if the SRV record
included another copy of the resolved name, it was reported once again.
This could trigger the use-after-free memory access after SRV resolve
handler freed the resolve context by calling ngx_resolve_name_done().
Now the SRV resolve handler is called only when all its subrequests are
completed.
Previously, each configured header was represented in one of two ways,
depending on whether or not its value included any variables.
If the value didn't include any variables, then it would be represented
as as a single script that contained complete header line with HTTP/1.1
delimiters, i.e.:
"Header: value\r\n"
But if the value included any variables, then it would be represented
as a series of three scripts: first contained header name and the ": "
delimiter, second evaluated to header value, and third contained only
"\r\n", i.e.:
"Header: "
"$value"
"\r\n"
This commit changes that, so that each configured header is represented
as a series of two scripts: first contains only header name, and second
contains (or evaluates to) only header value, i.e.:
"Header"
"$value"
or
"Header"
"value"
This not only makes things more consistent, but also allows header name
and value to be accessed separately.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
As per RFC 2616 / RFC 7233, any range request to an empty file
is expected to result in 416 Range Not Satisfiable response, as
there cannot be a "byte-range-spec whose first-byte-pos is less
than the current length of the entity-body". On the other hand,
this makes use of byte-range requests inconvenient in some cases,
as reported for the slice module here:
http://mailman.nginx.org/pipermail/nginx-devel/2017-June/010177.html
This commit changes range filter to instead return 200 if the file
is empty and the range requested starts at 0.
This change reworks 13a5f4765887 to only run posted requests once,
with nothing on stack. Running posted requests with other request
functions on stack may result in use-after-free in case of errors,
similar to the one reported in #788.
To only run posted request once, a separate function was introduced
to be used as ssl handshake handler in c->ssl->handler,
ngx_http_upstream_ssl_handshake_handler(). The ngx_http_run_posted_requests()
is only called in this function, and not in ngx_http_upstream_ssl_handshake()
which may be called directly on stack.
Additionaly, ngx_http_upstream_ssl_handshake_handler() now does appropriate
debug logging of the current subrequest, similar to what is done in other
event handlers.
Previously, the upstream resolve handler always called
ngx_http_run_posted_requests() to run posted requests after processing the
resolver response. However, if the handler was called directly from the
ngx_resolve_name() function (for example, if the resolver response was cached),
running posted requests from the handler could lead to the following errors:
- If the request was scheduled for termination, it could actually be terminated
in the resolve handler. Upper stack frames could reference the freed request
object in this case.
- If a significant number of requests were posted, and for each of them the
resolve handler was called directly from the ngx_resolve_name() function,
posted requests could be run recursively and lead to stack overflow.
Now ngx_http_run_posted_requests() is only called from asynchronously invoked
resolve handlers.
Trailers added using this directive are evaluated after response body
is processed by output filters (but before it's written to the wire),
so it's possible to use variables calculated from the response body
as the trailer value.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Example:
ngx_table_elt_t *h;
h = ngx_list_push(&r->headers_out.trailers);
if (h == NULL) {
return NGX_ERROR;
}
ngx_str_set(&h->key, "Fun");
ngx_str_set(&h->value, "with trailers");
h->hash = ngx_hash_key_lc(h->key.data, h->key.len);
The code above adds "Fun: with trailers" trailer to the response.
Modules that want to emit trailers must set r->expect_trailers = 1
in header filter, otherwise they might not be emitted for HTTP/1.1
responses that aren't already chunked.
This change also adds $sent_trailer_* variables.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
The current style in variable handlers returning NGX_OK is to either set
v->not_found to 1, or to initialize the entire ngx_http_variable_value_t
structure.
In theory, always setting v->valid = 1 for NGX_OK would be useful, which
would mean that the value was computed and is thus valid, including the
special case of v->not_found = 1. But currently that's not the case and
causes the (v->valid || v->not_found) check to access an uninitialized
v->valid value, which is safe only because its value doesn't matter when
v->not_found is set.
When evaluating a mapped $reset_uid variable in the userid filter,
if get_handler set to ngx_http_map_variable() returned an error,
this previously resulted in a NULL pointer dereference.
If memory allocation of a new r->uri.data storage failed, reset its length as
well. Request URI is used in ngx_http_finalize_request() for debug logging.
Previously, when using NGX_HTTP_SSI_ERROR, error was ignored in ssi processing,
thus timefmt could be accessed later in ngx_http_ssi_date_gmt_local_variable()
as part of "set" handler, or NULL format pointer could be passed to strftime().
Previously, SETTINGS ACK was sent immediately upon receipt of SETTINGS
frame, before already queued DATA frames created using old SETTINGS.
This incorrect behavior was source of interoperability issues, because
peers rely on the fact that new SETTINGS are in effect after receiving
SETTINGS ACK.
Reported by Feng Li.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Previously, new frames could be emitted in the middle of applying
new (and already acknowledged) SETTINGS params, which is illegal.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
On Cygwin and NetBSD 7.0+ struct in_pktinfo has no ipi_spec_dst field, which
caused nginx compilation error. Now presence of this field is ensured by the
IP_PKTINFO feature test.
The problem was introduced by dbb0c854e308 (1.13.0).
If the main request was finalized while a background request performed an
asynchronous operation, the main request ended up in ngx_http_writer() and was
not finalized until a network event or a timeout. For example, cache
background update with aio enabled made nginx unable to process further client
requests or close the connection, keeping it open until client closes it.
Now regular finalization of the main request is not suspended because of an
asynchronous operation in another request.
If a background request was terminated while an asynchronous operation was in
progress, background request's write event handler was changed to
ngx_http_request_finalizer() and never called again.
Now, whenever a request is terminated while an asynchronous operation is in
progress, connection error flag is set to make further finalizations of any
request with this connection lead to termination.
These issues appeared in 1aeaae6e9446 (not yet released).
In http these checks were changed in a6d6d762c554, though mail module
was missed at that time. Since then, the stream module was introduced
based on mail, using "== NGX_ERROR" check.
With OpenSSL 1.1.0+, the workaround for handshake buffer size as introduced
in a720f0b0e083 (ticket #413) no longer works, as OpenSSL no longer exposes
handshake buffers, see https://github.com/openssl/openssl/commit/2e7dc7cd688.
Moreover, it is no longer possible to adjust handshake buffers at all now.
To avoid additional RTT if handshake uses more than 4k we now set TCP_NODELAY
on SSL connections before handshake. While this still results in sub-optimal
network utilization due to incomplete packets being sent, it seems to be
better than nothing.
Previously, cache background update might not work as expected, making client
wait for it to complete before receiving the final part of a stale response.
This could happen if the response could not be sent to the client socket in one
filter chain call.
Now background cache update is done in a background subrequest. This type of
subrequest does not block any other subrequests or the main request.
Previously, the read event of the accepted connection was marked ready, but not
available. This made EPOLLRDHUP-related code (for example, in ngx_unix_recv())
expect more data from the socket, leading to unexpected behavior.
For example, if SSL, PROXY protocol and deferred accept were enabled on a listen
socket, the client connection was aborted due to unexpected return value of
c->recv().
If allocation of cleanup handler in the HTTP/2 header filter failed, then
a stream might be freed with a HEADERS frame left in the output queue.
Now the HEADERS frame is accounted in the queue before trying to allocate
the cleanup handler.
