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>
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;
The WINDOW_UPDATE frame could be left in the output queue for an indefinite
period of time resulting in the request timeout.
This might happen if reading of the body was triggered by an event unrelated
to client connection, e.g. by the limit_req timer.
Particularly this prevents sending WINDOW_UPDATE with zero delta
which can result in PROTOCOL_ERROR.
Also removed surplus setting of no_flow_control to 0.
The ngx_thread_pool_done object isn't volatile, and at least some
compilers assume that it is permitted to reorder modifications of
volatile and non-volatile objects. Added appropriate ngx_memory_barrier()
calls to make sure all modifications will happen before the lock is released.
Reported by Mindaugas Rasiukevicius,
http://mailman.nginx.org/pipermail/nginx-devel/2016-April/008160.html.
Refusing streams is known to be incorrectly handled at least by IE, Edge
and Safari. Make sure to provide appropriate logging to simplify fixing
this in the affected browsers.
After the 92464ebace8e change, it has been discovered that not all
clients follow the RFC and handle RST_STREAM with NO_ERROR properly.
Notably, Chrome currently interprets it as INTERNAL_ERROR and discards
the response.
As a workaround, instead of RST_STREAM the maximum stream window update
will be sent, which will let client to send up to 2 GB of a request body
data before getting stuck on flow control. All the received data will
be silently discarded.
See for details:
http://mailman.nginx.org/pipermail/nginx-devel/2016-April/008143.htmlhttps://bugs.chromium.org/p/chromium/issues/detail?id=603182
A client is allowed to send requests before receiving and acknowledging
the SETTINGS frame. Such a client having a wrong idea about the stream's
could send the request body that nginx isn't ready to process.
The previous behavior was to send RST_STREAM with FLOW_CONTROL_ERROR in
such case, but it didn't allow retrying requests that have been rejected.
This prevents forming empty records out of such buffers. Particularly it fixes
double end-of-stream records with chunked transfer encoding, or when HTTP/2 is
used and the END_STREAM flag has been sent without data. In both cases there
is an empty buffer at the end of the request body chain with the "last_buf"
flag set.
The canonical libfcgi, as well as php implementation, tolerates such records,
while the HHVM parser is more strict and drops the connection (ticket #950).
There are two improvements:
1. Support for request body filters;
2. Receiving of request body is started only after
the ngx_http_read_client_request_body() call.
The last one fixes the problem when the client_max_body_size value might not be
respected from the right location if the location was changed either during the
process of receiving body or after the whole body had been received.
RFC 7540 states that "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)."
This should prevent a client from blocking on the stream window, since it isn't
maintained for closed streams. Currently, quite big initial stream windows are
used, so such blocking is very unlikly, but that will be changed in the further
patches.
SSLeay_version() and SSLeay() are no longer available if OPENSSL_API_COMPAT
is set to 0x10100000L. Switched to using OpenSSL_version() instead.
Additionally, we now compare version strings instead of version numbers,
and this correctly works for LibreSSL as well.
OPENSSL_config() deprecated in OpenSSL 1.1.0. Additionally,
SSL_library_init(), SSL_load_error_strings() and OpenSSL_add_all_algorithms()
are no longer available if OPENSSL_API_COMPAT is set to 0x10100000L.
The OPENSSL_init_ssl() function is now used instead with appropriate
arguments to trigger the same behaviour. The configure test changed to
use SSL_CTX_set_options().
Deinitialization now happens automatically in OPENSSL_cleanup() called
via atexit(3), so we no longer call EVP_cleanup() and ENGINE_cleanup()
directly.
LibreSSL defines OPENSSL_VERSION_NUMBER to 0x20000000L, but uses an old
API derived from OpenSSL at the time LibreSSL forked. As a result, every
version check we use to test for new API elements in newer OpenSSL versions
requires an explicit check for LibreSSL.
To reduce clutter, redefine OPENSSL_VERSION_NUMBER to 0x1000107fL if
LibreSSL is used. The same is done by FreeBSD port of LibreSSL.
Correct error code for NGX_EXDEV on Windows is ERROR_NOT_SAME_DEVICE,
"The system cannot move the file to a different disk drive".
Previously used ERROR_WRONG_DISK is about wrong diskette in the drive and
is not appropriate.
