haproxy/src/haproxy.c

3962 lines
114 KiB
C
Raw Normal View History

/*
* HAProxy : High Availability-enabled HTTP/TCP proxy
[RELEASE] Released version 3.0-dev1 Released version 3.0-dev1 with the following main changes : - MINOR: channel: Use dedicated functions to deal with STREAMER flags - MEDIUM: applet: Handle channel's STREAMER flags on applets size - MINOR: applets: Use channel's field to compute amount of data received - MEDIUM: cache: Save body size of cached objects and track it on delivery - MEDIUM: cache: Add support for endp-to-endp fast-forwarding - MINOR: cache: Add global option to enable/disable zero-copy forwarding - MINOR: pattern: Use reference name as filename to read patterns from a file - MEDIUM: pattern: Add support for virtual and optional files for patterns - DOC: config: Add section about name format for maps and ACLs - DOC: management/lua: Update commands about map and acl - MINOR: promex: Add support for specialized front/back/li/srv metric names - MINOR: promex: Export active/backup metrics per-server - BUG/MINOR: ssl: Double free of OCSP Certificate ID - MINOR: ssl/cli: Add ha_(warning|alert) msgs to CLI ckch callback - BUG/MINOR: ssl: Wrong OCSP CID after modifying an SSL certficate - BUG/MINOR: lua: Wrong OCSP CID after modifying an SSL certficate (LUA) - DOC: configuration: typo req.ssl_hello_type - MINOR: hq-interop: add fastfwd support - CLEANUP: mux_quic: rename ffwd function with prefix qmux_strm_ - MINOR: mux-quic: add traces for 0-copy/fast-forward - BUG/MINOR: mworker/cli: fix set severity-output support - CLEANUP: mworker/cli: add comments about pcli_find_and_exec_kw() - BUG/MEDIUM: quic: Possible buffer overflow when building TLS records - BUILD: ssl: update types in wolfssl cert selection callback - MINOR: ssl: activate the certificate selection callback for WolfSSL - CI: github: switch to wolfssl git-c4b77ad for new PR - BUG/MEDIUM: map/acl: pat_ref_{set,delete}_by_id regressions - BUG/MINOR: ext-check: cannot use without preserve-env - CLEANUP: mux-quic: remove unused prototype - MINOR: mux-quic: clean up qcs Rx buffer allocation API - MINOR: mux-quic: clean up qcs Tx buffer allocation API - CLEANUP: mux-quic: clean up app ops callback definitions - MINOR: mux-quic: factorize QC_SF_UNKNOWN_PL_LENGTH set - MINOR: h3: complete traces for sending - MINOR: h3: adjust zero-copy sending related code - MINOR: hq-interop: use zero-copy to transfer single HTX data block - BUG/MEDIUM: quic: QUIC CID removed from tree without locking - BUG/MEDIUM: stconn: Block zero-copy forwarding if EOS/ERROR on consumer side - BUG/MEDIUM: mux-h1: Cound data from input buf during zero-copy forwarding - BUG/MEDIUM: mux-h1: Explicitly skip request's C-L header if not set originally - CLEANUP: mux-h1: Fix a trace message about C-L header addition - BUG/MEDIUM: mux-h2: Report too large HEADERS frame only when rxbuf is empty - BUG/MEDIUM: mux-quic: report early error on stream - DOC: config: add arguments to sample fetch methods in the table - DOC: config: also add arguments to the converters in the table - BUG/MINOR: resolvers: default resolvers fails when network not configured - SCRIPTS: mk-patch-list: produce a list of patches - DEV: patchbot: add the AI-based bot to pre-select candidate patches to backport - BUG/MEDIUM: mux-h2: Switch pending error to error if demux buffer is empty - BUG/MEDIUM: mux-h2: Only Report H2C error on read error if demux buffer is empty - BUG/MEDIUM: mux-h2: Don't report error on SE if error is only pending on H2C - BUG/MEDIUM: mux-h2: Don't report error on SE for closed H2 streams - DOC: config: Update documentation about local haproxy response - DEV: patchbot: use checked buttons as reference instead of internal table - DEV: patchbot: allow to show/hide backported patches - MINOR: h3: remove quic_conn only reference - BUG/MINOR: server: Use the configured address family for the initial resolution - MINOR: mux-quic: remove qcc_shutdown() from qcc_release() - MINOR: mux-quic: use qcc_release in case of init failure - MINOR: mux-quic: adjust error code in init failure - MINOR: h3: add traces for connection init stage - BUG/MINOR: h3: properly handle alloc failure on finalize - MINOR: h3: use INTERNAL_ERROR code for init failure - BUG/MAJOR: stconn: Disable zero-copy forwarding if consumer is shut or in error - MINOR: stats: store the parent proxy in stats ctx (http) - BUG/MEDIUM: stats: unhandled switching rules with TCP frontend - MEDIUM: proxy: set PR_O_HTTP_UPG on implicit upgrades - MINOR: proxy: monitor-uri works with tcp->http upgrades - OPTIM: server: eb lookup for server_find_by_name() - OPTIM: server: ebtree lookups for findserver_unique_* functions - MINOR: server/event_hdl: add server_inetaddr struct to facilitate event data usage - MINOR: server/event_hdl: update _srv_event_hdl_prepare_inetaddr prototype - BUG/MINOR: server/event_hdl: propagate map port info through inetaddr event - MINOR: server: ensure connection cleanup on server addr changes - CLEANUP: server/event_hdl: remove purge_conn hint in INETADDR event - MEDIUM: server: merge srv_update_addr() and srv_update_addr_port() logic - CLEANUP: server: remove unused server_parse_addr_change_request() function - CLEANUP: resolvers: remove duplicate func prototype - MINOR: resolvers: add unique numeric id to nameservers - MEDIUM: server: make server_set_inetaddr() updater serializable - MINOR: server/event_hdl: expose updater info through INETADDR event - MINOR: server: add dns hint in server_inetaddr_updater struct - MEDIUM: server/dns: clear RMAINT when addr resolves again - BUG/MINOR: server/dns: use server_set_inetaddr() to unset srv addr from DNS - BUG/MEDIUM: server/dns: perform svc_port updates atomically from SRV records - MEDIUM: peers: use server as stream target - CLEANUP: peers: remove unused sock_init_arg struct member - CLEANUP: peers: remove unused "proto" and "xprt" struct members - MINOR: peers: rely on srv->addr and remove peer->addr - DOC: config: add context hint for server keywords - MINOR: stktable: add table_process_entry helper function - MINOR: stktable: use {show,set,clear} table with ptr - MINOR: map: add map_*_key converters to provide the matching key - DOC: fix typo for fastfwd QUIC option - BUG/MINOR: mux-quic: always report error to SC on RESET_STREAM emission - MEDIUM: mux-quic: add BUG_ON if sending on locally closed QCS - BUG/MINOR: mux-quic: disable fast-fwd if connection on error - BUG/MINOR: quic: Wrong keylog callback setting. - BUG/MINOR: quic: Missing call to TLS message callbacks - MINOR: h3: check connection error during sending - BUG/MINOR: h3: close connection on header list too big - BUG/MINOR: h3: close connection on sending alloc errors - BUG/MINOR: h3: disable fast-forward on buffer alloc failure - Revert "MINOR: mux-quic: Disable zero-copy forwarding for send by default" - MINOR: stktable: stktable_data_ptr() cannot fail in table_process_entry() - CLEANUP: assorted typo fixes in the code and comments - CI: use semantic version compare for determing "latest" OpenSSL - CLEANUP: server: remove ambiguous check in srv_update_addr_port() - CLEANUP: resolvers: remove unused RSLV_UPD_OBSOLETE_IP flag - CLEANUP: resolvers: remove some more unused RSLV_UDP flags - MEDIUM: server: simplify snr_set_srv_down() to prevent confusions - MINOR: backend: export get_server_*() functions - MINOR: tcpcheck: export proxy_parse_tcpcheck() - MEDIUM: udp: allow to retrieve the frontend destination address - MINOR: global: export a way