Instead of highlighting directives in arbitrary positions, proper
parsing of nginx.conf syntax was implemented, matching what nginx does
internally. This allows vim to correctly highlight various complex cases,
including:
return 301 http://example.com/path#fragment";
and also avoids highlighting of parameters as directives, as in
server_name missing.semicolon.example.com
index index.php;
where "index" is not a directive but a parameter of the "server_name"
directive due to missing semicolon.
Most important downside of this approach seems to be that there is no
easy way to introduce directive-specific parameters. As such, only "listen"
directive parameters were preserved.
Abnormally exited workers may leave locked cache entries, this can
result in the cache size on disk exceeding max_size and shared memory
exhaustion.
This change mitigates the issue by ignoring locked entries during forced
expire. It also increases the visibility of the problem by logging such
entries.
Previously, an allocation error resulted in uninitialized memory access
when evaluating $upstream_http_ variables.
On a related note, see r->headers_out.headers cleanup work in 0cdee26605f3.
Oracle Developer Studio 12.5 introduced GCC-compatible __sync builtins.
Unfortunately, these builtins are neither GCC-compatible (they generate
warnings when used with volatile), nor working (unexpectedly fail on
unpredictable combinations of code layout and compiler flags). As such,
the gcc builtin atomic operations configure test explicitly disabled when
compiling with Sun C.
In ac9b1df5b246 (1.13.0) we attempted to allow renegotiation in client mode,
but when using OpenSSL 1.0.2 or older versions it was additionally disabled
by SSL3_FLAGS_NO_RENEGOTIATE_CIPHERS.
If initialization of a header failed for some reason after ngx_list_push(),
leaving the header as is can result in uninitialized memory access by
the header filter or the log module. The fix is to clear partially
initialized headers in case of errors.
For the Cache-Control header, the fix is to postpone pushing
r->headers_out.cache_control until its value is completed.
Previously, ngx_http_sub_header_filter() could fail with a partially
initialized context, later accessed in ngx_http_sub_body_filter()
if called from the perl content handler.
The issue had appeared in 2c045e5b8291 (1.9.4).
A better fix would be to handle ngx_http_send_header() errors in
the perl module, though this doesn't seem to be easy enough.
The SSL_CTRL_SET_CURVES_LIST macro is removed in the OpenSSL master branch.
SSL_CTX_set1_curves_list is preserved as compatibility with previous versions.
CVE-2009-3555 is no longer relevant and mitigated by the renegotiation
info extension (secure renegotiation). On the other hand, unexpected
renegotiation still introduces potential security risks, and hence we do
not allow renegotiation on the server side, as we never request renegotiation.
On the client side the situation is different though. There are backends
which explicitly request renegotiation, and disabled renegotiation
introduces interoperability problems. This change allows renegotiation
on the client side, and fixes interoperability problems as observed with
such backends (ticket #872).
Additionally, with TLSv1.3 the SSL_CB_HANDSHAKE_START flag is currently set
by OpenSSL when receiving a NewSessionTicket message, and was detected by
nginx as a renegotiation attempt. This looks like a bug in OpenSSL, though
this change also allows better interoperability till the problem is fixed.
Previously, the source IP address of a response UDP datagram could differ from
the original datagram destination address. This could happen if the server UDP
socket is bound to a wildcard address and the network interface chosen to output
the response packet has a different default address than the destination address
of the original packet. For example, if two addresses from the same network are
configured on an interface.
Now source address is set explicitly if a response is sent for a server UDP
socket bound to a wildcard address.
This change adds "http_429" parameter to "proxy_next_upstream" for
retrying rate-limited requests, and to "proxy_cache_use_stale" for
serving stale cached responses after being rate-limited.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
This change adds reason phrase in status line and pretty response body
when "429" status code is used in "return", "limit_conn_status" and/or
"limit_req_status" directives.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
When a slice subrequest was redirected to a new location, its context was lost.
After its completion, a new slice subrequest for the same slice was created.
This could lead to infinite loop. Now the slice module makes sure each slice
subrequest starts output with the slice context available.
With post_action or subrequests, it is possible that the timer set for
wev->delayed will expire while the active subrequest write event handler
is not ready to handle this. This results in request hangs as observed
with limit_rate / sendfile_max_chunk and post_action (ticket #776) or
subrequests (ticket #1228).
Moving the handling to the connection event handler fixes the hangs observed,
and also slightly simplifies the code.
Since limit_req uses connection's write event to delay request processing,
it can conflict with timers in other subrequests. In particular, even
if applied to an active subrequest, it can break things if wev->delayed
is already set (due to limit_rate or sendfile_max_chunk), since after
limit_req finishes the wev->delayed flag will be set and no timer will be
active.
Fix is to use the wev->delayed flag in limit_req as well. This ensures that
wev->delayed won't be set after limit_req finishes, and also ensures that
limit_req's timers will be properly handled by other subrequests if the one
delayed by limit_req is not active.
All streams in connection must be finalized before the connection
itself can be finalized and all related memory is freed. That's
not always possible on the current event loop iteration.
Thus when the last stream is finalized, it sets the special read
event handler ngx_http_v2_handle_connection_handler() and posts
the event.
Previously, this handler didn't check the connection state and
could call the regular event handler on a connection that was
already in finalization stage. In the worst case that could
lead to a segmentation fault, since some data structures aren't
supposed to be used during connection finalization. Particularly,
the waiting queue can contain already freed streams, so the
WINDOW_UPDATE frame received by that moment could trigger
accessing to these freed streams.
Now, the connection error flag is explicitly checked in
ngx_http_v2_handle_connection_handler().
In order to finalize stream the error flag is set on fake connection and
either "write" or "read" event handler is called. The read events of fake
connections are always ready, but it's not the case with the write events.
When the ready flag isn't set, the error flag can be not checked in some
cases and as a result stream isn't finalized. Now the ready flag is
explicilty set on write events for proper finalization in all cases.
Previously, flow control didn't account for padding in DATA frames,
which meant that its view of the world could drift from peer's view
by up to 256 bytes per received padded DATA frame, which could lead
to a deadlock.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Previously, its value accounted for payloads of HEADERS, CONTINUATION
and DATA frames, as well as frame headers of HEADERS and DATA frames,
but it didn't account for frame headers of CONTINUATION frames.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Previously, connection write handler was called, resulting in wake up
of the active subrequest. This change makes it possible to read data
in non-active subrequests as well. For example, this allows SSI to
process instructions in non-active subrequests earlier and start
additional subrequests if needed, reducing overall response time.
If the subrequest is already finalized, the handler set with aio_write
may still be used by sendfile in threads when using range requests
(see also e4c1f5b32868, and the original note in 9fd738b85fad). Calling
already finalized subrequest's r->write_event_handler in practice
results in request hang in some cases.
Fix is to trigger connection event handler if the subrequest was already
finalized.
The ngx_linux_sendfile() function is now used for both normal sendfile()
and sendfile in threads. The ngx_linux_sendfile_thread() function was
modified to use the same interface as ngx_linux_sendfile(), and is simply
called from ngx_linux_sendfile() when threads are enabled.
Special return code NGX_DONE is used to indicate that a thread task was
posted and no further actions are needed.
If number of bytes sent is less that what we were sending, we now always
retry sending. This is needed for sendfile() in threads as the number
of bytes we are sending might have been changed since the thread task
was posted. And this is also needed for Linux 4.3+, as sendfile() might
be interrupted at any time and provides no indication if it was interrupted
or not (ticket #1174).
If of.err is 0, it means that there was a memory allocation error
and no further logging and/or processing is needed. The of.failed
string can be only accessed if of.err is not 0.
