Emeric Brun [Thu, 28 May 2020 09:13:15 +0000 (11:13 +0200)]
MEDIUM: ring: add server statement to forward messages from a ring
This patch adds new statement "server" into ring section, and the
related "timeout connect" and "timeout server".
server <name> <address> [param*]
Used to configure a syslog tcp server to forward messages from ring buffer.
This supports for all "server" parameters found in 5.2 paragraph.
Some of these parameters are irrelevant for "ring" sections.
timeout connect <timeout>
Set the maximum time to wait for a connection attempt to a server to succeed.
Arguments :
<timeout> is the timeout value specified in milliseconds by default, but
can be in any other unit if the number is suffixed by the unit,
as explained at the top of this document.
timeout server <timeout>
Set the maximum time for pending data staying into output buffer.
Arguments :
<timeout> is the timeout value specified in milliseconds by default, but
can be in any other unit if the number is suffixed by the unit,
as explained at the top of this document.
Example:
global
log ring@myring local7
ring myring
description "My local buffer"
format rfc3164
maxlen 1200
size 32764
timeout connect 5s
timeout server 10s
server mysyslogsrv 127.0.0.1:6514
Willy Tarreau [Sat, 30 May 2020 04:59:07 +0000 (06:59 +0200)]
SCRIPTS: publish-release: pass -n to gzip to remove timestamp
It just appeared that the tar.gz we put online are not reproducible
because a timestamp is put by default into the archive. Passing "-n"
to gzip is sufficient to remove this timestamp, so let's do it, and
also make the gzip command configurable for more flexibility. Now
issuing the commands multiple times finally results in the same
archives being produced.
This should be backported to supported stable branches.
Willy Tarreau [Fri, 29 May 2020 15:23:05 +0000 (17:23 +0200)]
CLEANUP: pools: use the regular lock for the flush operation on lockless pools
Commit 04f5fe87d3d introduced an rwlock in the pools to deal with the risk
that pool_flush() dereferences an area being freed, and commit 899fb8abdcd
turned it into a spinlock. The pools already contain a spinlock in case of
locked pools, so let's use the same and simplify the code by removing ifdefs.
At this point I'm really suspecting that if pool_flush() would instead
rely on __pool_get_first() to pick entries from the pool, the concurrency
problem could never happen since only one user would get a given entry at
once, thus it could not be freed by another user. It's not certain this
would be faster however because of the number of atomic ops to retrieve
one entry compared to a locked batch.
MEDIUM: ssl: use TLSv1.2 as the minimum default on bind lines
Since HAProxy 1.8, the TLS default minimum version was set to TLSv1.0 to
avoid using the deprecated SSLv3.0. Since then, the standard changed and
the recommended TLS version is now TLSv1.2.
This patch changes the minimum default version to TLSv1.2 on bind lines.
If you need to use prior TLS version, this is still possible by
using the ssl-min-ver keyword.
BUG/MEDIUM: checks: Don't add a tcpcheck ruleset twice in the shared tree
When a tcpcheck ruleset is created, it is automatically inserted in a global
tree. Unfortunately for applicative health checks (redis, mysql...), the created
ruleset is inserted a second time during the directive parsing. The leads to a
infinite loop when haproxy is stopped when we try to scan the tree to release
all tcpcheck rulesets.
Now, only the function responsible to create the tcpcheck ruleset insert it into
the tree.
Willy Tarreau [Thu, 28 May 2020 16:07:10 +0000 (18:07 +0200)]
BUG/MINOR: nameservers: fix error handling in parsing of resolv.conf
In issue #657, Coverity found a bug in the "nameserver" parser for the
resolv.conf when "parse-resolv-conf" is set. What happens is that if an
unparsable address appears on a "nameserver" line, it will destroy the
previously allocated pointer before reporting the warning, then the next
"nameserver" line will dereference it again and wlil cause a crash. If
the faulty nameserver is the last one, it will only be a memory leak.
Let's just make sure we preserve the pointer when handling the error.
The patch also fixes a typo in the warning.
The bug was introduced in 1.9 with commit 44e609bfa ("MINOR: dns:
Implement `parse-resolv-conf` directive") so the fix needs to be backported
up to 1.9 or 2.0.
MINOR: checks: I/O callback function only rely on the data layer wake callback
Most of code in event_srv_chk_io() function is inherited from the checks before
the recent refactoring. Now, it is enough to only call wake_srv_chk(). Since the
refactoring, the removed code is dead and never called. wake_srv_chk() may only
return 0 if tcpcheck_main() returns 0 and the check status is unknown
(CHK_RES_UNKNOWN). When this happens, nothing is performed in event_srv_chk_io().
BUG/MEDIUM: checks: Don't blindly subscribe for receive if waiting for connect
When an health check is waiting for a connection establishment, it subscribe for
receive or send events, depending on the next rule to be evaluated. For
subscription for send events, there is no problem. It works as expected. For
subscription for receive events, It only works for HTTP checks because the
underlying multiplexer takes care to do a receive before subscribing again,
updating the fd state. For TCP checks, the PT multiplexer only forwards
subscriptions at the transport layer. So the check itself is woken up. This
leads to a subscribe/notify loop waiting the connection establishment or a
timeout, uselessly eating CPU.
Thus, when a check is waiting for a connect, instead of blindly resubscribe for
receive events when it is woken up, we now try to receive data.
This patch should fix the issue #635. No backport needed.
HTTP_1XX, HTTP_3XX and HTTP_4XX message templates are no longer used. Only
HTTP_302 and HTTP_303 are used during configuration parsing by "errorloc" family
directives. So these templates are removed from the generic http code. And
HTTP_302 and HTTP_303 templates are moved as static strings in the function
parsing "errorloc" directives.