There is no real difference though, as MoveFile() is able to copy files
between disk drives, and will fail with ERROR_ACCESS_DENIED when asked
to copy directories. The ERROR_NOT_SAME_DEVICE error is only used
by MoveFileEx() when called without the MOVEFILE_COPY_ALLOWED flag.
On Windows there are two possible error codes which correspond to
the EEXIST error code: ERROR_FILE_EXISTS used by CreateFile(CREATE_NEW),
and ERROR_ALREADY_EXISTS used by CreateDirectory().
MoveFile() seems to use both: ERROR_ALREADY_EXISTS when moving within
one filesystem, and ERROR_FILE_EXISTS when copying a file to a different
drive.
By default, requests with non-idempotent methods (POST, LOCK, PATCH)
are no longer retried in case of errors if a request was already sent
to a backend. Previous behaviour can be restored by using
"proxy_next_upstream ... non_idempotent".
Much like normal connections, cached connections are now tested against
u->conf->next_upstream, and u->state->status is now always set.
This allows to disable additional tries even with upstream keepalive
by using "proxy_next_upstream off".
Fixes various aspects of --test-build-devpoll, --test-build-eventport, and
--test-build-epoll.
In particular, if --test-build-devpoll was used on Linux, then "devpoll"
event method would be preferred over "epoll". Also, wrong definitions of
event macros were chosen.
This fixes buffer over-read while using variables in the "proxy_pass",
"fastcgi_pass", "scgi_pass", and "uwsgi_pass" directives, where result
of string evaluation isn't null-terminated.
Found with MemorySanitizer.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
On nginx reload or binary upgrade, an attempt is made to inherit listen sockets
from the previous configuration. Previously, no check for socket type was made
and the inherited socket could have the wrong type. On binary upgrade, socket
type was not detected at all. Wrong socket type could lead to errors on that
socket due to different logic and unsupported syscalls. For example, a UDP
socket, inherited as TCP, lead to the following error after arrival of a
datagram: "accept() failed (102: Operation not supported on socket)".
It allows to turn off accumulation of small pool allocations into a big
preallocated chunk of memory. This is useful for debugging memory access
with sanitizer, since such accumulation can cover buffer overruns from
being detected.
This structure cannot be allocated as a large block anyway, otherwise that will
result in infinite recursion, since each large allocation requires to allocate
another ngx_pool_large_t.
The room for the structure is guaranteed by the NGX_MIN_POOL_SIZE constant.
When a keys_zone is full then each next request to the cache is
penalized. That is, the cache has to evict older files to get a
slot from the keys_zone synchronously. The patch introduces new
behavior in this scenario. Manager will try to maintain available
free slots in the keys_zone by cleaning old files in the background.
The "aio_write" directive is introduced, which enables use of aio
for writing. Currently it is meaningful only with "aio threads".
Note that aio operations can be done by both event pipe and output
chain, so proper mapping between r->aio and p->aio is provided when
calling ngx_event_pipe() and in output filter.
In collaboration with Valentin Bartenev.
The ngx_thread_write_chain_to_file() function introduced, which
uses ngx_file_t thread_handler, thread_ctx and thread_task fields.
The task context structure (ngx_thread_file_ctx_t) is the same for
both reading and writing, and can be safely shared as long as
operations are serialized.
The task->handler field is now always set (and not only when task is
allocated), as the same task can be used with different handlers.
The thread_write flag is introduced in the ngx_temp_file_t structure
to explicitly enable use of ngx_thread_write_chain_to_file() in
ngx_write_chain_to_temp_file() when supported by caller.
In collaboration with Valentin Bartenev.
This simplifies the interface of the ngx_thread_read() function.
Additionally, most of the thread operations now explicitly set
file->thread_task, file->thread_handler and file->thread_ctx,
to facilitate use of thread operations in other places.
(Potential problems remain with sendfile in threads though - it uses
file->thread_handler as set in ngx_output_chain(), and it should not
be overwritten to an incompatible one.)
In collaboration with Valentin Bartenev.
If a write event happens after sendfile() but before we've got the
sendfile results in the main thread, this write event will be ignored.
And if no more events will happen, the connection will hang.
Removing the events works in the simple cases, but not always, as
in some cases events are added back by an unrelated code. E.g.,
the upstream module adds write event in the ngx_http_upstream_init()
to track client aborts.