to list build options - MINOR: debug: add features and build options to "show dev" - BUG/MINOR: server: fix server_find_by_name() usage during parsing - REGTESTS: check attach-srv out of order declaration - CLEANUP: quic: Remaining useless code into server part - BUILD: quic: Missing quic_ssl.h header protection - BUG/MEDIUM: h3: fix incorrect snd_buf return value - MINOR: h3: do not consider missing buf room as error on trailers - BUG/MEDIUM: stconn: Forward shutdown on write timeout only if it is forwardable - BUG/MEDIUM: stconn: Set fsb date if zero-copy forwarding is blocked during nego - BUG/MEDIUM: spoe: Never create new spoe applet if there is no server up - MINOR: mux-h2: support limiting the total number of H2 streams per connection - CLEANUP: mux-h2: remove the printfs from previous commit on h2 streams limit. - DEV: h2: add the ability to emit literals in mkhdr - DEV: h2: add the preface as well in supported output types - DEV: h2: support passing raw data for a frame - IMPORT: ebtree: implement and use flsnz_long() to count bits - IMPORT: ebtree: switch the sizes and offsets to size_t and ssize_t - IMPORT: ebtree: rework the fls macros to better deal with arch-specific ones - IMPORT: ebtree: make string_equal_bits turn back to unsigned char - IMPORT: ebtree: use unsigned ints for flznz() - IMPORT: ebtree: make string_equal_bits() return an unsigned
2024-01-06 08:09:35 -05:00
* Copyright 2000-2024 Willy Tarreau <willy@haproxy.org>.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*/
#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <ctype.h>
#include <dirent.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/tcp.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netdb.h>
#include <fcntl.h>
#include <errno.h>
#include <signal.h>
#include <stdarg.h>
#include <sys/resource.h>
#include <sys/utsname.h>
#include <sys/wait.h>
#include <time.h>
#include <syslog.h>
#include <grp.h>
#ifdef USE_THREAD
#include <pthread.h>
#endif
#ifdef USE_CPU_AFFINITY
#include <sched.h>
#if defined(__FreeBSD__) || defined(__DragonFly__)
#include <sys/param.h>
#ifdef __FreeBSD__
#include <sys/cpuset.h>
#endif
#endif
#endif
#if defined(USE_PRCTL)
#include <sys/prctl.h>
#endif
#if defined(USE_PROCCTL)
#include <sys/procctl.h>
#endif
#ifdef DEBUG_FULL
#include <assert.h>
#endif
MEDIUM: mworker: Add systemd `Type=notify` support This patch adds support for `Type=notify` to the systemd unit. Supporting `Type=notify` improves both starting as well as reloading of the unit, because systemd will be let known when the action completed. See this quote from `systemd.service(5)`: > Note however that reloading a daemon by sending a signal (as with the > example line above) is usually not a good choice, because this is an > asynchronous operation and hence not suitable to order reloads of > multiple services against each other. It is strongly recommended to > set ExecReload= to a command that not only triggers a configuration > reload of the daemon, but also synchronously waits for it to complete. By making systemd aware of a reload in progress it is able to wait until the reload actually succeeded. This patch introduces both a new `USE_SYSTEMD` build option which controls including the sd-daemon library as well as a `-Ws` runtime option which runs haproxy in master-worker mode with systemd support. When haproxy is running in master-worker mode with systemd support it will send status messages to systemd using `sd_notify(3)` in the following cases: - The master process forked off the worker processes (READY=1) - The master process entered the `mworker_reload()` function (RELOADING=1) - The master process received the SIGUSR1 or SIGTERM signal (STOPPING=1) Change the unit file to specify `Type=notify` and replace master-worker mode (`-W`) with master-worker mode with systemd support (`-Ws`). Future evolutions of this feature could include making use of the `STATUS` feature of `sd_notify()` to send information about the number of active connections to systemd. This would require bidirectional communication between the master and the workers and thus is left for future work.
2017-11-20 09:58:35 -05:00
#if defined(USE_SYSTEMD)
#include <systemd/sd-daemon.h>
#endif
#include <import/sha1.h>
#include <haproxy/acl.h>
#include <haproxy/action.h>
#include <haproxy/activity.h>
#include <haproxy/api.h>
#include <haproxy/arg.h>
#include <haproxy/auth.h>
#include <haproxy/base64.h>
#include <haproxy/capture-t.h>
#include <haproxy/cfgcond.h>
#include <haproxy/cfgdiag.h>
#include <haproxy/cfgparse.h>
#include <haproxy/chunk.h>
#include <haproxy/cli.h>
#include <haproxy/clock.h>
#include <haproxy/connection.h>
#ifdef USE_CPU_AFFINITY
#include <haproxy/cpuset.h>
#endif
#include <haproxy/debug.h>
#include <haproxy/dns.h>
#include <haproxy/dynbuf.h>
#include <haproxy/errors.h>
#include <haproxy/fd.h>
#include <haproxy/filters.h>
#include <haproxy/global.h>
#include <haproxy/hlua.h>
#include <haproxy/http_rules.h>
#if defined(USE_LINUX_CAP)
#include <haproxy/linuxcap.h>
#endif
#include <haproxy/list.h>
#include <haproxy/listener.h>
#include <haproxy/log.h>
#include <haproxy/mworker.h>
#include <haproxy/namespace.h>
#include <haproxy/net_helper.h>
#include <haproxy/openssl-compat.h>
#include <haproxy/quic_conn.h>
#include <haproxy/quic_tp-t.h>
#include <haproxy/pattern.h>
#include <haproxy/peers.h>
#include <haproxy/pool.h>
#include <haproxy/protocol.h>
#include <haproxy/proto_tcp.h>
#include <haproxy/proxy.h>
#include <haproxy/regex.h>
#include <haproxy/sample.h>
#include <haproxy/server.h>
#include <haproxy/session.h>
#include <haproxy/signal.h>
#include <haproxy/sock.h>
#include <haproxy/sock_inet.h>
#include <haproxy/ssl_sock.h>
#include <haproxy/stats-t.h>
#include <haproxy/stream.h>
#include <haproxy/task.h>
#include <haproxy/thread.h>
#include <haproxy/time.h>
#include <haproxy/tools.h>
#include <haproxy/trace.h>
#include <haproxy/uri_auth-t.h>
#include <haproxy/vars.h>
#include <haproxy/version.h>
BUILD: re-implement an initcall variant without using executable sections The current initcall implementation relies on dedicated sections (one section per init stage) to store the initcall descriptors. Then upon startup, these sections are scanned from beginning to end and all items found there are called in sequence. On platforms like AIX or Cygwin it seems difficult to figure the beginning and end of sections as the linker doesn't seem to provide the corresponding symbols. In order to replace this, this patch simply implements an array of single linked (one per init stage) which are fed using constructors for each register call. These constructors are declared static, with a name depending on their line number in the file, in order to avoid name clashes. The final effect is the same, except that the method is slightly more expensive in that it explicitly produces code to register these initcalls : $ size haproxy.sections haproxy.constructor text data bss dec hex filename 4060312 249176 1457652 5767140 57ffe4 haproxy.sections 4062862 260408 1457652 5780922 5835ba haproxy.constructor This mechanism is enabled as an alternative to the default one when build option USE_OBSOLETE_LINKER is set. This option is currently enabled by default only on AIX and Cygwin, and may be attempted for any target which fails to build complaining about missing symbols __start_init_* and/or __stop_init_*. Once confirmed as a reliable fix, this will likely have to be backported to 1.9 where AIX and Cygwin do not build anymore.