MINOR: http-rules: Use an action function to eval http-request auth rules
Now http-request auth rules are evaluated in a dedicated function and no longer
handled "in place" during the HTTP rules evaluation. Thus the action name
ACT_HTTP_REQ_AUTH is removed. In additionn, http_reply_40x_unauthorized() is
also removed. This part is now handled in the new action_ptr callback function.
MINOR: http-ana: Use proxy's error replies to emit 401/407 responses
There is no reason to not use proxy's error replies to emit 401/407
responses. The function http_reply_40x_unauthorized(), responsible to emit those
responses, is not really complex. It only adds a
WWW-Authenticate/Proxy-Authenticate header to a generic message.
So now, error replies can be defined for 401 and 407 status codes, using
errorfile or http-error directives. When an http-request auth rule is evaluated,
the corresponding error reply is used. For 401 responses, all occurrences of the
WWW-Authenticate header are removed and replaced by a new one with a basic
authentication challenge for the configured realm. For 407 responses, the same
is done on the Proxy-Authenticate header. If the error reply must not be
altered, "http-request return" rule must be used instead.
This reg-test checks that sending unique IDs via PPv2 works for servers
with the `alpn` option specified (issue #640). As a side effect it also
checks that PPv2 works with ALPN (issue #651).
It has been verified that the test fails without the following commits
applied and succeeds with them applied.
1f9a4ecea BUG/MEDIUM: backend: set the connection owner to the session when using alpn. 083fd42d5 BUG/MEDIUM: connection: Ignore PP2 unique ID for stream-less connections eb9ba3cb2 BUG/MINOR: connection: Always get the stream when available to send PP2 line
Without the first two commits HAProxy crashes during execution of the
test. Without the last commit the test will fail, because no unique ID
is received.
Willy Tarreau [Fri, 8 May 2020 06:38:24 +0000 (08:38 +0200)]
MEDIUM: pools: directly free objects when pools are too much crowded
During pool_free(), when the ->allocated value is 125% of needed_avg or
more, instead of putting the object back into the pool, it's immediately
freed using free(). By doing this we manage to significantly reduce the
amount of memory pinned in pools after transient traffic spikes.
During a test involving a constant load of 100 concurrent connections
each delivering 100 requests per second, the memory usage was a steady
21 MB RSS. Adding a 1 minute parallel load of 40k connections all looping
on 100kB objects made the memory usage climb to 938 MB before this patch.
With the patch it was only 660 MB. But when this parasit load stopped,
before the patch the RSS would remain at 938 MB while with the patch,
it went down to 480 then 180 MB after a few seconds, to stabilize around
69 MB after about 20 seconds.
This can be particularly important to improve reloads where the memory
has to be shared between the old and new process.
Another improvement would be welcome, we ought to have a periodic task
to check pools usage and continue to free up unused objects regardless
of any call to pool_free(), because the needed_avg value depends on the
past and will not cover recently refilled objects.
Willy Tarreau [Fri, 8 May 2020 06:31:56 +0000 (08:31 +0200)]
MINOR: pools: compute an estimate of each pool's average needed objects
This adds a sliding estimate of the pools' usage. The goal is to be able
to use this to start to more aggressively free memory instead of keeping
lots of unused objects in pools. The average is calculated as a sliding
average over the last 1024 consecutive measures of ->used during calls to
pool_free(), and is bumped up for 1/4 of its history from ->allocated when
allocation from the pool fails and results in a call to malloc().
The result is a floating value between ->used and ->allocated, that tries
to react fast to under-estimates that result in expensive malloc() but
still maintains itself well in case of stable usage, and progressively
goes down if usage shrinks over time.
This new metric is reported as "needed_avg" in "show pools".
Sadly due to yet another include dependency hell, we couldn't reuse the
functions from freq_ctr.h so they were temporarily duplicated into memory.h.
Olivier Houchard [Tue, 26 May 2020 23:26:06 +0000 (01:26 +0200)]
BUG/MEDIUM: backend: set the connection owner to the session when using alpn.
In connect_server(), if we can't create the mux immediately because we have
to wait until the alpn is negociated, store the session as the connection's
owner. conn_create_mux() expects it to be set, and provides it to the mux
init() method. Failure to do so will result to crashes later if the
connection is private, and even if we didn't do so it would prevent connection
reuse for private connections.
This should fix github issue #651.
BUG/MINOR: connection: Always get the stream when available to send PP2 line
When a PROXY protocol line must be sent, it is important to always get the
stream if it exists. It is mandatory to send an unique ID when the unique-id
option is enabled. In conn_si_send_proxy(), to get the stream, we first retrieve
the conn-stream attached to the backend connection. Then if the conn-stream data
callback is si_conn_cb, it is possible to get the stream. But for now, it only
works for connections with a multiplexer. Thus, for mux-less connections, the
unique ID is never sent. This happens for all SSL connections relying on the
alpn to choose the good multiplexer. But it is possible to use the context of
such connections to get the conn-stream.
The bug was introduced by the commit cf6e0c8a8 ("MEDIUM: proxy_protocol: Support
sending unique IDs using PPv2"). Thus, this patch must be backported to the same
versions as the commit above.
BUG/MEDIUM: connection: Ignore PP2 unique ID for stream-less connections
It is possible to send a unique ID when the PROXY protocol v2 is used. It relies
on the stream to do so. So we must be sure to have a stream. Locally initiated
connections may not be linked to a stream. For instance, outgoing connections
created by health checks have no stream. Moreover, the stream is not retrieved
for mux-less connections (this bug will be fixed in another commit).