Fix is to use wev->complete instead. It is now set to 0 before
a sendfile() task is posted, and it is set to 1 once a write event
happens. If on completion of the sendfile() task wev->complete is 1,
we know that an event happened while we were executing sendfile(), and
the socket is still ready for writing even if sendfile() did not sent
all the data or returned EAGAIN.
While sendfilev() is documented to return -1 with EINVAL set
if the file was truncated, at least Solaris 11 silently returns 0,
and this results in CPU hog. Added a test to complain appropriately
if 0 is returned.
The main proxy function ngx_stream_proxy_process() can terminate the stream
session. The code, following it, should check its return code to make sure the
session still exists. This happens in client and upstream initialization
functions. Swapping ngx_stream_proxy_process() call with the code, that
follows it, leaves the same problem vice versa.
In future ngx_stream_proxy_process() will call ngx_stream_proxy_next_upstream()
making it too complicated to know if stream session still exists after this
call.
Now ngx_stream_proxy_process() is called from posted event handlers in both
places with no code following it. The posted event is automatically removed
once session is terminated.
It can now be set to "off" conditionally, e.g. using the map
directive.
An empty value will disable the emission of the Server: header
and the signature in error messages generated by nginx.
Any other value is treated as "on", meaning that full nginx
version is emitted in the Server: header and error messages
generated by nginx.
If proxy_cache is enabled, and proxy_no_cache tests true, it was previously
possible for the client connection to be closed after a 304. The fix is to
recheck r->header_only after the final cacheability is determined, and end the
request if no longer cacheable.
Example configuration:
proxy_cache foo;
proxy_cache_bypass 1;
proxy_no_cache 1;
If a client sends If-None-Match, and the upstream server returns 200 with a
matching ETag, no body should be returned to the client. At the start of
ngx_http_upstream_send_response proxy_no_cache is not yet tested, thus cacheable
is still 1 and downstream_error is set.
However, by the time the downstream_error check is done in process_request,
proxy_no_cache has been tested and cacheable is set to 0. The client connection
is then closed, regardless of keepalive.
If caching was used, "zero size buf in output" alerts might appear
in logs if a client prematurely closed connection. Alerts appeared
in the following situation:
- writing to client returned an error, so event pipe
drained all busy buffers leaving body output filters
in an invalid state;
- when upstream response was fully received,
ngx_http_upstream_finalize_request() tried to flush
all pending data.
Fix is to avoid flushing body if p->downstream_error is set.
Sendfile handlers (aio preload and thread handler) are called within
ctx->output_filter() in ngx_output_chain(), and hence ctx->aio cannot
be set directly in ngx_output_chain(). Meanwhile, it must be set to
make sure loop within ngx_output_chain() will be properly terminated.
There are no known cases that trigger the problem, though in theory
something like aio + sub filter (something that needs body in memory,
and can also free some memory buffers) + sendfile can result in
"task already active" and "second aio post" alerts.
The fix is to set ctx->aio in ngx_http_copy_aio_sendfile_preload()
and ngx_http_copy_thread_handler().
For consistency, ctx->aio is no longer set explicitly in
ngx_output_chain_copy_buf(), as it's now done in
ngx_http_copy_thread_handler().
If sendfile in threads is used, it is possible that multiple
subrequests will trigger multiple ngx_linux_sendfile_thread() calls,
as operations are only serialized in output chain based on r->aio,
that is, on subrequest level.
This resulted in "task #N already active" alerts, in particular, when
running proxy_store.t with "aio threads; sendfile on;".
Fix is to tolerate duplicate calls, with an additional safety check
that the file is the same as previously used.
The same problem also affects "aio on; sendfile on;" on FreeBSD
(previously known as "aio sendfile;"), where aio->preload_handler()
could be called multiple times due to similar reasons, resulting in
"second aio post" alerts. Fix is the same as well.
It is also believed that similar problems can arise if a filter
calls the next body filter multiple times for some reason. These are
mostly theoretical though.
Previously, there were only three timeouts used globally for the whole HTTP/2
connection:
1. Idle timeout for inactivity when there are no streams in processing
(the "http2_idle_timeout" directive);
2. Receive timeout for incomplete frames when there are no streams in
processing (the "http2_recv_timeout" directive);
3. Send timeout when there are frames waiting in the output queue
(the "send_timeout" directive on a server level).
Reaching one of these timeouts leads to HTTP/2 connection close.