2019-03-29 16:30:17 -04:00
/* array of init calls for older platforms */
DECLARE_INIT_STAGES;
/* create a read_mostly section to hold variables which are accessed a lot
* but which almost never change. The purpose is to isolate them in their
* own cache lines where they don't risk to be perturbated by write accesses
* to neighbor variables. We need to create an empty aligned variable for
* this. The fact that the variable is of size zero means that it will be
* eliminated at link time if no other variable uses it, but alignment will
* be respected.
*/
empty_t __read_mostly_align HA_SECTION("read_mostly") ALIGNED(64);
#ifdef BUILD_FEATURES
char *build_features = BUILD_FEATURES;
#else
char *build_features = "";
#endif
/* list of config files */
static struct list cfg_cfgfiles = LIST_HEAD_INIT(cfg_cfgfiles);
int pid; /* current process id */
static unsigned long stopping_tgroup_mask; /* Thread groups acknowledging stopping */
/* global options */
struct global global = {
.hard_stop_after = TICK_ETERNITY,
MEDIUM: global: Add a "close-spread-time" option to spread soft-stop on time window The new 'close-spread-time' global option can be used to spread idle and active HTTP connction closing after a SIGUSR1 signal is received. This allows to limit bursts of reconnections when too many idle connections are closed at once. Indeed, without this new mechanism, in case of soft-stop, all the idle connections would be closed at once (after the grace period is over), and all active HTTP connections would be closed by appending a "Connection: close" header to the next response that goes over it (or via a GOAWAY frame in case of HTTP2). This patch adds the support of this new option for HTTP as well as HTTP2 connections. It works differently on active and idle connections. On active connections, instead of sending systematically the GOAWAY frame or adding the 'Connection: close' header like before once the soft-stop has started, a random based on the remainder of the close window is calculated, and depending on its result we could decide to keep the connection alive. The random will be recalculated for any subsequent request/response on this connection so the GOAWAY will still end up being sent, but we might wait a few more round trips. This will ensure that goaways are distributed along a longer time window than before. On idle connections, a random factor is used when determining the expire field of the connection's task, which should naturally spread connection closings on the time window (see h2c_update_timeout). This feature request was described in GitHub issue #1614. This patch should be backported to 2.5. It depends on "BUG/MEDIUM: mux-h2: make use of http-request and keep-alive timeouts" which refactorized the timeout management of HTTP2 connections.
2022-04-08 12:04:18 -04:00
.close_spread_time = TICK_ETERNITY,
.close_spread_end = TICK_ETERNITY,
.numa_cpu_mapping = 1,
.nbthread = 0,
.req_count = 0,
.loggers = LIST_HEAD_INIT(global.loggers),
.maxzlibmem = DEFAULT_MAXZLIBMEM * 1024U * 1024U,
.comp_rate_lim = 0,
.ssl_server_verify = SSL_SERVER_VERIFY_REQUIRED,
.unix_bind = {
.ux = {
.uid = -1,
.gid = -1,
.mode = 0,
}
},
.tune = {
.options = GTUNE_LISTENER_MQ_OPT,
.bufsize = (BUFSIZE + 2*sizeof(void *) - 1) & -(2*sizeof(void *)),
.maxrewrite = MAXREWRITE,
MAJOR: session: only wake up as many sessions as available buffers permit We've already experimented with three wake up algorithms when releasing buffers : the first naive one used to wake up far too many sessions, causing many of them not to get any buffer. The second approach which was still in use prior to this patch consisted in waking up either 1 or 2 sessions depending on the number of FDs we had released. And this was still inaccurate. The third one tried to cover the accuracy issues of the second and took into consideration the number of FDs the sessions would be willing to use, but most of the time we ended up waking up too many of them for nothing, or deadlocking by lack of buffers. This patch completely removes the need to allocate two buffers at once. Instead it splits allocations into critical and non-critical ones and implements a reserve in the pool for this. The deadlock situation happens when all buffers are be allocated for requests pending in a maxconn-limited server queue, because then there's no more way to allocate buffers for responses, and these responses are critical to release the servers's connection in order to release the pending requests. In fact maxconn on a server creates a dependence between sessions and particularly between oldest session's responses and latest session's requests. Thus, it is mandatory to get a free buffer for a response in order to release a server connection which will permit to release a request buffer. Since we definitely have non-symmetrical buffers, we need to implement this logic in the buffer allocation mechanism. What this commit does is implement a reserve of buffers which can only be allocated for responses and that will never be allocated for requests. This is made possible by the requester indicating how much margin it wants to leave after the allocation succeeds. Thus it is a cooperative allocation mechanism : the requester (process_session() in general) prefers not to get a buffer in order to respect other's need for response buffers. The session management code always knows if a buffer will be used for requests or responses, so that is not difficult : - either there's an applet on the initiator side and we really need the request buffer (since currently the applet is called in the context of the session) - or we have a connection and we really need the response buffer (in order to support building and sending an error message back) This reserve ensures that we don't take all allocatable buffers for requests waiting in a queue. The downside is that all the extra buffers are really allocated to ensure they can be allocated. But with small values it is not an issue. With this change, we don't observe any more deadlocks even when running with maxconn 1 on a server under severely constrained memory conditions. The code becomes a bit tricky, it relies on the scheduler's run queue to estimate how many sessions are already expected to run so that it doesn't wake up everyone with too few resources. A better solution would probably consist in having two queues, one for urgent requests and one for normal requests. A failed allocation for a session dealing with an error, a connection event, or the need for a response (or request when there's an applet on the left) would go to the urgent request queue, while other requests would go to the other queue. Urgent requests would be served from 1 entry in the pool, while the regular ones would be served only according to the reserve. Despite not yet having this, it works remarkably well. This mechanism is quite efficient, we don't perform too many wake up calls anymore. For 1 million sessions elapsed during massive memory contention, we observe about 4.5M calls to process_session() compared to 4.0M without memory constraints. Previously we used to observe up to 16M calls, which rougly means 12M failures. During a test run under high memory constraints (limit enforced to 27 MB instead of the 58 MB normally needed), performance used to drop by 53% prior to this patch. Now with this patch instead it *increases* by about 1.5%. The best effect of this change is that by limiting the memory usage to about 2/3 to 3/4 of what is needed by default, it's possible to increase performance by up to about 18% mainly due to the fact that pools are reused more often and remain hot in the CPU cache (observed on regular HTTP traffic with 20k objects, buffers.limit = maxconn/10, buffers.reserve = limit/2). Below is an example of scenario which used to cause a deadlock previously : - connection is received - two buffers are allocated in process_session() then released - one is allocated when receiving an HTTP request - the second buffer is allocated then released in process_session() for request parsing then connection establishment. - poll() says we can send, so the request buffer is sent and released - process session gets notified that the connection is now established and allocates two buffers then releases them - all other sessions do the same till one cannot get the request buffer without hitting the margin - and now the server responds. stream_interface allocates the response buffer and manages to get it since it's higher priority being for a response. - but process_session() cannot allocate the request buffer anymore => We could end up with all buffers used by responses so that none may be allocated for a request in process_session(). When the applet processing leaves the session context, the test will have to be changed so that we always allocate a response buffer regardless of the left side (eg: H2->H1 gateway). A final improvement would consists in being able to only retry the failed I/O operation without waking up a task, but to date all experiments to achieve this have proven not to be reliable enough.