Some combinations of options might cause the builds with the
--with-stream option to break due to invalid value of the
STREAM_INCS make variable, e.g.
auto/configure \
--with-stream \
--with-http_perl_module=dynamic \
--without-http_memcached_module \
--without-http_empty_gif_module \
--without-http_browser_module \
--without-http_upstream_hash_module \
--without-http_upstream_ip_hash_module \
--without-http_upstream_least_conn_module \
--without-http_upstream_keepalive_module \
--without-http_upstream_zone_module \
Explicit initialization of ngx_module_libs and ngx_module_link
matches what we already do when processing mail modules, and
is also required after the next change.
The directive configures a timeout to be used when gracefully shutting down
worker processes. When the timer expires, nginx will try to close all
the connections currently open to facilitate shutdown.
There is no need to cancel timers early if there are other timers blocking
shutdown anyway. Preserving such timers allows nginx to continue some
periodic work till the shutdown is actually possible.
With the new approach, timers with ev->cancelable are simply ignored when
checking if there are any timers left during shutdown.
The ev->timedout flag is set on first timer expiration, and never reset
after it. Due to this the code to stop the timer when the timer was
canceled never worked (except in a very specific time frame immediately
after start), and the timer was always armed again. This essentially
resulted in a buffer flush at the end of an event loop iteration.
This behaviour actually seems to be better than just stopping the flush
timer for the whole shutdown, so it is preserved as is instead of fixing
the code to actually remove the timer. It will be further improved by
upcoming changes to preserve cancelable timers if there are other timers
blocking shutdown.
Most notably, this fixes possible buffer overflows if number of large
client header buffers in a virtual server is different from the one in
the default server.
Reported by Daniil Bondarev.
Cloned subrequests should inherit r->content_handler. This way they will
be able to use the same location configuration as the original request
if there are "if" directives in the configuration.
Without r->content_handler inherited, the following configuration tries
to access a static file in the update request:
location / {
set $true 1;
if ($true) {
# nothing
}
proxy_pass http://backend;
proxy_cache one;
proxy_cache_use_stale updating;
proxy_cache_background_update on;
}
See http://mailman.nginx.org/pipermail/nginx/2017-February/053019.html for
initial report.
Dates in ISO 8601 format are in line with what is used in nginx.org news,
and are generally less ambiguous and easier to work with.
The changes.xml was converted using sed:
$ sed 's/date="\(..\)\.\(..\)\.\(....\)"/date="\3-\2-\1"/g' changes.xml
Appropriate changes to changes.xsls were introduced.
With "proxy_ignore_client_abort off" (the default), upstream module changes
r->read_event_handler to ngx_http_upstream_rd_check_broken_connection().
If the handler is not cleared during upstream finalization, it can be
triggered later, causing unexpected effects, if, for example, a request
was redirected to a different location using error_page or X-Accel-Redirect.
In particular, it makes "proxy_ignore_client_abort on" non-working after
a redirection in a configuration like this:
location = / {
error_page 502 = /error;
proxy_pass http://127.0.0.1:8082;
}
location /error {
proxy_pass http://127.0.0.1:8083;
proxy_ignore_client_abort on;
}
It is also known to cause segmentation faults with aio used, see
http://mailman.nginx.org/pipermail/nginx-ru/2015-August/056570.html.
Fix is to explicitly set r->read_event_handler to ngx_http_block_reading()
during upstream finalization, similar to how it is done in the request body
reading code and in the limit_req module.
This allows to store larger ETag values for proxy_cache_revalidate,
including ones generated as SHA256, and cache responses with longer
Vary (ticket #826).
In particular, this fixes caching of Amazon S3 responses with CORS
enabled, which now use "Vary: Origin, Access-Control-Request-Headers,
Access-Control-Request-Method".
Cache version bumped accordingly.
Previously, slice subrequest location was selected based on request URI.
If request is then redirected to a new location, its context array is cleared,
making the slice module loose current slice range information. This lead to
broken output. Now subrequests with the NGX_HTTP_SUBREQUEST_CLONE flag are
created for slices. Such subrequests stay in the same location as the parent
request and keep the right slice context.
Previously, there was no way to enable the proxy_cache_use_stale behavior by
reading the backend response. Now, stale-while-revalidate and stale-if-error
Cache-Control extensions (RFC 5861) are supported. They specify, how long a
stale response can be used when a cache entry is being updated, or in case of
an error.
The function may leave error in the error queue while returning success,
e.g., when taking a DSO reference to itself as of OpenSSL 1.1.0d:
https://git.openssl.org/?p=openssl.git;a=commit;h=4af9f7f
Notably, this fixes alert seen with statically linked OpenSSL on some platforms.
While here, check OPENSSL_init_ssl() return value.
Previously, buffer size was not changed from the one saved during
initial ngx_ssl_create_connection(), even if the buffer itself was not
yet created. Fix is to change c->ssl->buffer_size in the SNI callback.
Note that it should be also possible to update buffer size even in non-SNI
virtual hosts as long as the buffer is not yet allocated. This looks
like an overcomplication though.
The ngx_event_pipe() function wasn't called on write events with
wev->delayed set. As a result, threaded writing results weren't
properly collected in ngx_event_pipe_write_to_downstream() when a
write event was triggered for a completed write.
Further, this wasn't detected, as p->aio was reset by a thread completion
handler, and results were later collected in ngx_event_pipe_read_upstream()
instead of scheduling a new write of additional data. If this happened
on the last reading from an upstream, last part of the response was never
written to the cache file.
Similar problems might also happen in case of timeouts when writing to
client, as this also results in ngx_event_pipe() not being called on write
events. In this scenario socket leaks were observed.
Fix is to check if p->writing is set in ngx_event_pipe_read_upstream(), and
therefore collect results of previous write operations in case of read events
as well, similar to how we do so in ngx_event_pipe_write_downstream().
This is enough to fix the wev->delayed case. Additionally, we now call
ngx_event_pipe() from ngx_http_upstream_process_request() if there are
uncollected write operations (p->writing and !p->aio). This also fixes
the wev->timedout case.
The ngx_chain_coalesce_file() function may produce more bytes to send then
requested in the limit passed, as it aligns the last file position
to send to memory page boundary. As a result, (limit - send) may become
negative. This resulted in big positive number when converted to size_t
while calling ngx_output_chain_to_iovec().
Another part of the problem is in ngx_chain_coalesce_file(): it changes cl
to the next chain link even if the current buffer is only partially sent
due to limit.
Therefore, if a file buffer was not expected to be fully sent due to limit,
and was followed by a memory buffer, nginx called sendfile() with a part
of the file buffer, and the memory buffer in trailer. If there were enough
room in the socket buffer, this resulted in a part of the file buffer being
skipped, and corresponding part of the memory buffer sent instead.
The bug was introduced in 8e903522c17a (1.7.8). Configurations affected
are ones using limits, that is, limit_rate and/or sendfile_max_chunk, and
memory buffers after file ones (may happen when using subrequests or
with proxying with disk buffering).
Fix is to explicitly check if (send < limit) before constructing trailer
with ngx_output_chain_to_iovec(). Additionally, ngx_chain_coalesce_file()
was modified to preserve unfinished file buffers in cl.
Closing up to 32 connections might be too aggressive if worker_connections
is set to a comparable number (and/or there are only a small number of
reusable connections). If an occasional connection shorage happens in
such a configuration, it leads to closing all reusable connections instead
of gradually reducing keepalive timeout to a smaller value. To improve
granularity in such configurations we now close no more than 1/8 of all
reusable connections at once.