This left a number of scenarios when a connection can get stuck without any
processing and timeouts:
1. A client has sent the headers block partially so nginx starts processing
a new stream but cannot continue without the rest of HEADERS and/or
CONTINUATION frames;
2. When nginx waits for the request body;
3. All streams are stuck on exhausted connection or stream windows.
The first idea that was rejected was to detect when the whole connection
gets stuck because of these situations and set the global receive timeout.
The disadvantage of such approach would be inconsistent behaviour in some
typical use cases. For example, if a user never replies to the browser's
question about where to save the downloaded file, the stream will be
eventually closed by a timeout. On the other hand, this will not happen
if there's some activity in other concurrent streams.
Now almost all the request timeouts work like in HTTP/1.x connections, so
the "client_header_timeout", "client_body_timeout", and "send_timeout" are
respected. These timeouts close the request.
The global timeouts work as before.
Previously, the c->write->delayed flag was abused to avoid setting timeouts on
stream events. Now, the "active" and "ready" flags are manipulated instead to
control the processing of individual streams.
This is required for implementing per request timeouts.
Previously, the temporary pool was used only during skipping of
headers and the request pool was used otherwise. That required
switching of pools if the request was closed while parsing.
It wasn't a problem since the request could be closed only after
the validation of the fully parsed header. With the per request
timeouts, the request can be closed at any moment, and switching
of pools in the middle of parsing header name or value becomes a
problem.
To overcome this, the temporary pool is now always created and
used. Special checks are added to keep it when either the stream
is being processed or until header block is fully parsed.
Since 667aaf61a778 (1.1.17) the ngx_http_parse_header_line() function can return
NGX_HTTP_PARSE_INVALID_HEADER when a header contains NUL character. In this
case the r->header_end pointer isn't properly initialized, but the log message
in ngx_http_process_request_headers() hasn't been adjusted. It used the pointer
in size calculation, which might result in up to 2k buffer over-read.
Found with afl-fuzz.
This fixes "called a function you should not call" and
"shutdown while in init" errors as observed with OpenSSL 1.0.2f
due to changes in how OpenSSL handles SSL_shutdown() during
SSL handshakes.
When the "pending" value is zero, the "buf" will be right shifted
by the width of its type, which results in undefined behavior.
Found by Coverity (CID 1352150).
Changes to NGX_MODULE_V1 and ngx_module_t in 85dea406e18f (1.9.11)
broke all modules written in C++, because ISO C++11 does not allow
conversion from string literal to char *.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
The auto/module script is extended to understand ngx_module_link=DYNAMIC.
When set, it links the module as a shared object rather than statically
into nginx binary. The module can later be loaded using the "load_module"
directive.
New auto/module parameter ngx_module_order allows to define module loading
order in complex cases. By default the order is set based on ngx_module_type.
3rd party modules can be compiled dynamically using the --add-dynamic-module
configure option, which will preset ngx_module_link to "DYNAMIC" before
calling the module config script.
Win32 support is rudimentary, and only works when using MinGW gcc (which
is able to handle exports/imports automatically).
In collaboration with Ruslan Ermilov.
Due to greater priority of the unary plus operator over the ternary operator
the expression didn't work as expected. That might result in one byte less
allocation than needed for the HEADERS frame buffer.
The previous code only parsed the first answer, without checking its
type, and required a compressed RR name.
The new code checks the RR type, supports responses with multiple
answers, and doesn't require the RR name to be compressed.
This has a side effect in limited support of CNAME. If a response
includes both CNAME and PTR RRs, like when recursion is enabled on
the server, PTR RR is handled.
Full CNAME support in PTR response is not implemented in this change.
Previously, a global server balancer was used to assign the next DNS server to
send a query to. That could lead to a non-uniform distribution of servers per
request. A request could be assigned to the same dead server several times in a
row and wait longer for a valid server or even time out without being processed.
Now each query is sent to all servers sequentially in a circle until a
response is received or timeout expires. Initial server for each request is
still globally balanced.
When several requests were waiting for a response, then after getting
a CNAME response only the last request's context had the name updated.
Contexts of other requests had the wrong name. This name was used by
ngx_resolve_name_done() to find the node to remove the request context
from. When the name was wrong, the request could not be properly
cancelled, its context was freed but stayed linked to the node's waiting
list. This happened e.g. when the first request was aborted or timed
out before the resolving completed. When it completed, this triggered
a use-after-free memory access by calling ctx->handler of already freed
request context. The bug manifests itself by
"could not cancel <name> resolving" alerts in error_log.