2014-11-26 19:11:56 -05:00
.reserved_bufs = RESERVED_BUFS,
.pattern_cache = DEFAULT_PAT_LRU_SIZE,
.pool_low_ratio = 20,
.pool_high_ratio = 25,
.max_http_hdr = MAX_HTTP_HDR,
#ifdef USE_OPENSSL
.sslcachesize = SSLCACHESIZE,
#endif
.comp_maxlevel = 1,
#ifdef DEFAULT_IDLE_TIMER
.idle_timer = DEFAULT_IDLE_TIMER,
#else
.idle_timer = 1000, /* 1 second */
#endif
.nb_stk_ctr = MAX_SESS_STKCTR,
.default_shards = -2, /* by-group */
#ifdef USE_QUIC
.quic_backend_max_idle_timeout = QUIC_TP_DFLT_BACK_MAX_IDLE_TIMEOUT,
.quic_frontend_max_idle_timeout = QUIC_TP_DFLT_FRONT_MAX_IDLE_TIMEOUT,
.quic_frontend_max_streams_bidi = QUIC_TP_DFLT_FRONT_MAX_STREAMS_BIDI,
.quic_retry_threshold = QUIC_DFLT_RETRY_THRESHOLD,
.quic_max_frame_loss = QUIC_DFLT_MAX_FRAME_LOSS,
.quic_streams_buf = 30,
#endif /* USE_QUIC */
},
#ifdef USE_OPENSSL
#ifdef DEFAULT_MAXSSLCONN
.maxsslconn = DEFAULT_MAXSSLCONN,
#endif
#endif
/* others NULL OK */
};
/*********************************************************************/
int stopping; /* non zero means stopping in progress */
int killed; /* non zero means a hard-stop is triggered */
int jobs = 0; /* number of active jobs (conns, listeners, active tasks, ...) */
int unstoppable_jobs = 0; /* number of active jobs that can't be stopped during a soft stop */
int active_peers = 0; /* number of active peers (connection attempts and connected) */
int connected_peers = 0; /* number of connected peers (verified ones) */
int arg_mode = 0; /* MODE_DEBUG etc as passed on command line ... */
char *change_dir = NULL; /* set when -C is passed */
char *check_condition = NULL; /* check condition passed to -cc */
/* Here we store information about the pids of the processes we may pause
* or kill. We will send them a signal every 10 ms until we can bind to all
* our ports. With 200 retries, that's about 2 seconds.
*/
#define MAX_START_RETRIES 200
static int *oldpids = NULL;
static int oldpids_sig; /* use USR1 or TERM */
/* Path to the unix socket we use to retrieve listener sockets from the old process */
static const char *old_unixsocket;
int atexit_flag = 0;
int nb_oldpids = 0;
const int zero = 0;
const int one = 1;
const struct linger nolinger = { .l_onoff = 1, .l_linger = 0 };
char hostname[MAX_HOSTNAME_LEN];
char *localpeer = NULL;
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
static char *kwd_dump = NULL; // list of keyword dumps to produce
static char **old_argv = NULL; /* previous argv but cleaned up */
struct list proc_list = LIST_HEAD_INIT(proc_list);
int master = 0; /* 1 if in master, 0 if in child */
unsigned int rlim_fd_cur_at_boot = 0;
unsigned int rlim_fd_max_at_boot = 0;
/* per-boot randomness */
unsigned char boot_seed[20]; /* per-boot random seed (160 bits initially) */
/* takes the thread config in argument or NULL for any thread */
static void *run_thread_poll_loop(void *data);
/* bitfield of a few warnings to emit just once (WARN_*) */
unsigned int warned = 0;
/* set if experimental features have been used for the current process */
unsigned int tainted = 0;
unsigned int experimental_directives_allowed = 0;
int check_kw_experimental(struct cfg_keyword *kw, const char *file, int linenum,
char **errmsg)
{
if (kw->flags & KWF_EXPERIMENTAL) {
if (!experimental_directives_allowed) {
memprintf(errmsg, "parsing [%s:%d] : '%s' directive is experimental, must be allowed via a global 'expose-experimental-directives'",
file, linenum, kw->kw);
return 1;
}
mark_tainted(TAINTED_CONFIG_EXP_KW_DECLARED);
}
return 0;
}
/* master CLI configuration (-S flag) */
struct list mworker_cli_conf = LIST_HEAD_INIT(mworker_cli_conf);
/* These are strings to be reported in the output of "haproxy -vv". They may
* either be constants (in which case must_free must be zero) or dynamically
* allocated strings to pass to free() on exit, and in this case must_free
* must be non-zero.
*/
struct list build_opts_list = LIST_HEAD_INIT(build_opts_list);
struct build_opts_str {
struct list list;
const char *str;
int must_free;
};
/*********************************************************************/
/* general purpose functions ***************************************/
/*********************************************************************/
/* used to register some build option strings at boot. Set must_free to
* non-zero if the string must be freed upon exit.
*/
void hap_register_build_opts(const char *str, int must_free)
{
struct build_opts_str *b;
b = calloc(1, sizeof(*b));
if (!b) {
fprintf(stderr, "out of memory\n");
exit(1);
}
b->str = str;
b->must_free = must_free;
LIST_APPEND(&build_opts_list, &b->list);
}
/* returns the first build option when <curr> is NULL, or the next one when
* <curr> is passed the last returned value. NULL when there is no more entries
* in the list. Otherwise the returned pointer is &opt->str so the caller can
* print it as *ret.
*/
const char **hap_get_next_build_opt(const char **curr)
{
struct build_opts_str *head, *start;
head = container_of(&build_opts_list, struct build_opts_str, list);
if (curr)
start = container_of(curr, struct build_opts_str, str);
else
start = head;
start = container_of(start->list.n, struct build_opts_str, list);
if (start == head)
return NULL;
return &start->str;
}
/* used to make a new feature appear in the build_features list at boot time.
* The feature must be in the format "XXX" without the leading "+" which will
* be automatically appended.
*/
void hap_register_feature(const char *name)
{
static int must_free = 0;
int new_len = strlen(build_features) + 2 + strlen(name);
char *new_features;
new_features = malloc(new_len + 1);
if (!new_features)
return;
strlcpy2(new_features, build_features, new_len);
snprintf(new_features, new_len + 1, "%s +%s", build_features, name);
if (must_free)
ha_free(&build_features);
build_features = new_features;
must_free = 1;
}
#define VERSION_MAX_ELTS 7
/* This function splits an haproxy version string into an array of integers.
* The syntax of the supported version string is the following:
*
* <a>[.<b>[.<c>[.<d>]]][-{dev,pre,rc}<f>][-*][-<g>]
*
* This validates for example:
* 1.2.1-pre2, 1.2.1, 1.2.10.1, 1.3.16-rc1, 1.4-dev3, 1.5-dev18, 1.5-dev18-43
* 2.4-dev18-f6818d-20
*
* The result is set in a array of <VERSION_MAX_ELTS> elements. Each letter has
* one fixed place in the array. The tags take a numeric value called <e> which
* defaults to 3. "dev" is 1, "rc" and "pre" are 2. Numbers not encountered are
* considered as zero (henxe 1.5 and 1.5.0 are the same).
*
* The resulting values are:
* 1.2.1-pre2 1, 2, 1, 0, 2, 2, 0
* 1.2.1 1, 2, 1, 0, 3, 0, 0
* 1.2.10.1 1, 2, 10, 1, 3, 0, 0
* 1.3.16-rc1 1, 3, 16, 0, 2, 1, 0
* 1.4-dev3 1, 4, 0, 0, 1, 3, 0
* 1.5-dev18 1, 5, 0, 0, 1, 18, 0
* 1.5-dev18-43 1, 5, 0, 0, 1, 18, 43
* 2.4-dev18-f6818d-20 2, 4, 0, 0, 1, 18, 20
*
* The function returns non-zero if the conversion succeeded, or zero if it
* failed.