Suggested by Joel Cunningham.
A missing check could cause ngx_stream_ssl_handler() to be applied
to a non-ssl session, which resulted in a null pointer dereference
if ssl_verify_client is enabled.
The bug had appeared in 1.11.8 (41cb1b64561d).
Previously, an unavailable peer was considered recovered after a successful
proxy session to this peer. Until then, only a single client connection per
fail_timeout was allowed to be proxied to the peer.
Since stream sessions can be long, it may take indefinite time for a peer to
recover, limiting the ability of the peer to receive new connections.
Now, a peer is considered recovered after a successful TCP connection is
established to it. Balancers are notified of this event via the notify()
callback.
OpenSSL 1.1.0 now uses normal "nmake; nmake install" instead of using
custom "ms\do_ms.bat" script and "ms\nt.mak" makefile. And Configure
now requires --prefix to be absolute, and no longer derives --openssldir
from prefix (so it's specified explicitly). Generated libraries are now
called "libcrypto.lib" and "libssl.lib" instead of "libeay32.lib"
and "ssleay32.lib". Appropriate tests added to support both old and new
variants.
Additionally, openssl/lhash.h now triggers warning C4090 ('function' :
different 'const' qualifiers), so the warning was disabled.
There are lots of C4244 warnings (conversion from 'type1' to 'type2',
possible loss of data), so they were disabled.
The same applies to C4267 warnings (conversion from 'size_t' to 'type',
possible loss of data), most notably - conversion from ngx_str_t.len to
ngx_variable_value_t.len (which is unsigned:28). Additionally, there
is at least one case when it is not possible to fix the warning properly
without introducing win32-specific code: recv() on win32 uses "int len",
while POSIX defines "size_t len".
The ssize_t type now properly defined for 64-bit compilation with MSVC.
Caught by warning C4305 (truncation from '__int64' to 'ssize_t'), on
"cutoff = NGX_MAX_SIZE_T_VALUE / 10" in ngx_atosz()).
Several C4334 warnings (result of 32-bit shift implicitly converted to 64 bits)
were fixed by adding explicit conversions.
Several C4214 warnings (nonstandard extension used: bit field types other
than int) in ngx_http_script.h fixed by changing bit field types from
uintptr_t to unsigned.
Most notably, warning W8012 (comparing signed and unsigned values) reported
in multiple places where an unsigned value of small type (e.g., u_short) is
promoted to an int and compared to an unsigned value.
Warning W8072 (suspicious pointer arithmetic) disabled, it is reported
when we increment base pointer in ngx_shm_alloc().
These types are available with MSVC (at least since 2003, in stddef.h),
all variants of GCC (in stdint.h) and Watcom C. We need to define them
only for Borland C.
This allows to set a different one from command line when needed.
For example, to configure nginx with gcc as a compiler one could
use "make -f misc/GNUmakefile win32 CC=gcc".
This implies ticket key size of 80 bytes instead of previously used 48,
as both HMAC and AES keys are 32 bytes now. When an old 48-byte ticket key
is provided, we fall back to using backward-compatible AES128 encryption.
OpenSSL switched to using AES256 in 1.1.0, and we are providing equivalent
security. While here, order of HMAC and AES keys was reverted to make
the implementation compatible with keys used by OpenSSL with
SSL_CTX_set_tlsext_ticket_keys().
Prodded by Christian Klinger.
The current version of HTTP/1.1 standard allows relative references in
redirects (https://tools.ietf.org/html/rfc7231#section-7.1.2).
Allow this form for redirects generated by nginx by introducing the new
directive absolute_redirect.
SSL version 3.0 can be specified by the client at the record level for
compatibility reasons. Previously, ssl_preread module rejected such
connections, presuming they don't have SNI. Now SSL 3.0 is allowed at
the record level.
The resolver handles SRV requests in two stages. In the first
stage it gets all SRV RRs, and in the second stage it resolves
the names from SRV RRs into addresses.
Previously, if a response to an SRV request was cached, the
queries to resolve names were not limited by a timeout. If a
response to any of these queries was not received, the SRV
request could never complete.
If a response to an SRV request was not cached, and some of the
queries to resolve names timed out, NGX_RESOLVE_TIMEDOUT was
returned instead of successfully resolved addresses.
To fix both issues, resolving of names is now always limited by
a timeout.
Changeset e7cb5deb951d breaks build on CentOS 5 with "dereferencing
type-punned pointer will break strict-aliasing rules" warning. It is
backed out.
Instead, to keep builds with BoringSSL happy, type of the "value"
variable changed to "char *", and an explicit cast added before calling
ngx_parse_http_time().
A bug was introduced by 82efcedb310b that could lead to timing out of
responses or segmentation fault, when accept_mutex was enabled.
The output queue in HTTP/2 can contain frames from different streams.
When the queue is sent, all related write handlers need to be called.
In order to do so, the streams were added to the h2c->posted queue
after handling sent frames. Then this queue was processed in
ngx_http_v2_write_handler().
If accept_mutex is enabled, the event's "ready" flag is set but its
handler is not called immediately. Instead, the event is added to
the ngx_posted_events queue. At the same time in this queue can be
events from upstream connections. Such events can result in sending
output queue before ngx_http_v2_write_handler() is triggered. And
at the time ngx_http_v2_write_handler() is called, the output queue
can be already empty with some streams added to h2c->posted.
But after 82efcedb310b, these streams weren't processed if all frames
have already been sent and the output queue was empty. This might lead
to a situation when a number of streams were get stuck in h2c->posted
queue for a long time. Eventually these streams might get closed by
the send timeout.
In the worst case this might also lead to a segmentation fault, if
already freed stream was left in the h2c->posted queue. This could
happen if one of the streams was terminated but wasn't closed, due to
the HEADERS frame or a partially sent DATA frame left in the output
queue. If this happened the ngx_http_v2_filter_cleanup() handler
removed the stream from the h2c->waiting or h2c->posted queue on
termination stage, before the frame has been sent, and the stream
was again added to the h2c->posted queue after the frame was sent.
In order to fix all these problems and simplify the code, write
events of fake stream connections are now added to ngx_posted_events
instead of using a custom h2c->posted queue.
By default, "map" creates cacheable variables [1]. With this
parameter it creates a non-cacheable variable.
An original idea was to deduce the cacheability of the "map"
variable by checking the cacheability of variables specified
in source and resulting values, but it turned to be too hard.
For example, a cacheable variable can be overridden with the
"set" directive or with the SSI "set" command. Also, keeping
"map" variables cacheable by default is good for performance
reasons. This required adding a new parameter.
[1] Before db699978a33f (1.11.0), the cacheability of the
"map" variable could vary depending on the cacheability of
variables specified in resulting values (ticket #1090).
This is believed to be a bug rather than a feature.
Removed code that would cause an endless loop, and removed condition
check that is always false. The first page in the slot list is
guaranteed to satisfy an allocation.
On exit environment allocated from a pool is no longer available, leading
to a segmentation fault if, for example, a library tries to use it from
an atexit() handler.
Fix is to allocate environment via ngx_alloc() instead, and explicitly
free it using a pool cleanup handler if it's no longer used (e.g., on
configuration reload).
In Perl 5.8.6 the default was switched to use putenv() when used as
embedded library unless "PL_use_safe_putenv = 0" is explicitly used
in the code. Therefore, for modern versions of Perl it is no longer
necessary to restore previous environment when calling perl_destruct().