Unfortunately, in make_proxy_line_v2() function, the stream is not tested before
generating the unique-id. This bug leads to a segfault when a health check is
performed for a server with the PROXY protocol v2 and the unique-id option
enabled. It also crashes for servers using SSL connections with alpn. The bug
was introduced by the commit cf6e0c8a8 ("MEDIUM: proxy_protocol: Support sending
unique IDs using PPv2")
This patch should fix the issue #640. It must be backported to the same versions
as the commit above.
Bertrand Jacquin [Fri, 22 May 2020 16:03:46 +0000 (17:03 +0100)]
BUG/MEDIUM: contrib/spoa: do not register python3.8 if --embed fail
spoa server fails to build when python3.8 is not available. If
python3-config --embed fails, the output of the command is registered in
check_python_config. However when it's later used to define
PYTHON_DEFAULT_INC and PYTHON_DEFAULT_LIB it's content does not match
and fallback to python2.7
Content of check_python_config when building with python3.6:
MINOR: checks: Remove useless tests on the connection and conn-stream
During an health check execution, the conn-stream and the conncetion may only be
NULL before the evaluation of the first rule, which is always a connect, or if
the first conn-stream allocation failed. Thus, in tcpcheck_main(), useless tests
on the conn-stream or on the connection have been removed. A comment has been
added to make it clear.
BUG/MEDIUM: checks: Refresh the conn-stream and the connection after a connect
When a connect rule is evaluated, the conn-stream and the connection must be
refreshed in tcpcheck_main(). Otherwise, in case of synchronous connect, these
variables point on outdated values (NULL for the first connect or released
otherwise).
Emeric Brun [Mon, 25 May 2020 13:01:04 +0000 (15:01 +0200)]
MEDIUM: ring: new section ring to declare custom ring buffers.
It is possible to globally declare ring-buffers, to be used as target for log
servers or traces.
ring <ringname>
Creates a new ring-buffer with name <ringname>.
description <text>
The descritpition is an optional description string of the ring. It will
appear on CLI. By default, <name> is reused to fill this field.
format <format>
Format used to store events into the ring buffer.
Arguments:
<format> is the log format used when generating syslog messages. It may be
one of the following :
iso A message containing only the ISO date, followed by the text.
The PID, process name and system name are omitted. This is
designed to be used with a local log server.
raw A message containing only the text. The level, PID, date, time,
process name and system name are omitted. This is designed to be
used in containers or during development, where the severity
only depends on the file descriptor used (stdout/stderr). This
is the default.
rfc3164 The RFC3164 syslog message format. This is the default.
(https://tools.ietf.org/html/rfc3164)
rfc5424 The RFC5424 syslog message format.
(https://tools.ietf.org/html/rfc5424)
short A message containing only a level between angle brackets such as
'<3>', followed by the text. The PID, date, time, process name
and system name are omitted. This is designed to be used with a
local log server. This format is compatible with what the systemd
logger consumes.
timed A message containing only a level between angle brackets such as
'<3>', followed by ISO date and by the text. The PID, process
name and system name are omitted. This is designed to be
used with a local log server.
maxlen <length>
The maximum length of an event message stored into the ring,
including formatted header. If an event message is longer than
<length>, it will be truncated to this length.
size <size>
This is the optional size in bytes for the ring-buffer. Default value is
set to BUFSIZE.
Example:
global
log ring@myring local7
ring myring
description "My local buffer"
format rfc3164
maxlen 1200
Note: ring names are resolved during post configuration processing.
Tim Duesterhus [Tue, 19 May 2020 11:49:42 +0000 (13:49 +0200)]
MEDIUM: lua: Add `ifexist` parameter to `set_var`
As discussed in GitHub issue #624 Lua scripts should not use
variables that are never going to be read, because the memory
for variable names is never going to be freed.
Add an optional `ifexist` parameter to the `set_var` function
that allows a Lua developer to set variables that are going to
be ignored if the variable name was not used elsewhere before.
Usually this mean that there is no `var()` sample fetch for the
variable in question within the configuration.
With "MINOR: lua: Use vars_unset_by_name_ifexist()" the last user was
removed and as outlined in that commit there is no good reason for this
function to exist.
May be backported together with the commit mentioned above.
REGTESTS: checks: Fix tls_health_checks when IPv6 addresses are used
In tls_health_checks.vtc, when IPv6 addresses are used, A config error is
reported because of the "addr" server parameter. Because there is no specified
port, the IPv6 address must be enclosed into brackets to be properly parsed. It
also works with IPv4 addresses. But instead, a dummy port is added to the addr
parameter. This way, we also check the port parameter, when specified, is used
in priority over the port found in the addr parameter.
MINOR: checks: Remove dead code from process_chk_conn()
With the checks refactoring, all connections are performed into a tcp-check
ruleset. So, in process_chk_conn(), it is no longer required to check
synchronous error on the connect when an health check is started. This part is
now handled in tcpcheck_main(). And because it is impossible to start a health
check with a connection, all tests on the connection during the health check
startup can be safely removed.
This patch should fix the issue #636. No backport needed.
BUG/MINOR: http-htx: Don't forget to release the http reply in release function
The http reply must be released in the function responsible to release it. This
leak was introduced when the http return was refactored to use http reply.