*/
int split_version(const char *version, unsigned int *value)
{
const char *p, *s;
char *error;
int nelts;
/* Initialize array with zeroes */
for (nelts = 0; nelts < VERSION_MAX_ELTS; nelts++)
value[nelts] = 0;
value[4] = 3;
p = version;
/* If the version number is empty, return false */
if (*p == '\0')
return 0;
/* Convert first number <a> */
value[0] = strtol(p, &error, 10);
p = error + 1;
if (*error == '\0')
return 1;
if (*error == '-')
goto split_version_tag;
if (*error != '.')
return 0;
/* Convert first number <b> */
value[1] = strtol(p, &error, 10);
p = error + 1;
if (*error == '\0')
return 1;
if (*error == '-')
goto split_version_tag;
if (*error != '.')
return 0;
/* Convert first number <c> */
value[2] = strtol(p, &error, 10);
p = error + 1;
if (*error == '\0')
return 1;
if (*error == '-')
goto split_version_tag;
if (*error != '.')
return 0;
/* Convert first number <d> */
value[3] = strtol(p, &error, 10);
p = error + 1;
if (*error == '\0')
return 1;
if (*error != '-')
return 0;
split_version_tag:
/* Check for commit number */
if (*p >= '0' && *p <= '9')
goto split_version_commit;
/* Read tag */
if (strncmp(p, "dev", 3) == 0) { value[4] = 1; p += 3; }
else if (strncmp(p, "rc", 2) == 0) { value[4] = 2; p += 2; }
else if (strncmp(p, "pre", 3) == 0) { value[4] = 2; p += 3; }
else
goto split_version_commit;
/* Convert tag number */
value[5] = strtol(p, &error, 10);
p = error + 1;
if (*error == '\0')
return 1;
if (*error != '-')
return 0;
split_version_commit:
/* Search the last "-" */
s = strrchr(p, '-');
if (s) {
s++;
if (*s == '\0')
return 0;
value[6] = strtol(s, &error, 10);
if (*error != '\0')
value[6] = 0;
return 1;
}
/* convert the version */
value[6] = strtol(p, &error, 10);
if (*error != '\0')
value[6] = 0;
return 1;
}
/* This function compares the current haproxy version with an arbitrary version
* string. It returns:
* -1 : the version in argument is older than the current haproxy version
* 0 : the version in argument is the same as the current haproxy version
* 1 : the version in argument is newer than the current haproxy version
*
* Or some errors:
* -2 : the current haproxy version is not parsable
* -3 : the version in argument is not parsable
*/
int compare_current_version(const char *version)
{
unsigned int loc[VERSION_MAX_ELTS];
unsigned int mod[VERSION_MAX_ELTS];
int i;
/* split versions */
if (!split_version(haproxy_version, loc))
return -2;
if (!split_version(version, mod))
return -3;
/* compare versions */
for (i = 0; i < VERSION_MAX_ELTS; i++) {
if (mod[i] < loc[i])
return -1;
else if (mod[i] > loc[i])
return 1;
}
return 0;
}
void display_version()
{
struct utsname utsname;
printf("HAProxy version %s %s - https://haproxy.org/\n"
PRODUCT_STATUS "\n", haproxy_version, haproxy_date);
if (strlen(PRODUCT_URL_BUGS) > 0) {
char base_version[20];
int dots = 0;
char *del;
/* only retrieve the base version without distro-specific extensions */
for (del = haproxy_version; *del; del++) {
if (*del == '.')
dots++;
else if (*del < '0' || *del > '9')
break;
}
strlcpy2(base_version, haproxy_version, del - haproxy_version + 1);
if (dots < 2)
printf("Known bugs: https://github.com/haproxy/haproxy/issues?q=is:issue+is:open\n");
else
printf("Known bugs: " PRODUCT_URL_BUGS "\n", base_version);
}
if (uname(&utsname) == 0) {
printf("Running on: %s %s %s %s\n", utsname.sysname, utsname.release, utsname.version, utsname.machine);
}
}
static void display_build_opts()
{
const char **opt;
printf("Build options :"
#ifdef BUILD_TARGET
"\n TARGET = " BUILD_TARGET
#endif
#ifdef BUILD_CPU
"\n CPU = " BUILD_CPU
#endif
#ifdef BUILD_CC
"\n CC = " BUILD_CC
#endif
#ifdef BUILD_CFLAGS
"\n CFLAGS = " BUILD_CFLAGS
#endif
#ifdef BUILD_OPTIONS
"\n OPTIONS = " BUILD_OPTIONS
#endif
#ifdef BUILD_DEBUG
"\n DEBUG = " BUILD_DEBUG
#endif
"\n\nFeature list : %s"
"\n\nDefault settings :"
"\n bufsize = %d, maxrewrite = %d, maxpollevents = %d"
"\n\n",
build_features, BUFSIZE, MAXREWRITE, MAX_POLL_EVENTS);
for (opt = NULL; (opt = hap_get_next_build_opt(opt)); puts(*opt))
;
putchar('\n');
list_pollers(stdout);
putchar('\n');
list_mux_proto(stdout);
putchar('\n');
list_services(stdout);
putchar('\n');
list_filters(stdout);
putchar('\n');
}
/*
* This function prints the command line usage and exits
*/
static void usage(char *name)
{
display_version();
fprintf(stderr,
"Usage : %s [-f <cfgfile|cfgdir>]* [ -vdV"
"D ] [ -n <maxconn> ] [ -N <maxpconn> ]\n"
" [ -p <pidfile> ] [ -m <max megs> ] [ -C <dir> ] [-- <cfgfile>*]\n"
" -v displays version ; -vv shows known build options.\n"
" -d enters debug mode ; -db only disables background mode.\n"
" -dM[<byte>,help,...] debug memory (default: poison with <byte>/0x50)\n"
" -dt activate traces on stderr\n"
" -V enters verbose mode (disables quiet mode)\n"
" -D goes daemon ; -C changes to <dir> before loading files.\n"
" -W master-worker mode.\n"
MEDIUM: mworker: Add systemd `Type=notify` support This patch adds support for `Type=notify` to the systemd unit. Supporting `Type=notify` improves both starting as well as reloading of the unit, because systemd will be let known when the action completed. See this quote from `systemd.service(5)`: > Note however that reloading a daemon by sending a signal (as with the > example line above) is usually not a good choice, because this is an > asynchronous operation and hence not suitable to order reloads of > multiple services against each other. It is strongly recommended to > set ExecReload= to a command that not only triggers a configuration > reload of the daemon, but also synchronously waits for it to complete. By making systemd aware of a reload in progress it is able to wait until the reload actually succeeded. This patch introduces both a new `USE_SYSTEMD` build option which controls including the sd-daemon library as well as a `-Ws` runtime option which runs haproxy in master-worker mode with systemd support. When haproxy is running in master-worker mode with systemd support it will send status messages to systemd using `sd_notify(3)` in the following cases: - The master process forked off the worker processes (READY=1) - The master process entered the `mworker_reload()` function (RELOADING=1) - The master process received the SIGUSR1 or SIGTERM signal (STOPPING=1) Change the unit file to specify `Type=notify` and replace master-worker mode (`-W`) with master-worker mode with systemd support (`-Ws`). Future evolutions of this feature could include making use of the `STATUS` feature of `sd_notify()` to send information about the number of active connections to systemd. This would require bidirectional communication between the master and the workers and thus is left for future work.