For Perl compiled with threads, without PERL_SET_INTERP() the PL_curinterp
remains set to the first interpreter created (that is, one created at
original start). As a result after a reload Perl thinks that operations
are done withing a thread, and, most notably, denies to change environment.
For example, the following code properly works on original start,
but fails after a reload:
perl 'sub {
my $r = shift;
$r->send_http_header("text/plain");
$ENV{TZ} = "UTC";
$r->print("tz: " . $ENV{TZ} . " (localtime " . (localtime()) . ")\n");
$ENV{TZ} = "Europe/Moscow";
$r->print("tz: " . $ENV{TZ} . " (localtime " . (localtime()) . ")\n");
return OK;
}';
To fix this, PERL_SET_INTERP() added anywhere where PERL_SET_CONTEXT()
was previously used.
Note that PERL_SET_INTERP() doesn't seem to be documented anywhere.
Yet it is used in some other software, and also seems to be the only
solution possible.
Atom size is the sum of atom header size and atom data size. The
specification says that the first 4 bytes are set to one when
the atom size is greater than the maximum unsigned 32-bit value.
Which means atom header size should be considered when the
comparison takes place between atom data size and 0xffffffff.
The variable contains a list of curves as supported by the client.
Known curves are listed by their names, unknown ones are shown
in hex, e.g., "0x001d:prime256v1:secp521r1:secp384r1".
Note that OpenSSL uses session data for SSL_get1_curves(), and
it doesn't store full list of curves supported by the client when
serializing a session. As a result $ssl_curves is only available
for new sessions (and will be empty for reused ones).
The variable is only meaningful when using OpenSSL 1.0.2 and above.
With older versions the variable is empty.
The variable contains list of ciphers as supported by the client.
Known ciphers are listed by their names, unknown ones are shown
in hex, e.g., ""AES128-SHA:AES256-SHA:0x00ff".
The variable is fully supported only when using OpenSSL 1.0.2 and above.
With older version there is an attempt to provide some information
using SSL_get_shared_ciphers(). It only lists known ciphers though.
Moreover, as OpenSSL uses session data for SSL_get_shared_ciphers(),
and it doesn't store relevant data when serializing a session. As
a result $ssl_ciphers is only available for new sessions (and not
available for reused ones) when using OpenSSL older than 1.0.2.
Now in case of a verification failure $ssl_client_verify contains
"FAILED:<reason>", similar to Apache's SSL_CLIENT_VERIFY, e.g.,
"FAILED:certificate has expired".
Detailed description of possible errors can be found in the verify(1)
manual page as provided by OpenSSL.
Normally, the epoll module calls the read and write handlers depending
on whether EPOLLIN and EPOLLOUT are reported by epoll_wait(). No error
processing is done in the module, the handlers are expected to get an
error when doing I/O.
If an error event is reported without EPOLLIN and EPOLLOUT, the module
set both EPOLLIN and EPOLLOUT to ensure the error event is handled at
least in one active handler.
This works well unless the error is delivered along with only one of
EPOLLIN or EPOLLOUT, and the corresponding handler does not do any I/O.
For example, it happened when getting EPOLLERR|EPOLLOUT from
epoll_wait() upon receiving "ICMP port unreachable" while proxying UDP.
As the write handler had nothing to send it was not able to detect and
log an error, and did not switch to the next upstream.
The fix is to unconditionally set EPOLLIN and EPOLLOUT in case of an
error event. In the aforementioned case, this causes the read handler
to be called which does recv() and detects an error.
In addition to the epoll module, analogous changes were made in
devpoll/eventport/poll.
Previously, a request body bigger than "client_body_buffer_size" wasn't written
into a temporary file if it has been pre-read entirely. The preread buffer
is freed after processing, thus subsequent use of it might result in sending
corrupted body or cause a segfault.
Dependencies of dynamic modules are added to NGX_ADDON_DEPS (and
it is now used for dynamic modules) to be in line with what happens
in case of static compilation.
To avoid duplication, MAIL_DEPS and STREAM_DEPS are no longer passed
to auto/module when these modules are compiled as dynamic ones. Mail
and stream dependencies are handled explicitly via corresponding
variables.
On Linux, the rename syscall can be slow due to a global file system lock,
acquired for the entire rename operation, unless both old and new files are
in the same directory. To address this temporary files are now created
in the same directory as the expected resulting cache file when using the
"use_temp_path=off" parameter.
This change mostly reverts 99639bfdfa2a and 3281de8142f5, restoring the
behaviour as of a9138c35120d (with minor changes).
Holding a cache node lock doesn't make sense as we can't use caching
anyway, and results in "ignore long locked inactive cache entry" alerts
if a node is locked for a long time.
The same is done for unbuffered connections, as they can be alive for
a long time as well.
It configures a threshold in bytes, above which client range
requests are not cached. In such a case the client's Range
header is passed directly to a proxied server.
As the pointer to the first argument was tested instead of the argument
itself, array of arguments was always created, even if there were no
arguments. Fix is to test args[0] instead of args.
Found by Coverity (CID 1356862).
The only thing that default_port comparison did in the current
code is prevented implicit upstreams to the same address/port
from being aliased for http and https, e.g.:
proxy_pass http://10.0.0.1:12345;
proxy_pass https://10.0.0.1:12345;
This is inconsistent because it doesn't work for a similar case
with uswgi_pass:
uwsgi_pass uwsgi://10.0.0.1:12345;
uwsgi_pass suwsgi://10.0.0.1:12345;
or with an explicit upstream:
upstream u {
server 10.0.0.1:12345;
}
proxy_pass http://u;
proxy_pass https://u;
Before c9059bd5445b, default_port comparison was needed to
differentiate implicit upstreams in
proxy_pass http://example.com;
and
proxy_pass https://example.com;
as u->port was not set.
When an upstream{} block follows a proxy_pass reference to it,
such an upstream inherited port and default_port settings from
proxy_pass. This was different from when they came in another
order (see ticket #1059). Explicit upstreams should not have
port and default_port in any case.
This fixes the following case:
server { location / { proxy_pass http://u; } ... }
upstream u { server 127.0.0.1; }
server { location / { proxy_pass https://u; } ... }
but not the following:
server { location / { proxy_pass http://u; } ... }
server { location / { proxy_pass https://u; } ... }
upstream u { server 127.0.0.1; }
If proxy_pass (and friends) with variables evaluates an upstream
specified with literal address, nginx always created a per-request
upstream.
Now, if there's a matching upstream specified in the configuration
(either implicit or explicit), it will be used instead.
This fixes inconsistency in what is stored in the "host" field.
Normally it would contain the "host" part of the parsed URL
(e.g., proxy_pass with variables), but for the case of an
implicit upstream specified with literal address it contained
the text representation of the socket address (that is, host
including port for IP).
Now the "host" field always contains the "host" part of the URL,
while the text representation of the socket address is stored
in the newly added "name" field.
The ngx_http_upstream_create_round_robin_peer() function was
modified accordingly in a way to be compatible with the code
that does not know about the new "name" field.
The "stream" code was similarly modified except for not adding
compatibility in ngx_stream_upstream_create_round_robin_peer().
This change is also a prerequisite for the next change.
The new directive "http2_max_requests" is introduced. From users point of
view it works quite similar to "keepalive_requests" but has significantly
bigger default value that is more suitable for HTTP/2.