Willy Tarreau [Fri, 22 May 2020 14:19:04 +0000 (16:19 +0200)]
[RELEASE] Released version 2.2-dev8
Released version 2.2-dev8 with the following main changes :
- MINOR: checks: Improve report of unexpected errors for expect rules
- MEDIUM: checks: Add matching on log-format string for expect rules
- DOC: Fix req.body and co documentation to be accurate
- MEDIUM: checks: Remove dedicated sample fetches and use response ones instead
- CLEANUP: checks: sort and rename tcpcheck_expect_type types
- MINOR: checks: Use dedicated actions to send log-format strings in send rules
- MINOR: checks: Simplify matching on HTTP headers in HTTP expect rules
- MINOR: checks/sample: Remove unnecessary tests on the sample session
- REGTEST: checks: Adapt SSL error message reported when connection is rejected
- MINOR: mworker: replace ha_alert by ha_warning when exiting successfuly
- MINOR: checks: Support log-format string to set the URI for HTTP send rules
- MINOR: checks: Support log-format string to set the body for HTTP send rules
- DOC: Be more explicit about configurable check ok/error/timeout status
- MINOR: checks: Make matching on HTTP headers for expect rules less obscure
- BUG/MEDIUM: lua: Fix dumping of stick table entries for STD_T_DICT
- BUG/MINOR: config: Make use_backend and use-server post-parsing less obscur
- REGTESTS: make the http-check-send test require version 2.2
- BUG/MINOR: http-ana: fix NTLM response parsing again
- BUG/MEDIUM: http_ana: make the detection of NTLM variants safer
- BUG/MINOR: cfgparse: Abort parsing the current line if an invalid \x sequence is encountered
- MINOR: cfgparse: Improve error message for invalid \x sequences
- CI: travis-ci: enable arm64 builds again
- MEDIUM: ssl: increase default-dh-param to 2048
- CI: travis-ci: skip pcre2 on arm64 build
- CI: travis-ci: extend the build time for SSL to 60 minutes
- CLEANUP: config: drop unused setting CONFIG_HAP_MEM_OPTIM
- CLEANUP: config: drop unused setting CONFIG_HAP_INLINE_FD_SET
- CLENAUP: config: move CONFIG_HAP_LOCKLESS_POOLS out of config.h
- CLEANUP: remove THREAD_LOCAL from config.h
- CI: travis-ci: upgrade LibreSSL versions
- DOC: assorted typo fixes in the documentation
- CI: extend spellchecker whitelist
- CLEANUP: assorted typo fixes in the code and comments
- MAJOR: contrib: porting spoa_server to support python3
- BUG/MEDIUM: checks: Subscribe to I/O events on an unfinished connect
- BUG/MINOR: checks: Don't subscribe to I/O events if it is already done
- BUG/MINOR: checks: Rely on next I/O oriented rule when waiting for a connection
- MINOR: checks: Don't try to send outgoing data if waiting to be able to send
- MINOR: sample: Move aes_gcm_dec implementation into sample.c
- MINOR: sample: Add digest and hmac converters
- BUG/MEDIUM: checks: Subscribe to I/O events only if a mux was installed
- BUG/MINOR: sample/ssl: Fix digest converter for openssl < 1.1.0
- BUG/MINOR: pools: use %u not %d to report pool stats in "show pools"
- BUG/MINOR: pollers: remove uneeded free in global init
- CLEANUP: select: enhance readability in init
- BUG/MINOR: soft-stop: always wake up waiting threads on stopping
- MINOR: soft-stop: let the first stopper only signal other threads
- BUILD: select: only declare existing local labels to appease clang
- BUG/MEDIUM: streams: Remove SF_ADDR_SET if we're retrying due to L7 retry.
- BUG/MEDIUM: stream: Only allow L7 retries when using HTTP.
- DOC: retry-on can only be used with mode http
- MEDIUM: ssl: allow to register callbacks for SSL/TLS protocol messages
- MEDIUM: ssl: split ssl_sock_msgcbk() and use a new callback mechanism
- MINOR: ssl: add a new function ssl_sock_get_ssl_object()
- MEDIUM: ssl: use ssl_sock_get_ssl_object() in fetchers where appropriate
- REORG: ssl: move macros and structure definitions to ssl_sock.h
- CLEANUP: ssl: remove the shsess_* macros
- REORG: move the crt-list structures in their own .h
- REORG: ssl: move the ckch structures to types/ssl_ckch.h
- CLEANUP: ssl: add ckch prototypes in proto/ssl_ckch.h
- REORG: ssl: move crtlist functions to src/ssl_crtlist.c
- CLEANUP: ssl: avoid circular dependencies in ssl_crtlist.h
- REORG: ssl: move the ckch_store related functions to src/ssl_ckch.c
- REORG: ssl: move ckch_inst functions to src/ssl_ckch.c
- REORG: ssl: move the crt-list CLI functions in src/ssl_crtlist.c
- REORG: ssl: move the CLI 'cert' functions to src/ssl_ckch.c
- REORG: ssl: move ssl configuration to cfgparse-ssl.c
- MINOR: ssl: remove static keyword in some SSL utility functions
- REORG: ssl: move ssl_sock_ctx and fix cross-dependencies issues
- REORG: ssl: move sample fetches to src/ssl_sample.c
- REORG: ssl: move utility functions to src/ssl_utils.c
- DOC: ssl: update MAINTAINERS file
- CI: travis-ci: switch arm64 builds to use openssl from distro
- MINOR: stats: Prepare for more accurate moving averages
- MINOR: stats: Expose native cum_req metric for a server
- MEDIUM: stats: Enable more accurate moving average calculation for stats
- BUILD: ssl: include buffer common headers for ssl_sock_ctx
- BUILD: ssl: include errno.h in ssl_crtlist.c
- CLEANUP: acl: remove unused assignment
- DOC/MINOR: halog: Add long help info for ic flag
- BUILD: ssl: fix build without OPENSSL_NO_ENGINE
- DOC: SPOE is no longer experimental
- BUG/MINOR: cache: Don't needlessly test "cache" keyword in parse_cache_flt()
- MINOR: config: Don't dump keywords if argument is NULL
- MEDIUM: checks: Make post-41 the default mode for mysql checks
- BUG/MINOR: logs: prevent double line returns in some events.