When a request was responded with a CNAME, the request context kept
the pointer to the original node's rn->u.cname. If the original node
expired before the resolving timed out or completed with an error,
this would trigger a use-after-free memory access via ctx->name in
ctx->handler().
The fix is to keep ctx->name unmodified. The name from context
is no longer used by ngx_resolve_name_done(). Instead, we now keep
the pointer to resolver node to which this request is linked.
Keeping the original name intact also improves logging.
When several requests were waiting for a response, then after getting
a CNAME response only the last request was properly processed, while
others were left waiting.
If one or more requests were waiting for a response, then after
getting a CNAME response, the timeout event on the first request
remained active, pointing to the wrong node with an empty
rn->waiting list, and that could cause either null pointer
dereference or use-after-free memory access if this timeout
expired.
If several requests were waiting for a response, and the first
request terminated (e.g., due to client closing a connection),
other requests were left without a timeout and could potentially
wait indefinitely.
This is fixed by introducing per-request independent timeouts.
This change also reverts 954867a2f0a6 and 5004210e8c78.
If enabled, workers are bound to available CPUs, each worker to once CPU
in order. If there are more workers than available CPUs, remaining are
bound in a loop, starting again from the first available CPU.
The optional mask parameter defines which CPUs are available for automatic
binding.
In collaboration with Vladimir Homutov.
With main request buffered, it's possible, that a slice subrequest will send
output before it. For example, while main request is waiting for aio read to
complete, a slice subrequest can start an aio operation as well. The order
in which aio callbacks are called is undetermined.
Skip SSL_CTX_set_tlsext_servername_callback in case of renegotiation.
Do nothing in SNI callback as in this case it will be supplied with
request in c->data which isn't expected and doesn't work this way.
This was broken by b40af2fd1c16 (1.9.6) with OpenSSL master branch and LibreSSL.
Splits a request into subrequests, each providing a specific range of response.
The variable "$slice_range" must be used to set subrequest range and proper
cache key. The directive "slice" sets slice size.
The following example splits requests into 1-megabyte cacheable subrequests.
server {
listen 8000;
location / {
slice 1m;
proxy_cache cache;
proxy_cache_key $uri$is_args$args$slice_range;
proxy_set_header Range $slice_range;
proxy_cache_valid 200 206 1h;
proxy_pass http://127.0.0.1:9000;
}
}
If an upstream with variables evaluated to address without a port,
then instead of a "no port in upstream" error an attempt was made
to connect() which failed with EADDRNOTAVAIL.
This fixes suboptimal behavior caused by surplus lseek() for sequential writes
on systems without pwrite(). A consecutive read after write might result in an
error on systems without pread() and pwrite().
Fortunately, at the moment there are no widely used systems without these
syscalls.
The HEADERS frame is always represented by more than one buffer since
b930e598a199, but the handling code hasn't been adjusted.
Only the first buffer of HEADERS frame was checked and if it had been
sent while others had not, the rest of the frame was dropped, resulting
in broken connection.
Before b930e598a199, the problem could only be seen in case of HEADERS
frame with CONTINUATION.
The r->invalid_header flag wasn't reset once an invalid header appeared in a
request, resulting in all subsequent headers in the request were also marked
as invalid.
The directive toggles conversion of HEAD to GET for cacheable proxy requests.
When disabled, $request_method must be added to cache key for consistency.
By default, HEAD is converted to GET as before.
OpenSSL doesn't check if the negotiated protocol has been announced.
As a result, the client might force using HTTP/2 even if it wasn't
enabled in configuration.
It caused inconsistency between setting "in_closed" flag and the moment when
the last DATA frame was actually read. As a result, the body buffer might not
be initialized properly in ngx_http_v2_init_request_body(), which led to a
segmentation fault in ngx_http_v2_state_read_data(). Also it might cause
start processing of incomplete body.
This issue could be triggered when the processing of a request was delayed,
e.g. in the limit_req or auth_request modules.
The code failed to ensure that "s" is within the buffer passed for
parsing when checking for "ms", and this resulted in unexpected errors when
parsing non-null-terminated strings with trailing "m". The bug manifested
itself when the expires directive was used with variables.
Found by Roman Arutyunyan.
Now it limits only the maximum length of literal string (either raw or
compressed) in HPACK request header fields. It's easier to understand
and to describe in the documentation.