2017-11-20 09:58:35 -05:00
#if defined(USE_SYSTEMD)
" -Ws master-worker mode with systemd notify support.\n"
#endif
" -q quiet mode : don't display messages\n"
" -c check mode : only check config files and exit\n"
" -cc check condition : evaluate a condition and exit\n"
" -n sets the maximum total # of connections (uses ulimit -n)\n"
" -m limits the usable amount of memory (in MB)\n"
" -N sets the default, per-proxy maximum # of connections (%d)\n"
" -L set local peer name (default to hostname)\n"
" -p writes pids of all children to this file\n"
" -dC[[key],line] display the configuration file, if there is a key, the file will be anonymised\n"
#if defined(USE_EPOLL)
" -de disables epoll() usage even when available\n"
#endif
#if defined(USE_KQUEUE)
" -dk disables kqueue() usage even when available\n"
#endif
#if defined(USE_EVPORTS)
" -dv disables event ports usage even when available\n"
#endif
#if defined(USE_POLL)
" -dp disables poll() usage even when available\n"
#endif
#if defined(USE_LINUX_SPLICE)
" -dS disables splice usage (broken on old kernels)\n"
#endif
#if defined(USE_GETADDRINFO)
" -dG disables getaddrinfo() usage\n"
#endif
#if defined(SO_REUSEPORT)
" -dR disables SO_REUSEPORT usage\n"
#endif
#if defined(HA_HAVE_DUMP_LIBS)
" -dL dumps loaded object files after config checks\n"
#endif
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
" -dK{class[,...]} dump registered keywords (use 'help' for list)\n"
" -dr ignores server address resolution failures\n"
" -dV disables SSL verify on servers side\n"
" -dW fails if any warning is emitted\n"
" -dD diagnostic mode : warn about suspicious configuration statements\n"
" -dF disable fast-forward\n"
" -dZ disable zero-copy forwarding\n"
" -sf/-st [pid ]* finishes/terminates old pids.\n"
" -x <unix_socket> get listening sockets from a unix socket\n"
" -S <bind>[,<bind options>...] new master CLI\n"
"\n",
name, cfg_maxpconn);
exit(1);
}
/*********************************************************************/
/* more specific functions ***************************************/
/*********************************************************************/
/* sends the signal <sig> to all pids found in <oldpids>. Returns the number of
* pids the signal was correctly delivered to.
*/
int tell_old_pids(int sig)
{
int p;
int ret = 0;
for (p = 0; p < nb_oldpids; p++)
if (kill(oldpids[p], sig) == 0)
ret++;
return ret;
}
/*
* remove a pid forom the olpid array and decrease nb_oldpids
* return 1 pid was found otherwise return 0
*/
int delete_oldpid(int pid)
{
int i;
for (i = 0; i < nb_oldpids; i++) {
if (oldpids[i] == pid) {
oldpids[i] = oldpids[nb_oldpids - 1];
oldpids[nb_oldpids - 1] = 0;
nb_oldpids--;
return 1;
}
}
return 0;
}
/*
* When called, this function reexec haproxy with -sf followed by current
* children PIDs and possibly old children PIDs if they didn't leave yet.
*/
static void mworker_reexec(int hardreload)
{
char **next_argv = NULL;
int old_argc = 0; /* previous number of argument */
int next_argc = 0;
int i = 0;
char *msg = NULL;
struct rlimit limit;
struct mworker_proc *current_child = NULL;
mworker_block_signals();
setenv("HAPROXY_MWORKER_REEXEC", "1", 1);
mworker_cleanup_proc();
mworker_proc_list_to_env(); /* put the children description in the env */
/* ensure that we close correctly every listeners before reexecuting */
mworker_cleanlisteners();
/* during the reload we must ensure that every FDs that can't be
* reuse (ie those that are not referenced in the proc_list)
* are closed or they will leak. */
/* close the listeners FD */
mworker_cli_proxy_stop();
if (fdtab)
deinit_pollers();
#ifdef HAVE_SSL_RAND_KEEP_RANDOM_DEVICES_OPEN
/* close random device FDs */
RAND_keep_random_devices_open(0);
#endif
/* restore the initial FD limits */
limit.rlim_cur = rlim_fd_cur_at_boot;
limit.rlim_max = rlim_fd_max_at_boot;
if (raise_rlim_nofile(&limit, &limit) != 0) {
ha_warning("Failed to restore initial FD limits (cur=%u max=%u), using cur=%u max=%u\n",
rlim_fd_cur_at_boot, rlim_fd_max_at_boot,
(unsigned int)limit.rlim_cur, (unsigned int)limit.rlim_max);
}
/* compute length */
while (old_argv[old_argc])
old_argc++;
/* 1 for haproxy -sf, 2 for -x /socket */
next_argv = calloc(old_argc + 1 + 2 + mworker_child_nb() + 1,
sizeof(*next_argv));
if (next_argv == NULL)
goto alloc_error;
/* copy the program name */
next_argv[next_argc++] = old_argv[0];
/* insert the new options just after argv[0] in case we have a -- */
if (getenv("HAPROXY_MWORKER_WAIT_ONLY") == NULL) {
/* add -sf <PID>* to argv */
if (mworker_child_nb() > 0) {
struct mworker_proc *child;
if (hardreload)
next_argv[next_argc++] = "-st";
else
next_argv[next_argc++] = "-sf";
list_for_each_entry(child, &proc_list, list) {
if (!(child->options & PROC_O_LEAVING) && (child->options & PROC_O_TYPE_WORKER))
current_child = child;
if (!(child->options & (PROC_O_TYPE_WORKER|PROC_O_TYPE_PROG)) || child->pid <= -1)
continue;
if ((next_argv[next_argc++] = memprintf(&msg, "%d", child->pid)) == NULL)
goto alloc_error;
msg = NULL;
}
}
if (current_child) {
/* add the -x option with the socketpair of the current worker */
next_argv[next_argc++] = "-x";
if ((next_argv[next_argc++] = memprintf(&msg, "sockpair@%d", current_child->ipc_fd[0])) == NULL)
goto alloc_error;
msg = NULL;
}
}
/* copy the previous options */
for (i = 1; i < old_argc; i++)
next_argv[next_argc++] = old_argv[i];
signal(SIGPROF, SIG_IGN);
execvp(next_argv[0], next_argv);
ha_warning("Failed to reexecute the master process [%d]: %s\n", pid, strerror(errno));
ha_free(&next_argv);
return;
alloc_error:
ha_free(&next_argv);
ha_warning("Failed to reexecute the master process [%d]: Cannot allocate memory\n", pid);
return;
}
/* reexec haproxy in waitmode */
static void mworker_reexec_waitmode()
{
setenv("HAPROXY_MWORKER_WAIT_ONLY", "1", 1);
mworker_reexec(0);
}
/* reload haproxy and emit a warning */
void mworker_reload(int hardreload)
{
struct mworker_proc *child;
struct per_thread_deinit_fct *ptdf;
ha_notice("Reloading HAProxy%s\n", hardreload?" (hard-reload)":"");
/* close the poller FD and the thread waker pipe FD */
list_for_each_entry(ptdf, &per_thread_deinit_list, list)
ptdf->fct();
/* increment the number of reloads */
list_for_each_entry(child, &proc_list, list) {
child->reloads++;
}
#if defined(USE_SYSTEMD)
if (global.tune.options & GTUNE_USE_SYSTEMD)
sd_notify(0, "RELOADING=1\nSTATUS=Reloading Configuration.\n");
#endif
mworker_reexec(hardreload);
}
static void mworker_loop()
{