- MEDIUM: sink: build header in sink_write for log formats
- MEDIUM: logs: buffer targets now rely on new sink_write
- MEDIUM: sink: add global statement to create a new ring (sink buffer)
- MEDIUM: hpack: use a pool for the hpack table
- BUG/MAJOR: mux-fcgi: Stop sending loop if FCGI stream is blocked for any reason
- BUG/MEDIUM: ring: write-lock the ring while attaching/detaching
- MINOR: applet: adopt the wait list entry from the CLI
- MINOR: ring: make the applet code not depend on the CLI
- Revert "MEDIUM: sink: add global statement to create a new ring (sink buffer)"
- CI: travis-ci: fix libslz download URL
- MINOR: ssl: split config and runtime variable for ssl-{min,max}-ver
- CLEANUP: http_ana: Remove unused TXN flags
- BUG/MINOR: http-rules: Mark http return rules as final
- MINOR: http-htx: Add http_reply type based on what is used for http return rules
- CLEANUP: http-htx: Rename http_error structure into http_error_msg
- MINOR: http-rules: Use http_reply structure for http return rules
- MINOR: http-htx: Use a dedicated function to release http_reply objects
- MINOR: http-htx: Use a dedicated function to parse http reply arguments
- MINOR: http-htx: Use a dedicated function to check http reply validity
- MINOR: http-ana: Use a dedicated function to send a response from an http reply
- MEDIUM: http-rules: Rely on http reply for http deny/tarpit rules
- MINOR: http-htx: Store default error messages in a global http reply array
- MINOR: http-htx: Store messages of an http-errors section in a http reply array
- MINOR: http-htx: Store errorloc/errorfile messages in http replies
- MINOR: proxy: Add references on http replies for proxy error messages
- MINOR: http-htx: Use http reply from the http-errors section
- MINOR: http-ana: Use a TXN flag to prevent after-response ruleset evaluation
- MEDIUM: http-ana: Use http replies for HTTP error messages
- CLEANUP: http-htx: Remove unused storage of error messages in buffers
- MINOR: htx: Add a function to copy a buffer in an HTX message
- CLEANUP: channel: Remove channel_htx_copy_msg() function
- MINOR: http-ana: Add a function to write an http reply in an HTX message
- MINOR: http-htx/proxy: Add http-error directive using http return syntax
- DOC: Fix "errorfile" description in the configuration manual
- BUG/MINOR: checks: Respect check-ssl param when a port or an addr is specified
- BUILD: hpack: make sure the hpack table can still be built standalone
- CONTRIB: hpack: make use of the simplified standalone HPACK API
- MINOR: connection: add pp2-never-send-local to support old PP2 behavior
Willy Tarreau [Fri, 22 May 2020 11:53:29 +0000 (13:53 +0200)]
MINOR: connection: add pp2-never-send-local to support old PP2 behavior
A bug in the PROXY protocol v2 implementation was present in HAProxy up to
version 2.1, causing it to emit a PROXY command instead of a LOCAL command
for health checks. This is particularly minor but confuses some servers'
logs. Sadly, the bug was discovered very late and revealed that some servers
which possibly only tested their PROXY protocol implementation against
HAProxy fail to properly handle the LOCAL command, and permanently remain in
the "down" state when HAProxy checks them. When this happens, it is possible
to enable this global option to revert to the older (bogus) behavior for the
time it takes to contact the affected components' vendors and get them fixed.
This option is disabled by default and acts on all servers having the
"send-proxy-v2" statement.
Older versions were reverted to the old behavior and should not attempt to
be fixed by default again. However a variant of this patch could possibly
be implemented to ask to explicitly send LOCAL if needed by some servers.
More context here:
https://www.mail-archive.com/haproxy@formilux.org/msg36890.html
https://www.mail-archive.com/haproxy@formilux.org/msg37218.html
Willy Tarreau [Fri, 22 May 2020 10:09:16 +0000 (12:09 +0200)]
CONTRIB: hpack: make use of the simplified standalone HPACK API
As reported in oss-fuzz issue 22388, the contrib/hpack code doesn't
build anymore for now. Thanks to previous patch we can now just define
HPACK_STANDALONE in the few files which include hpack-tbl.h and we can
declare a dummy pool to hold the DHT size. This is enough to make the
code work again outside of the haproxy process. It was successfully
tested using:
Willy Tarreau [Fri, 22 May 2020 10:05:27 +0000 (12:05 +0200)]
BUILD: hpack: make sure the hpack table can still be built standalone
Recent commit 2bdcc70fa7 ("MEDIUM: hpack: use a pool for the hpack table")
made the hpack code finally use a pool with very unintrusive code that was
assumed to be trivial enough to adjust if the code needed to be reused
outside of haproxy. Unfortunately the code in contrib/hpack already uses
it and broke the oss-fuzz tests as it doesn't build anymore.
This patch adds an HPACK_STANDALONE macro to decide if we should use the
pools or malloc+free. The resulting macros are called hpack_alloc() and
hpack_free() respectively, and the size must be passed into the pool
itself.
BUG/MINOR: checks: Respect check-ssl param when a port or an addr is specified
When a check port or a check address is specified, the check transport layer is
ignored. So it is impossible to do a SSL check in this case. This bug was
introduced by the commit 8892e5d30 ("BUG/MEDIUM: server/checks: Init server
check during config validity check").