Previous code has been based on assumption that the header block can only be
splitted at the borders of individual headers. That wasn't the case and might
result in emitting frames bigger than the frame size limit.
The current approach is to split header blocks by the frame size limit.
Previously, nginx worker would crash because of a double free
if client disconnected or timed out before sending all headers.
Found with afl-fuzz.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Previously, streams that were indirectly reprioritized (either because of
a new exclusive dependency on their parent or because of removal of their
parent from the dependency tree), didn't have their pointer to the parent
node updated.
This broke detection of circular dependencies and, as a result, nginx
worker would crash due to stack overflow whenever such dependency was
introduced.
Found with afl-fuzz.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
Per RFC7540, a stream cannot depend on itself.
Previously, this requirement was enforced on PRIORITY frames, but not on
HEADERS frames and due to the implementation details nginx worker would
crash (stack overflow) while opening self-dependent stream.
Found with afl-fuzz.
Signed-off-by: Piotr Sikora <piotrsikora@google.com>
As setitimer() isn't available on Windows, time wasn't updated at all
if timer_resolution was used with the select event method. Fix is
to ignore timer_resolution in such cases.
This context is needed for shared sessions cache to work in configurations
with multiple virtual servers sharing the same port. Unfortunately, OpenSSL
does not provide an API to access the session context, thus storing it
separately.
In collaboration with Vladimir Homutov.
If no space left in buffer after adding formatting symbols, error message
could be left without terminating null. The fix is to output message using
actual length.
The code for displaying version info and configuration info seemed to be
cluttering up the main function. I was finding it hard to read main. This
extracts out all of the logic for displaying version and configuration info
into its own function, thus making main easier to read.
RAND_pseudo_bytes() is deprecated in the OpenSSL master branch, so the only
use was changed to RAND_bytes(). Access to internal structures is no longer
possible, so now we don't try to set SSL3_FLAGS_NO_RENEGOTIATE_CIPHERS even
if it's defined.
Since an output buffer can only be used for either reading or sending, small
amounts of data left from the previous operation (due to some limits) must be
sent before nginx will be able to read further into the buffer. Using only
one output buffer can result in suboptimal behavior that manifests itself in
forming and sending too small chunks of data. This is particularly painful
with SPDY (or HTTP/2) where each such chunk needs to be prefixed with some
header.
The default flow-control window in HTTP/2 is 64k minus one bytes. With one
32k output buffer this results is one byte left after exhausting the window.
With two 32k buffers the data will be read into the second free buffer before
sending, thus the minimum output is increased to 32k + 1 bytes which is much
better.
A configuration like
server { server_name .foo^@; }
server { server_name .foo; }
resulted in a segmentation fault during construction of server names hash.
Reported by Markus Linnala.
Found with afl-fuzz.
A configuration with a named location inside a zero-length prefix
or regex location used to trigger a segmentation fault, as
ngx_http_core_location() failed to properly detect if a nested location
was created. Example configuration to reproduce the problem:
location "" {
location @foo {}
}
Fix is to not rely on a parent location name length, but rather check
command type we are currently parsing.
Identical fix is also applied to ngx_http_rewrite_if(), which used to
incorrectly assume the "if" directive is on server{} level in such
locations.
Reported by Markus Linnala.
Found with afl-fuzz.
This prevents a potential attack that discloses cached data if an attacker
will be able to craft a hash collision between some cache key the attacker
is allowed to access and another cache key with protected data.
See http://mailman.nginx.org/pipermail/nginx-devel/2015-September/007288.html.
Thanks to Gena Makhomed and Sergey Brester.
The value of NGX_ERROR, returned from filter handlers, was treated as a generic
upstream error and changed to NGX_HTTP_INTERNAL_SERVER_ERROR before calling
ngx_http_finalize_request(). This resulted in "header already sent" alert
if header was already sent in filter handlers.
The problem appeared in 54e9b83d00f0 (1.7.5).
This overflow has become possible after the change in 06e850859a26,
since concurrent subrequests are not limited now and each of them is
counted in r->main->count.
Resolved warnings about declarations that hide previous local declarations.
Warnings about WSASocketA() being deprecated resolved by explicit use of
WSASocketW() instead of WSASocket(). When compiling without IPv6 support,
WinSock deprecated warnings are disabled to allow use of gethostbyname().