This allows to correctly parse "start" and "end" arguments without
null-termination (ticket #475), and also fixes rounding errors observed
with strtod() when using i387 instructions.
Originally, the variables kept a result of X509_NAME_oneline(),
which is, according to the official documentation, a legacy
function. It produces a non standard output form and has
various quirks and inconsistencies.
The RFC2253 compliant behavior is introduced for these variables.
The original variables are available through $ssl_client_s_dn_legacy
and $ssl_client_i_dn_legacy.
Previously, while shutting down gracefully, the HTTP/2 connections were
closed in transition to idle state after all active streams have been
processed. That might never happen if the client continued opening new
streams.
Now, nginx sends GOAWAY to all HTTP/2 connections and ignores further
attempts to open new streams. A worker process will quit as soon as
processing of already opened streams is finished.
BoringSSL changed SSL_set_tlsext_host_name() to be a real function
with a (const char *) argument, so it now triggers a warning due to
conversion from (u_char *). Added an explicit cast to silence the
warning.
Prodded by Piotr Sikora, Alessandro Ghedini.
This is needed to allow TLS client certificate auth to work. With
ssl_verify_client configured, the auth daemon can choose to allow the
connection to proceed based on the certificate data.
This has been tested with Thunderbird for IMAP only. I've not yet found a
client that will do client certificate auth for POP3 or SMTP, and the method is
not really documented anywhere that I can find. That said, its simple enough
that the way I've done is probably right.
When headers are set at the "http" level and not redefined in
a server block, we now preserve conf->headers into the "http"
section configuration to inherit it to all servers.
The same applies to conf->headers_cache, though it may not be effective
if no servers use cache at the "server" level as conf->headers_cache
is only initialized if cache is enabled on a given level.
Similar changes made in fastcgi/scgi/uwsgi to preserve conf->params
and conf->params_cache.
When headers to hide are set at the "http" level and not redefined in
a server block, we now preserve compiled headers hash into the "http"
section configuration to inherit this hash to all servers.
Dependency on cache settings existed prior to 2728c4e4a9ae (0.8.44)
as Set-Cookie header was automatically hidden from responses when
using cache. This is no longer the case, and hide_headers_hash can
be safely inherited regardless of cache settings.
With this change it is now possible to load modules compiled without
the "--with-http_ssl_module" configure option into nginx binary compiled
with it, and vice versa (if a module doesn't use ssl-specific functions),
assuming both use the "--with-compat" option.
With this change it is now possible to load modules compiled without
the "--with-file-aio" configure option into nginx binary compiled with it,
and vice versa, assuming both use the "--with-compat" option.
With this change it is now possible to load modules compiled without
the "--with-threads" configure option into nginx binary compiled with it,
and vice versa (if a module does not use thread-specific functions),
assuming both use the "--with-compat" option.
It is used at least by SOAP (M-POST method, defined by RFC 2774) and
by WebDAV versioning (VERSION-CONTROL and BASELINE-CONTROL methods,
defined by RFC 3253).
Previously, user access bits were always set to "rw" unconditionally,
even with "user:r" explicitly specified. With this change we only add
default user access bits (0600) if they weren't set explicitly.
Duplicate processing was possible if the address set by realip was
listed in set_realip_from, and there was an internal redirect so module
context was cleared. This resulted in exactly the same address being set,
so this wasn't a problem before the $realip_remote_addr variable was
introduced, though now results in incorrect $realip_remote_addr being
picked.
Fix is to use ngx_http_realip_get_module_ctx() to look up module context
even if it was cleared. Additionally, the order of checks was switched to
check the configuration first as it looks more effective.
The new parameters "manager_files", "manager_sleep"
and "manager_threshold" were added to proxy_cache_path
and friends.
Note that ngx_path_manager_pt was changed to return ngx_msec_t
instead of time_t (API change).
Explicit checks for OPENSSL_VERSION_NUMBER replaced with checks
for X509_CHECK_FLAG_ALWAYS_CHECK_SUBJECT, thus allowing X509_check_host()
to be used with other libraries. In particular, X509_check_host() was
introduced in LibreSSL 2.5.0.
IPv6 now compiled-in automatically if support is found. If there is a need
to disable it for some reason, --with-cc-opt="-DNGX_HAVE_INET6=0" can be used
for this.
When the last_buf flag is cleared for add_after_body to append more data from a
subrequest, other filters may still have buffered data, which should be flushed
at this point. For example, the sub_filter may have a partial match buffered,
which will only be flushed after the subrequest is done, ending up with
interleaved data in output.
Setting last_in_chain instead of last_buf flushes the data and fixes the order
of output buffers.
The last_buf flag should only be set in the last buffer of the main request.
Otherwise, several last_buf flags can appear in output. This can, for example,
break the chunked filter, which will include several final chunks in output.
The IPV6_V6ONLY macro is now checked only while parsing appropriate flag
and when using the macro.
The ipv6only field in listen structures is always initialized to 1,
even if not supported on a given platform. This is expected to prevent
a module compiled without IPV6_V6ONLY from accidentally creating dual
sockets if loaded into main binary with proper IPV6_V6ONLY support.
It is to be used as a bitmask with various bits set/reset when appropriate.
Any bit set means that the peer should not be used, that is, exactly what
current checks do, no additional changes required.
Previously flags passed by --with-ld-opt were not used when building perl
module, which meant hardening flags provided by package build systems were not
applied.
All the errors that prevent loading configuration must be printed on the "emerg"
log level. Previously, nginx might silently fail to load configuration in some
cases as the default log level is "error".
The ssl_preread module extracts information from the SSL Client Hello message
without terminating SSL. Currently, only $ssl_preread_server_name variable
is supported, which contains server name from the SNI extension.
In this phase, head of a stream is read and analysed before proceeding to the
content phase. Amount of data read is controlled by the module implementing
the phase, but not more than defined by the "preread_buffer_size" directive.
The time spent on processing preread is controlled by the "preread_timeout"
directive.
The typical preread phase module will parse the beginning of a stream and set
variable that may be used by the content phase, for example to make routing
decision.
Previously, it was not possible to use the stream context
inside ngx_stream_init_connection() handlers. Now, limit_conn,
access handlers, as well as those added later, can create
their own contexts.
Previously, it was possible that some system calls could be
invoked while holding the accept mutex. This is clearly
wrong as it prevents incoming connections from being accepted
as quickly as possible.
Keeps the full address of the upstream server. If several servers were
contacted during proxying, their addresses are separated by commas,
e.g. "192.168.1.1:80, 192.168.1.2:80".
The stream session status is one of the following:
200 - normal completion
403 - access forbidden
500 - internal server error
502 - bad gateway
503 - limit conn
This fixes a problem with aio threads and sendfile with aio_write switched
off, as observed with range requests after fc72784b1f52 (1.9.13). Potential
problems with sendfile in threads were previously described in 9fd738b85fad,
and this seems to be one of them.
The problem occurred as file's thread_handler was set to NULL by event pipe
code after a sendfile thread task was scheduled. As a result, no sendfile
completion code was executed, and the same buffer was additionally sent
using non-threaded sendfile. Fix is to avoid modifying file's thread_handler
if aio_write is switched off.
Note that with "aio_write on" it is still possible that sendfile will use
thread_handler as set by event pipe. This is believed to be safe though,
as handlers used are compatible.
When c->recv_chain() returns an error, it is possible that we already
have some data previously read, e.g., in preread buffer. And in some
cases it may be even a complete response. Changed c->recv_chain() error
handling to process the data, much like it is already done if kevent
reports about an error.