This patch should fix the issue #643. It must be backported to all branches
where the above commit was backported.
DOC: Fix "errorfile" description in the configuration manual
"errorfile" description is outdated. Now, error messages are parsed at startup
and converted to HTX messages. So they must be valid according to the HTTP
standards. In addition, there is now a comment about the reserved buffer space
that should remain available to allow header rewrites.
MINOR: http-htx/proxy: Add http-error directive using http return syntax
The http-error directive can now be used instead of errorfile to define an error
message in a proxy section (including default sections). This directive uses the
same syntax that http return rules. The only real difference is the limitation
on status code that may be specified. Only status codes supported by errorfile
directives are supported for this new directive. Parsing of errorfile directive
remains independent from http-error parsing. But functionally, it may be
expressed in terms of http-errors :
errorfile <status> <file> ==> http-errror status <status> errorfile <file>
MINOR: http-ana: Add a function to write an http reply in an HTX message
When an error response is sent to a client, the write of the http reply in the
channel buffer and its sending are performed in different functions. The
http_reply_to_htx() function is used to write an http reply in HTX message. This
way, it could be possible to use the http replies in a different context.
MINOR: htx: Add a function to copy a buffer in an HTX message
The htx_copy_msg() function can now be used to copy the HTX message stored in a
buffer in an existing HTX message. It takes care to not overwrite existing
data. If the destination message is empty, a raw copy is performed. All the
message is copied or nothing.
This function is used instead of channel_htx_copy_msg().
MEDIUM: http-ana: Use http replies for HTTP error messages
When HAProxy returns an http error message, the corresponding http reply is now
used instead of the buffer containing the corresponding HTX message. So,
http_error_message() function now returns the http reply to use for a given
stream. And the http_reply_and_close() function now relies on
http_reply_message() to send the response to the client.
MINOR: http-ana: Use a TXN flag to prevent after-response ruleset evaluation
The txn flag TX_CONST_REPLY may now be used to prevent after-response ruleset
evaluation. It is used if this ruleset evaluation failed on an internal error
response. Before, it was done incrementing the parameter <final>. But it is not
really convenient if an intermediary function is used to produce the
response. Using a txn flag could also be a good way to prevent after-response
ruleset evaluation in a different context.
MINOR: http-htx: Use http reply from the http-errors section
When an http reply is configured to use an error message from an http-errors
section, instead of referencing the error message, the http reply is used. To do
so the new http reply type HTTP_REPLY_INDIRECT has been added.
MINOR: proxy: Add references on http replies for proxy error messages
Error messages defined in proxy section or inherited from a default section are
now also referenced using an array of http replies. This is done during the
configuration validity check.
MINOR: http-htx: Store errorloc/errorfile messages in http replies
During configuration parsing, error messages resulting of parsing of errorloc
and errorfile directives are now also stored as an http reply. So, for now,
these messages are stored as a buffer and as an http reply. To be able to
release all these http replies when haproxy is stopped, a global list is
used. We must do that because the same http reply may be referenced several
times by different proxies if it is defined in a default section.
MINOR: http-htx: Store messages of an http-errors section in a http reply array
Error messages specified in an http-errors section is now also stored in an
array of http replies. So, for now, these messages are stored as a buffer and as
a http reply.
MINOR: http-htx: Store default error messages in a global http reply array
Default error messages are stored as a buffer, in http_err_chunks global array.
Now, they are also stored as a http reply, in http_err_replies global array.
MEDIUM: http-rules: Rely on http reply for http deny/tarpit rules
"http-request deny", "http-request tarpit" and "http-response deny" rules now
use the same syntax than http return rules and internally rely on the http
replies. The behaviour is not the same when no argument is specified (or only
the status code). For http replies, a dummy response is produced, with no
payload. For old deny/tarpit rules, the proxy's error messages are used. Thus,
to be compatible with existing configuration, the "default-errorfiles" parameter
is implied. For instance :
MINOR: http-ana: Use a dedicated function to send a response from an http reply
The http_reply_message() function may be used to send an http reply to a
client. This function is responsile to convert the reply in HTX, to push it in
the response buffer and to forward it to the client. It is also responsible to
terminate the transaction.
This function is used during evaluation of http return rules.
MINOR: http-htx: Use a dedicated function to check http reply validity
A dedicated function is added to check the validity of an http reply object,
after parsing. It is used to check the validity of http return rules.
For now, this function is only used to find the right error message in an
http-errors section for http replies of type HTTP_REPLY_ERRFILES (using
"errorfiles" argument). On success, such replies are updated to point on the
corresponding error message and their type is set to HTTP_REPLY_ERRMSG. If an
unknown http-errors section is referenced, anx error is returned. If a unknown
error message is referenced inside an existing http-errors section, a warning is
emitted and the proxy's error messages are used instead.
MINOR: http-htx: Use a dedicated function to parse http reply arguments
A dedicated function to parse arguments and create an http_reply object is
added. It is used to parse http return rule. Thus, following arguments are
parsed by this function :
MINOR: http-rules: Use http_reply structure for http return rules
No real change here. Instead of using an internal structure to the action rule,
the http return rules are now stored as an http reply. The main change is about
the action type. It is now always set to ACT_CUSTOM. The http reply type is used
to know how to evaluate the rule.
CLEANUP: http-htx: Rename http_error structure into http_error_msg
The structure owns an error message, most of time loaded from a file, and
converted to HTX. It is created when an errorfile or errorloc directive is
parsed. It is renamed to avoid ambiguities with http_reply structure.