The following configuration with alias, nested location and try_files
resulted in wrong file being used. Request "/foo/test.gif" tried to
use "/tmp//foo/test.gif" instead of "/tmp/test.gif":
location /foo/ {
alias /tmp/;
location ~ gif {
try_files $uri =405;
}
}
Additionally, rev. c985d90a8d1f introduced a regression if
the "/tmp//foo/test.gif" file was found (ticket #768). Resulting URI
was set to "gif?/foo/test.gif", as the code used clcf->name of current
location ("location ~ gif") instead of parent one ("location /foo/").
Fix is to use r->uri instead of clcf->name in all cases in the
ngx_http_core_try_files_phase() function. It is expected to be
already matched and identical to the clcf->name of the right
location.
If alias was used in a location given by a regular expression,
nginx used to do wrong thing in try_files if a location name (i.e.,
regular expression) was an exact prefix of URI. The following
configuration triggered a segmentation fault on a request to "/mail":
location ~ /mail {
alias /path/to/directory;
try_files $uri =404;
}
Reported by Per Hansson.
Iterating through all connections takes a lot of CPU time, especially
with large number of worker connections configured. As a result
nginx processes used to consume CPU time during graceful shutdown.
To mitigate this we now only do a full scan for idle connections when
shutdown signal is received.
Transitions of connections to idle ones are now expected to be
avoided if the ngx_exiting flag is set. The upstream keepalive module
was modified to follow this.
If nginx was used under OpenVZ and a container with nginx was suspended
and resumed, configuration tests started to fail because of EADDRINUSE
returned from listen() instead of bind():
# nginx -t
nginx: the configuration file /etc/nginx/nginx.conf syntax is ok
nginx: [emerg] listen() to 0.0.0.0:80, backlog 511 failed (98: Address already in use)
nginx: configuration file /etc/nginx/nginx.conf test failed
With this change EADDRINUSE errors returned by listen() are handled
similarly to errors returned by bind(), and configuration tests work
fine in the same environment:
# nginx -t
nginx: the configuration file /etc/nginx/nginx.conf syntax is ok
nginx: configuration file /etc/nginx/nginx.conf test is successful
More details about OpenVZ suspend/resume bug:
https://bugzilla.openvz.org/show_bug.cgi?id=2470
OCSP responses may contain no nextUpdate. As per RFC 6960, this means
that nextUpdate checks should be bypassed. Handle this gracefully by
using NGX_MAX_TIME_T_VALUE as "valid" in such a case.
The problem was introduced by 6893a1007a7c (1.9.2).
Reported by Matthew Baldwin.
Broken by 6893a1007a7c (1.9.2) during introduction of strict OCSP response
validity checks. As stapling file is expected to be returned unconditionally,
fix is to set its validity to the maximum supported time.
Reported by Faidon Liambotis.
Once upstream is connected, the upstream buffer is allocated. Previously, the
proxy module used the buffer allocation status to check if upstream is
connected. Now it's enough to check the flag.
If the -T option is passed, additionally to configuration test, configuration
files are output to stdout.
In the debug mode, configuration files are kept in memory and can be accessed
using a debugger.
The function is now called ngx_parse_http_time(), and can be used by
any code to parse HTTP-style date and time. In particular, it will be
used for OCSP stapling.
For compatibility, a macro to map ngx_http_parse_time() to the new name
provided for a while.
With this change it's no longer needed to pass -D_GNU_SOURCE manually,
and -D_FILE_OFFSET_BITS=64 is set to use 64-bit off_t.
Note that nginx currently fails to work properly with master process
enabled on GNU Hurd, as fcntl(F_SETOWN) returns EOPNOTSUPP for sockets
as of GNU Hurd 0.6. Additionally, our strerror() preloading doesn't
work well with GNU Hurd, as it uses large numbers for most errors.
When configured, an individual listen socket on a given address is
created for each worker process. This allows to reduce in-kernel lock
contention on configurations with high accept rates, resulting in better
performance. As of now it works on Linux and DragonFly BSD.
Note that on Linux incoming connection requests are currently tied up
to a specific listen socket, and if some sockets are closed, connection
requests will be reset, see https://lwn.net/Articles/542629/. With
nginx, this may happen if the number of worker processes is reduced.
There is no such problem on DragonFly BSD.
Based on previous work by Sepherosa Ziehau and Yingqi Lu.