/* Busy polling makes no sense in the master :-) */
global.tune.options &= ~GTUNE_BUSY_POLLING;
signal_unregister(SIGTTIN);
signal_unregister(SIGTTOU);
signal_unregister(SIGUSR1);
signal_unregister(SIGHUP);
signal_unregister(SIGQUIT);
signal_register_fct(SIGTERM, mworker_catch_sigterm, SIGTERM);
signal_register_fct(SIGUSR1, mworker_catch_sigterm, SIGUSR1);
signal_register_fct(SIGTTIN, mworker_broadcast_signal, SIGTTIN);
signal_register_fct(SIGTTOU, mworker_broadcast_signal, SIGTTOU);
signal_register_fct(SIGINT, mworker_catch_sigterm, SIGINT);
signal_register_fct(SIGHUP, mworker_catch_sighup, SIGHUP);
signal_register_fct(SIGUSR2, mworker_catch_sighup, SIGUSR2);
signal_register_fct(SIGCHLD, mworker_catch_sigchld, SIGCHLD);
mworker_unblock_signals();
mworker_cleantasks();
mworker_catch_sigchld(NULL); /* ensure we clean the children in case
some SIGCHLD were lost */
jobs++; /* this is the "master" job, we want to take care of the
signals even if there is no listener so the poll loop don't
leave */
fork_poller();
run_thread_poll_loop(NULL);
}
/*
* Reexec the process in failure mode, instead of exiting
*/
void reexec_on_failure()
{
struct mworker_proc *child;
if (!atexit_flag)
return;
/* get the info of the children in the env */
if (mworker_env_to_proc_list() < 0) {
exit(EXIT_FAILURE);
}
/* increment the number of failed reloads */
list_for_each_entry(child, &proc_list, list) {
child->failedreloads++;
}
BUG/MEDIUM: mworker: close unused transferred FDs on load failure When the master process is reloaded on a new config, it will try to connect to the previous process' socket to retrieve all known listening FDs to be reused by the new listeners. If listeners were removed, their unused FDs are simply closed. However there's a catch. In case a socket fails to bind, the master will cancel its startup and swithc to wait mode for a new operation to happen. In this case it didn't close the possibly remaining FDs that were left unused. It is very hard to hit this case, but it can happen during a troubleshooting session with fat fingers. For example, let's say a config runs like this: frontend ftp bind 1.2.3.4:20000-29999 The admin wants to extend the port range down to 10000-29999 and by mistake ends up with: frontend ftp bind 1.2.3.41:20000-29999 Upon restart the bind will fail if the address is not present, and the master will then switch to wait mode without releasing the previous FDs for 1.2.3.4:20000-29999 since they're now apparently unused. Then once the admin fixes the config and does: frontend ftp bind 1.2.3.4:10000-29999 The service will start, but will bind new sockets, half of them overlapping with the previous ones that were not properly closed. This may result in a startup error (if SO_REUSEPORT is not enabled or not available), in a FD number exhaustion (if the error is repeated many times), or in connections being randomly accepted by the process if they sometimes land on the old FD that nobody listens on. This patch will need to be backported as far as 1.8, and depends on previous patch: MINOR: sock: move the unused socket cleaning code into its own function Note that before 2.3 most of the code was located inside haproxy.c, so the patch above should probably relocate the function there instead of sock.c.
2022-01-28 12:40:06 -05:00
/* do not keep unused FDs retrieved from the previous process */
sock_drop_unused_old_sockets();
usermsgs_clr(NULL);
setenv("HAPROXY_LOAD_SUCCESS", "0", 1);
ha_warning("Loading failure!\n");
#if defined(USE_SYSTEMD)
/* the sd_notify API is not able to send a reload failure signal. So
* the READY=1 signal still need to be sent */
if (global.tune.options & GTUNE_USE_SYSTEMD)
sd_notify(0, "READY=1\nSTATUS=Reload failed!\n");
#endif
mworker_reexec_waitmode();
}
/*
* Exit with an error message upon a wait-mode failure.
*/
void exit_on_waitmode_failure()
{
if (!atexit_flag)
return;
ha_alert("Non-recoverable mworker wait-mode error, exiting.\n");
}
/*
* upon SIGUSR1, let's have a soft stop. Note that soft_stop() broadcasts
* a signal zero to all subscribers. This means that it's as easy as
* subscribing to signal 0 to get informed about an imminent shutdown.
*/
static void sig_soft_stop(struct sig_handler *sh)
{
soft_stop();
signal_unregister_handler(sh);
pool_gc(NULL);
}
/*
* upon SIGTTOU, we pause everything
*/
static void sig_pause(struct sig_handler *sh)
{
if (protocol_pause_all() & ERR_FATAL) {
const char *msg = "Some proxies refused to pause, performing soft stop now.\n";
ha_warning("%s", msg);
send_log(NULL, LOG_WARNING, "%s", msg);
soft_stop();
}
pool_gc(NULL);
}
/*
* upon SIGTTIN, let's have a soft stop.
*/
static void sig_listen(struct sig_handler *sh)
{
if (protocol_resume_all() & ERR_FATAL) {
const char *msg = "Some proxies refused to resume, probably due to a conflict on a listening port. You may want to try again after the conflicting application is stopped, otherwise a restart might be needed to resume safe operations.\n";
ha_warning("%s", msg);
send_log(NULL, LOG_WARNING, "%s", msg);
}
}
/*
* this function dumps every server's state when the process receives SIGHUP.
*/
static void sig_dump_state(struct sig_handler *sh)
{
struct proxy *p = proxies_list;
ha_warning("SIGHUP received, dumping servers states.\n");
while (p) {
struct server *s = p->srv;
send_log(p, LOG_NOTICE, "SIGHUP received, dumping servers states for proxy %s.\n", p->id);
while (s) {
chunk_printf(&trash,
"SIGHUP: Server %s/%s is %s. Conn: %d act, %d pend, %lld tot.",
p->id, s->id,
(s->cur_state != SRV_ST_STOPPED) ? "UP" : "DOWN",
s->cur_sess, s->queue.length, s->counters.cum_sess);
ha_warning("%s\n", trash.area);
send_log(p, LOG_NOTICE, "%s\n", trash.area);
s = s->next;
}
/* FIXME: those info are a bit outdated. We should be able to distinguish between FE and BE. */
if (!p->srv) {
chunk_printf(&trash,
"SIGHUP: Proxy %s has no servers. Conn: act(FE+BE): %d+%d, %d pend (%d unass), tot(FE+BE): %lld+%lld.",
p->id,
p->feconn, p->beconn, p->totpend, p->queue.length, p->fe_counters.cum_conn, p->be_counters.cum_conn);
} else if (p->srv_act == 0) {
chunk_printf(&trash,
"SIGHUP: Proxy %s %s ! Conn: act(FE+BE): %d+%d, %d pend (%d unass), tot(FE+BE): %lld+%lld.",
p->id,
(p->srv_bck) ? "is running on backup servers" : "has no server available",
p->feconn, p->beconn, p->totpend, p->queue.length, p->fe_counters.cum_conn, p->be_counters.cum_conn);
} else {
chunk_printf(&trash,
"SIGHUP: Proxy %s has %d active servers and %d backup servers available."