MINOR: http-htx: Add http_reply type based on what is used for http return rules
The http_reply structure is added. It represents a generic HTTP message used as
internal response by HAProxy. It is based on the structure used to store http
return rules. The aim is to store all error messages using this structure, as
well as http return and http deny rules.
BUG/MINOR: http-rules: Mark http return rules as final
For HTTP rules, this flag is only used to trigger a warning during HAProxy
startup when a final rule without ACL is not the last one. So this patch is
marked as a bug, but its impact is really limited.
No backport needed because http return rules were introduced in the 2.2.
TX_CLDENY, TX_CLALLOW, TX_SVDENY and TX_SVALLOW flags are unused. Only
TX_CLTARPIT is used to make the difference between an http deny rule and an http
tarpit rule. So these unused flags are removed.
MINOR: ssl: split config and runtime variable for ssl-{min,max}-ver
In the CLI command 'show ssl crt-list', the ssl-min-ver and the
ssl-min-max arguments were always displayed because the dumped versions
were the actual version computed and used by haproxy, instead of the
version found in the configuration.
To fix the problem, this patch separates the variables to have one with
the configured version, and one with the actual version used. The dump
only shows the configured version.
Ilya Shipitsin [Mon, 18 May 2020 20:47:07 +0000 (01:47 +0500)]
CI: travis-ci: fix libslz download URL
let us switch to github mirror, travis cannot download original libslz
error is:
fatal: unable to access 'http://git.1wt.eu/git/libslz.git/': Failed
to connect to git.1wt.eu port 80: Connection timed out The command
"git clone http://git.1wt.eu/git/libslz.git/" failed and exited with
128...
Willy Tarreau [Tue, 19 May 2020 17:14:42 +0000 (19:14 +0200)]
MINOR: ring: make the applet code not depend on the CLI
The ring to applet communication was only made to deal with CLI functions
but it's generic. Let's have generic appctx functions and have the CLI
rely on these instead. This patch introduces ring_attach_appctx() and
ring_detach_appctx().
Willy Tarreau [Tue, 19 May 2020 15:07:30 +0000 (17:07 +0200)]
MINOR: applet: adopt the wait list entry from the CLI
A few fields, including a generic list entry, were added to the CLI context
by commit 300decc8d9 ("MINOR: cli: extend the CLI context with a list and
two offsets"). It turns out that the list entry (l0) is solely used to
consult rings and that the generic ring_write() code is restricted to a
consumer on the CLI due to this, which was not the initial intent. Let's
make it a general purpose wait_entry field that is properly initialized
during appctx_init(). This will allow any applet to wait on a ring, not
just the CLI.
Willy Tarreau [Tue, 19 May 2020 17:21:45 +0000 (19:21 +0200)]
BUG/MEDIUM: ring: write-lock the ring while attaching/detaching
The LIST_ADDQ() and LIST_DEL_INIT() calls made to attach/detach a waiter
to the ring were made under a read lock which was sufficient in front of
the writer's write lock. But it's not sufficient against other readers!
Thus theorically multiple "show events" on the same ring buffer on the
CLI could result in a crash, even though for now I couldn't manage to
reproduce it.
This fixes commit 1d181e489c ("MEDIUM: ring: implement a wait mode for
watchers") so it must be backported to 2.1, possibly further if the ring
code gets backported.
BUG/MAJOR: mux-fcgi: Stop sending loop if FCGI stream is blocked for any reason
Because of a typo error in conditions to exit the sending loop, it is possible
to loop infinitely in fcgi_snd_buf() function. Instead of checking the FCGI
stream is not blocked to continue sending data, the FCGI connection is used. So
it is possible to have a stream blocked because there is not enough space in the
mux buffers to copy more data but continue to loop to send more data.
This patch should fix the issue #637. It must be backported to 2.1.
Willy Tarreau [Tue, 19 May 2020 09:31:11 +0000 (11:31 +0200)]
MEDIUM: hpack: use a pool for the hpack table
Instead of using malloc/free to allocate an HPACK table, let's declare
a pool. However the HPACK size is configured by the H2 mux, so it's
also this one which allocates it after post_check.
Emeric Brun [Mon, 11 May 2020 12:41:31 +0000 (14:41 +0200)]
MEDIUM: sink: add global statement to create a new ring (sink buffer)
This patch adds the new global statement:
ring <name> [desc <desc>] [format <format>] [size <size>] [maxlen <length>]
Creates a named ring buffer which could be used on log line for instance.
<desc> is an optionnal description string of the ring. It will appear on
CLI. By default, <name> is reused to fill this field.
<format> is the log format used when generating syslog messages. It may be
one of the following :
iso A message containing only the ISO date, followed by the text.
The PID, process name and system name are omitted. This is
designed to be used with a local log server.
raw A message containing only the text. The level, PID, date, time,
process name and system name are omitted. This is designed to be
used in containers or during development, where the severity only
depends on the file descriptor used (stdout/stderr). This is
the default.
rfc3164 The RFC3164 syslog message format. This is the default.
(https://tools.ietf.org/html/rfc3164)
rfc5424 The RFC5424 syslog message format.
(https://tools.ietf.org/html/rfc5424)
short A message containing only a level between angle brackets such as
'<3>', followed by the text. The PID, date, time, process name
and system name are omitted. This is designed to be used with a
local log server. This format is compatible with what the systemd
logger consumes.
timed A message containing only a level between angle brackets such as
'<3>', followed by ISO date and by the text. The PID, process
name and system name are omitted. This is designed to be
used with a local log server.
<length> is the maximum length of event message stored into the ring,
including formatted header. If the event message is longer
than <length>, it would be truncated to this length.