This change, in particular, fixes processing of small responses
when an upstream fails to properly close a connection with lingering and
therefore the connection is reset, but the response is already fully
obtained by nginx (see ticket #1037).
Previously, the realip module could be left with uninitialized context after an
error in the ngx_http_realip_set_addr() function. That context could be later
accessed by $realip_remote_addr and $realip_remote_port variable handlers.
This prevents theoretical resource leak, since those threads are never joined.
Found with ThreadSanitizer.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
In recent Perl versions unpack("C*") unpacks wide characters by default,
likely since perl 5.10 (seen at least in perl 5.20). Replaced with
unpack("U0C*") instead to unpack bytes.
While here, improved style and updated my email.
If the range includes two or more /16 networks and does
not start at the /16 boundary, the last subrange was not
removed (see 91cff7f97a50 for details).
Return 1 in the SSL_CTX_set_tlsext_ticket_key_cb() callback function
to indicate that a new session ticket is created, as per documentation.
Until 1.1.0, OpenSSL didn't make a distinction between non-negative
return values.
See https://git.openssl.org/?p=openssl.git;a=commitdiff;h=5c753de for details.
The IP_BIND_ADDRESS_NO_PORT option is set on upstream sockets
if proxy_bind does not specify a port. The SO_REUSEADDR option
is set on UDP upstream sockets if proxy_bind specifies a port.
Due to checking of the wrong port, IP_BIND_ADDRESS_NO_PORT was
never set, and SO_REUSEPORT was always set.
Unlike $upstream_response_length that only counts the body size,
the new variable also counts the size of response header and data
received after switching protocols when proxying WebSockets.
The change in b91bcba29351 was not enough to fix random() seeding.
On Windows, the srand() seeds the PRNG only in the current thread,
and worse, is not inherited from the calling thread. Due to this,
worker threads were not properly seeded.
Reported by Marc Bevand.
If PCRE is disabled, captures were treated as normal variables in
ngx_http_script_compile(), while code calculating flushes array length in
ngx_http_compile_complex_value() did not account captures as variables.
This could lead to write outside of the array boundary when setting
last element to -1.
Found with AddressSanitizer.
It fixes potential connection leak if some unsent data was left in the SSL
buffer. Particularly, that could happen when a client canceled the stream
after the HEADERS frame has already been created. In this case no other
frames might be produced and the HEADERS frame alone didn't flush the buffer.
Checking for return value of c->send_chain() isn't sufficient since there
are data can be left in the SSL buffer. Now the wew->ready flag is used
instead.
In particular, this fixed a connection leak in cases when all streams were
closed, but there's still some data to be sent in the SSL buffer and the
client forgot about the connection.
Particularly this fixes alerts on OS X and NetBSD systems when HTTP/2 is
configured over plain TCP sockets.
On these systems calling writev() with no data leads to EINVAL errors
being logged as "writev() failed (22: Invalid argument) while processing
HTTP/2 connection".
Previously, if the worker process exited, GOAWAY was sent to connections in
idle state, but connections with active streams were closed without GOAWAY.
This flag appeared in Linux 4.5 and is useful for avoiding thundering herd
problem.
The current Linux kernel implementation walks the list of exclusive waiters,
and queues an event to each epfd, until it finds the first waiter that has
threads blocked on it via epoll_wait().
Now it is believed that the accept mutex brings more harm than benefits.
Especially in various benchmarks it often results in situation where only
one worker grabs all connections.
On non-aligned platforms, properly cast argument before left-shifting it in
ngx_http_v2_parse_uint32 that is used with u_char. Otherwise it propagates
to int to hold the value and can step over the sign bit. Usually, on known
compilers, this results in negation. Furthermore, a subsequent store into a
wider type, that is ngx_uint_t on 64-bit platforms, results in sign-extension.
In practice, this can be observed in debug log as a very large exclusive bit
value, when client sent PRIORITY frame with exclusive bit set:
: *14 http2 PRIORITY frame sid:5 on 1 excl:8589934591 weight:17
Found with UndefinedBehaviorSanitizer.
Previously, when a buffer was processed by the sub filter, its final bytes
could be buffered by the filter even if they don't match any pattern.
This happened because the Boyer-Moore algorithm, employed by the sub filter
since b9447fc457b4 (1.9.4), matches the last characters of patterns prior to
checking other characters. If the last character is out of scope, initial
bytes of a potential match are buffered until the last character is available.
Now, after receiving a flush or recycled buffer, the filter performs
additional checks to reduce the number of buffered bytes. The potential match
is checked against the initial parts of all patterns. Non-matching bytes are
not buffered. This improves processing of a chunked response from upstream
by sending the entire chunks without buffering unless a partial match is found
at the end of a chunk.
This reduces the number of moving parts in ABI compatibility checks.
Additionally, it also allows to use OpenSSL in FIPS mode while still
using md5 for non-security tasks.
The option is only set if the socket is bound to a specific port to allow
several such sockets coexist at the same time. This is required, for example,
when nginx acts as a transparent proxy and receives two datagrams from the same
client in a short time.
The feature is only implemented for Linux.
The following two types of bind addresses are supported in addition to
$remote_addr and address literals:
- $remote_addr:$remote_port
- [$remote_addr]:$remote_port
In both cases client remote address with port is used in upstream socket bind.
This patch moves various OpenSSL-specific function calls into the
OpenSSL module and introduces ngx_ssl_ciphers() to make nginx more
crypto-library-agnostic.
When the stream is terminated the HEADERS frame can still wait in the output
queue. This frame can't be removed and must be sent to the client anyway,
since HTTP/2 uses stateful compression for headers. So in order to postpone
closing and freeing memory of such stream the special close stream handler
is set to the write event. After the HEADERS frame is sent the write event
is called and the stream will be finally closed.
Some events like receiving a RST_STREAM can trigger the read handler of such
stream in closing state and cause unexpected processing that can result in
another attempt to finalize the request. To prevent it the read handler is
now set to ngx_http_empty_handler.
Thanks to Amazon.
There is no reason to add the "Content-Length: 0" header to a proxied request
without body if the header isn't presented in the original request.
Thanks to Amazon.
According to RFC 7540, an endpoint should not send more than one RST_STREAM
frame for any stream.
Also, now all the data frames will be skipped while termination.
The ngx_http_v2_finalize_connection() closes current stream, but that is an
invalid operation while processing unbuffered upload. This results in access
to already freed memory, since the upstream module sets a cleanup handler that
also finalizes the request.
A special last buffer with cl->buf->pos set to NULL can be present in
a chain when writing request body if chunked encoding was used. This
resulted in a NULL pointer dereference if it happened to be the only
buffer left after a do...while loop iteration in ngx_write_chain_to_file().
The problem originally appeared in nginx 1.3.9 with chunked encoding
support. Additionally, rev. 3832b608dc8d (nginx 1.9.13) changed the
minimum number of buffers to trigger this from IOV_MAX (typically 1024)
to NGX_IOVS_PREALLOCATE (typically 64).
Fix is to skip such buffers in ngx_chain_to_iovec(), much like it is
done in other places.
Previously, the stream's window was kept zero in order to prevent a client
from sending the request body before it was requested (see 887cca40ba6a for
details). Until such initial window was acknowledged all requests with
data were rejected (see 0aa07850922f for details).
That approach revealed a number of problems:
1. Some clients (notably MS IE/Edge, Safari, iOS applications) show an error
or even crash if a stream is rejected;
2. This requires at least one RTT for every request with body before the
client receives window update and able to send data.