<name> is the ring identifier, which follows the same naming convention as
proxies and servers.
<size> is the optionnal size in bytes. Default value is set to BUFSIZE.
Note: Historically sink's name and desc were refs on const strings. But with new
configurable rings a dynamic allocation is needed.
Emeric Brun [Wed, 6 May 2020 15:23:59 +0000 (17:23 +0200)]
MEDIUM: logs: buffer targets now rely on new sink_write
Before this path, they rely directly on ring_write bypassing
a part of the sink API.
Now the maxlen parameter of the log will apply only on the text
message part (and not the header, for this you woud prefer
to use the maxlen parameter on the sink/ring).
sink_write prototype was also reviewed to return the number of Bytes
written to be compliant with the other write functions.
Emeric Brun [Wed, 6 May 2020 12:33:46 +0000 (14:33 +0200)]
MEDIUM: sink: build header in sink_write for log formats
This patch extends the sink_write prototype and code to
handle the rfc5424 and rfc3164 header.
It uses header building tools from log.c. Doing this some
functions/vars have been externalized.
facility and minlevel have been removed from the struct sink
and passed to args at sink_write because they depends of the log
and not of the sink (they remained unused by rest of the code
until now).
Emeric Brun [Tue, 12 May 2020 17:33:15 +0000 (19:33 +0200)]
BUG/MINOR: logs: prevent double line returns in some events.
Historically some messages used to already contain the trailing LF but
not all, and __do_send_log adds a new one in needed cases. It also does
trim a trailing LF in certain cases while computing the max message
length, as a result of subtracting 1 to the available room in the
destination buffer. But the way it's done is wrong since some messages
still contain it.
So the code was fixed to always trim the trailing LF from messages if
present, and then only subtract 1 from the destination buffer room
instead of the size..
Note: new sink API is not designed to receive a trailing LF on
event messages
This could be backported to relevant stable versions with particular
care since the logic of the code changed a bit since 1.6 and there
may be other locations that need to be adjusted.
MEDIUM: checks: Make post-41 the default mode for mysql checks
MySQL 4.1 is old enough to be the default mode for mysql checks. So now, once a
username is defined, post-41 mode is automatically used. To do mysql checks on
previous MySQL version, the argument "pre-41" must be used.
Note, it is a compatibility breakage for everyone using an antique and
unsupported MySQL version.
MINOR: config: Don't dump keywords if argument is NULL
Helper functions are used to dump bind, server or filter keywords. These
functions are used to report errors during the configuration parsing. To have a
coherent API, these functions are now prepared to handle a null pointer as
argument. If so, no action is performed and functions immediately return.
This patch should fix the issue #631. It is not a bug. There is no reason to
backport it.
BUG/MINOR: cache: Don't needlessly test "cache" keyword in parse_cache_flt()
parse_cache_flt() is the registered callback for the "cache" filter keyword. It
is only called when the "cache" keyword is found on a filter line. So, it is
useless to test the filter name in the callback function.
This patch should fix the issue #634. It may be backported as far as 1.9.
Ilya Shipitsin [Sat, 16 May 2020 17:42:12 +0000 (22:42 +0500)]
CLEANUP: acl: remove unused assignment
Coverity found unused variable assignment
CID 1299671 (#1 of 1): Unused value (UNUSED_VALUE)assigned_pointer:
Assigning value from args[arg + 1] to word here, but that stored
value is overwritten before it can be used.
958 word = args[arg + 1];
959 arg = arg_end;
William Dauchy [Sun, 17 May 2020 11:41:53 +0000 (13:41 +0200)]
BUILD: ssl: include buffer common headers for ssl_sock_ctx
since commit c0cdaffaa338 ("REORG: ssl: move ssl_sock_ctx and fix
cross-dependencies issues"), `struct ssl_sock_ctx` was moved in
ssl_sock.h. As it contains a `struct buffer`, including
`common/buffer.h` is now mandatory. I encountered an issue while
including ssl_sock.h on another patch:
include/types/ssl_sock.h:240:16: error: field ‘early_buf’ has incomplete type
240 | struct buffer early_buf; /* buffer to store the early data received */
no backport needed.
Fixes: c0cdaffaa338 ("REORG: ssl: move ssl_sock_ctx and fix
cross-dependencies issues") Signed-off-by: William Dauchy <w.dauchy@criteo.com>
Marcin Deranek [Fri, 15 May 2020 16:32:51 +0000 (18:32 +0200)]
MINOR: stats: Expose native cum_req metric for a server
Expose native cum_req metric for a server: so far it was calculated as a
sum or all responses. Rename it from Cum. HTTP Responses to Cum. HTTP
Requests to be consistent with Frontend and Backend.
Marcin Deranek [Fri, 15 May 2020 16:26:18 +0000 (18:26 +0200)]
MINOR: stats: Prepare for more accurate moving averages
Add swrate_add_dynamic function which is similar to swrate_add, but more
accurate when calculating moving averages when not enough samples have
been processed yet.
CI: travis-ci: switch arm64 builds to use openssl from distro
there are ongoing arm64 failures in travis-ci. we suspect that
build stucks on scripts/build-ssl.sh which is called during build.
let us try to use openssl from ubuntu distro and not build it
ourselves. to achive that we add "install" section without
build-ssl.sh
REORG: ssl: move ssl_sock_ctx and fix cross-dependencies issues
In order to move all SSL sample fetches in another file, moving the
ssl_sock_ctx definition in a .h file is required.
Unfortunately it became a cross dependencies hell to solve, because of
the struct wait_event field, so <types/connection.h> is needed which
created other problems.