To overcome these problems the new directive "http2_body_preread_size" is
introduced. It sets the initial window and configures a special per stream
preread buffer that is used to save all incoming data before the body is
requested and processed.
If the directive's value is lower than the default initial window (65535),
as previously, all streams with data will be rejected until the new window
is acknowledged. Otherwise, no special processing is used and all requests
with data are welcome right from the connection start.
The default value is chosen to be 64k, which is bigger than the default
initial window. Setting it to zero is fully complaint to the previous
behavior.
Now, the module extracts optional port which may accompany an
IP address. This custom extension is introduced, among other
things, in order to facilitate logging of original client ports.
Addresses with ports are expected to be in the RFC 3986 format,
that is, with IPv6 addresses in square brackets. E.g.,
"X-Real-IP: [2001:0db8::1]:12345" sets client port ($remote_port)
to 12345.
Previously, when the client address was changed to the one from
the PROXY protocol header, the client port ($remote_port) was
reset to zero. Now the client port is also changed to the one
from the PROXY protocol header.
The 6f8254ae61b8 change inadvertently fixed the duplicate port
detection similar to how it was fixed for mail in b2920b517490.
It also revealed another issue: the socket type (tcp vs. udp)
wasn't taken into account.
Since 4fbef397c753 nginx rejects with the 400 error any attempts of
requesting different host over the same connection, if the relevant
virtual server requires verification of a client certificate.
While requesting hosts other than negotiated isn't something legal
in HTTP/1.x, the HTTP/2 specification explicitly permits such requests
for connection reuse and has introduced a special response code 421.
According to RFC 7540 Section 9.1.2 this code can be sent by a server
that is not configured to produce responses for the combination of
scheme and authority that are included in the request URI. And the
client may retry the request over a different connection.
Now this code is used for requests that aren't authorized in current
connection. After receiving the 421 response a client will be able
to open a new connection, provide the required certificate and retry
the request.
Unfortunately, not all clients currently are able to handle it well.
Notably Chrome just shows an error, while at least the latest version
of Firefox retries the request over a new connection.
Using the same DH parameters on multiple servers is believed to be subject
to precomputation attacks, see http://weakdh.org/. Additionally, 1024 bits
are not enough in the modern world as well. Let users provide their own
DH parameters with the ssl_dhparam directive if they want to use EDH ciphers.
Note that SSL_CTX_set_dh_auto() as provided by OpenSSL 1.1.0 uses fixed
DH parameters from RFC 5114 and RFC 3526, and therefore subject to the same
precomputation attacks. We avoid using it as well.
This change also fixes compilation with OpenSSL 1.1.0-pre5 (aka Beta 2),
as OpenSSL developers changed their policy after releasing Beta 1 and
broke API once again by making the DH struct opaque (see ticket #860).
OpenSSL 1.0.2+ allows configuring a curve list instead of a single curve
previously supported. This allows use of different curves depending on
what client supports (as available via the elliptic_curves extension),
and also allows use of different curves in an ECDHE key exchange and
in the ECDSA certificate.
The special value "auto" was introduced (now the default for ssl_ecdh_curve),
which means "use an internal list of curves as available in the OpenSSL
library used". For versions prior to OpenSSL 1.0.2 it maps to "prime256v1"
as previously used. The default in 1.0.2b+ prefers prime256v1 as well
(and X25519 in OpenSSL 1.1.0+).
As client vs. server preference of curves is controlled by the
same option as used for ciphers (SSL_OP_CIPHER_SERVER_PREFERENCE),
the ssl_prefer_server_ciphers directive now controls both.
The SSL_CTX_add0_chain_cert() function as introduced in OpenSSL 1.0.2 now
used instead of SSL_CTX_add_extra_chain_cert().
SSL_CTX_add_extra_chain_cert() adds extra certs for all certificates
in the context, while SSL_CTX_add0_chain_cert() only to a particular
certificate. There is no difference unless multiple certificates are used,
though it is important when using multiple certificates.
Additionally, SSL_CTX_select_current_cert() is now called before using
a chain to make sure correct chain will be returned.
A pointer to a previously configured certificate now stored in a certificate.
This makes it possible to iterate though all certificates configured in
the SSL context. This is now used to configure OCSP stapling for all
certificates, and in ngx_ssl_session_id_context().
As SSL_CTX_use_certificate() frees previously loaded certificate of the same
type, and we have no way to find out if it's the case, X509_free() calls
are now posponed till ngx_ssl_cleanup_ctx().
Note that in OpenSSL 1.0.2+ this can be done without storing things in exdata
using the SSL_CTX_set_current_cert() and SSL_CTX_get0_certificate() functions.
These are not yet available in all supported versions though, so it's easier
to continue to use exdata for now.
This makes it possible to properly return OCSP staple with multiple
certificates configured.
Note that it only works properly in OpenSSL 1.0.1d+, 1.0.0k, 0.9.8y+.
In older versions SSL_get_certificate() fails to return correct certificate
when the certificate status callback is called.
Both minor and major versions are now limited to 999 maximum. In case of
r->http_minor, this limit is already implied by the code. Major version,
r->http_major, in theory can be up to 65535 with current code, but such
values are very unlikely to become real (and, additionally, such values
are not allowed by RFC 7230), so the same test was used for r->http_major.
When it's known that the kernel supports EPOLLRDHUP, there is no need in
additional recv() call to get EOF or error when the flag is absent in the
event generated by the kernel. A special runtime test is done at startup
to detect if EPOLLRDHUP is actually supported by the kernel because
epoll_ctl() silently ignores unknown flags.
With this knowledge it's now possible to drop the "ready" flag for partial
read. Previously, the "ready" flag was kept until the recv() returned EOF
or error. In particular, this change allows the lingering close heuristics
(which relies on the "ready" flag state) to actually work on Linux, and not
wait for more data in most cases.
The "available" flag is now used in the read event with the semantics similar
to the corresponding counter in kqueue.
This parameter lets binding the proxy connection to a non-local address.
Upstream will see the connection as coming from that address.
When used with $remote_addr, upstream will accept the connection from real
client address.
Example:
proxy_bind $remote_addr transparent;
Sergey Kandaurov
Maxim Dounin
Roman Arutyunyan
Yugo Horie
Valentin Bartenev
BullerDu
Ciel Zhao
Jiuzhou Cui
Murilo Andrade
Ruslan Ermilov
Gena Makhomed
Aleksei Bavshin
Vladimir Homutov
Marcus Ball
Pavel Pautov
Martin Duke
Mariano Di Martino
Alexey Radkov
Awdhesh Mathpal
Rob Mueller
Igor Ippolitov
balus
Eran Kornblau
Quantum
Jinhua Tan
Dmitry Volyntsev
Daniil Bondarev
Sergey A. Osokin
Ilya Leoshkevich
Niklas Keller
Nikolay Morozov
Chanhun Jeong
Maxim Konovalov
chronolaw
Nova DasSarma
Valentin Bartenev
Vadim Filimonov
Debayan Ghosh
Patryk Lesiewicz
hucongcong
Alessandro Ghedini
Piotr Sikora
Alex Zhang
Samuel Martin
Bart Warmerdam
Orgad Shaneh
Igor Sysoev
Simon Leblanc
othree
Thibault Charbonnier
Armin Grodon
Dmitry Lazurkin
Rob N ★
Konstantin Pavlov
Andrei Belov
Tim Taubert
Otto Kekäläinen