" Conn: act(FE+BE): %d+%d, %d pend (%d unass), tot(FE+BE): %lld+%lld.",
p->id, p->srv_act, p->srv_bck,
p->feconn, p->beconn, p->totpend, p->queue.length, p->fe_counters.cum_conn, p->be_counters.cum_conn);
}
ha_warning("%s\n", trash.area);
send_log(p, LOG_NOTICE, "%s\n", trash.area);
p = p->next;
}
}
static void dump(struct sig_handler *sh)
{
/* dump memory usage then free everything possible */
dump_pools();
pool_gc(NULL);
}
/*
* This function dup2 the stdio FDs (0,1,2) with <fd>, then closes <fd>
* If <fd> < 0, it opens /dev/null and use it to dup
*
* In the case of chrooting, you have to open /dev/null before the chroot, and
* pass the <fd> to this function
*/
static void stdio_quiet(int fd)
{
if (fd < 0)
fd = open("/dev/null", O_RDWR, 0);
if (fd > -1) {
fclose(stdin);
fclose(stdout);
fclose(stderr);
dup2(fd, 0);
dup2(fd, 1);
dup2(fd, 2);
if (fd > 2)
close(fd);
return;
}
ha_alert("Cannot open /dev/null\n");
exit(EXIT_FAILURE);
}
/* This function checks if cfg_cfgfiles contains directories.
* If it finds one, it adds all the files (and only files) it contains
* in cfg_cfgfiles in place of the directory (and removes the directory).
* It adds the files in lexical order.
* It adds only files with .cfg extension.
* It doesn't add files with name starting with '.'
*/
static void cfgfiles_expand_directories(void)
{
struct wordlist *wl, *wlb;
char *err = NULL;
list_for_each_entry_safe(wl, wlb, &cfg_cfgfiles, list) {
struct stat file_stat;
struct dirent **dir_entries = NULL;
int dir_entries_nb;
int dir_entries_it;
if (stat(wl->s, &file_stat)) {
ha_alert("Cannot open configuration file/directory %s : %s\n",
wl->s,
strerror(errno));
exit(1);
}
if (!S_ISDIR(file_stat.st_mode))
continue;
/* from this point wl->s is a directory */
dir_entries_nb = scandir(wl->s, &dir_entries, NULL, alphasort);
if (dir_entries_nb < 0) {
ha_alert("Cannot open configuration directory %s : %s\n",
wl->s,
strerror(errno));
exit(1);
}
/* for each element in the directory wl->s */
for (dir_entries_it = 0; dir_entries_it < dir_entries_nb; dir_entries_it++) {
struct dirent *dir_entry = dir_entries[dir_entries_it];
char *filename = NULL;
char *d_name_cfgext = strstr(dir_entry->d_name, ".cfg");
/* don't add filename that begin with .
* only add filename with .cfg extension
*/
if (dir_entry->d_name[0] == '.' ||
!(d_name_cfgext && d_name_cfgext[4] == '\0'))
goto next_dir_entry;
if (!memprintf(&filename, "%s/%s", wl->s, dir_entry->d_name)) {
ha_alert("Cannot load configuration files %s : out of memory.\n",
filename);
exit(1);
}
if (stat(filename, &file_stat)) {
ha_alert("Cannot open configuration file %s : %s\n",
wl->s,
strerror(errno));
exit(1);
}
/* don't add anything else than regular file in cfg_cfgfiles
* this way we avoid loops
*/
if (!S_ISREG(file_stat.st_mode))
goto next_dir_entry;
if (!list_append_word(&wl->list, filename, &err)) {
ha_alert("Cannot load configuration files %s : %s\n",
filename,
err);
exit(1);
}
next_dir_entry:
free(filename);
free(dir_entry);
}
free(dir_entries);
/* remove the current directory (wl) from cfg_cfgfiles */
free(wl->s);
LIST_DELETE(&wl->list);
free(wl);
}
free(err);
}
/*
* copy and cleanup the current argv
* Remove the -sf /-st / -x parameters
* Return an allocated copy of argv
*/
static char **copy_argv(int argc, char **argv)
{
char **newargv, **retargv;
newargv = calloc(argc + 2, sizeof(*newargv));
if (newargv == NULL) {
ha_warning("Cannot allocate memory\n");
return NULL;
}
retargv = newargv;
/* first copy argv[0] */
*newargv++ = *argv++;
argc--;
while (argc > 0) {
if (**argv != '-') {
/* non options are copied but will fail in the argument parser */
*newargv++ = *argv++;
argc--;
} else {
char *flag;
flag = *argv + 1;
if (flag[0] == '-' && flag[1] == 0) {
/* "--\0" copy every arguments till the end of argv */
*newargv++ = *argv++;
argc--;
while (argc > 0) {
*newargv++ = *argv++;
argc--;
}
} else {
switch (*flag) {
case 's':
/* -sf / -st and their parameters are ignored */
if (flag[1] == 'f' || flag[1] == 't') {
argc--;
argv++;
/* The list can't contain a negative value since the only
way to know the end of this list is by looking for the
next option or the end of the options */
while (argc > 0 && argv[0][0] != '-') {
argc--;
argv++;
}
} else {
argc--;
argv++;
}
break;
case 'x':
/* this option and its parameter are ignored */
argc--;
argv++;
if (argc > 0) {
argc--;
argv++;
}
break;
case 'C':
case 'n':
case 'm':
case 'N':
case 'L':
case 'f':
case 'p':
case 'S':
/* these options have only 1 parameter which must be copied and can start with a '-' */
*newargv++ = *argv++;
argc--;
if (argc == 0)
goto error;
*newargv++ = *argv++;
argc--;
break;
default:
/* for other options just copy them without parameters, this is also done
* for options like "--foo", but this will fail in the argument parser.
* */
*newargv++ = *argv++;
argc--;
break;
}
}
}
}
return retargv;
error:
free(retargv);
return NULL;
}
/* Performs basic random seed initialization. The main issue with this is that
* srandom_r() only takes 32 bits and purposely provides a reproducible sequence,
* which means that there will only be 4 billion possible random sequences once
* srandom() is called, regardless of the internal state. Not calling it is
* even worse as we'll always produce the same randoms sequences. What we do
* here is to create an initial sequence from various entropy sources, hash it
* using SHA1 and keep the resulting 160 bits available globally.
*
* We initialize the current process with the first 32 bits before starting the
* polling loop, where all this will be changed to have process specific and
* thread specific sequences.
BUG/MEDIUM: random: implement a thread-safe and process-safe PRNG This is the replacement of failed attempt to add thread safety and per-process sequences of random numbers initally tried with commit 1c306aa84d ("BUG/MEDIUM: random: implement per-thread and per-process random sequences"). This new version takes a completely different approach and doesn't try to work around the horrible OS-specific and non-portable random API anymore. Instead it implements "xoroshiro128**", a reputedly high quality random number generator, which is one of the many variants of xorshift, which passes all quality tests and which is described here: http://prng.di.unimi.it/ While not cryptographically secure, it is fast and features a 2^128-1 period. It supports fast jumps allowing to cut the period into smaller non-overlapping sequences, which we use here to support up to 2^32 processes each having their own, non-overlapping sequence of 2^96 numbers (~7*10^28). This is enough to provide 1 billion randoms per second and per process for 2200 billion years. The implementation was made thread-safe either by using a double 64-bit CAS on platforms supporting it (x86_64, aarch64) or by using a local lock for the time needed to perform the shift operations. This ensures that all threads pick numbers from the same pool so that it is not needed to assign per-thread ranges. For processes we use the fast jump method to advance the sequence by 2^96 for each process. Before this patch, the following config: global nbproc 8 frontend f bind :4445 mode http log stdout format raw daemon log-format "%[uuid] %pid" redirect location / Would produce this output: a4d0ad64-2645-4b74-b894-48acce0669af 12987 a4d0ad64-2645-4b74-b894-48acce0669af 12992 a4d0ad64-2645-4b74-b894-48acce0669af 12986 a4d0ad64-2645-4b74-b894-48acce0669af 12988 a4d0ad64-2645-4b74-b894-48acce0669af 12991 a4d0ad64-2645-4b74-b894-48acce0669af 12989 a4d0ad64-2645-4b74-b894-48acce0669af 12990 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12987 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12992 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12986 (...) And now produces: f94b29b3-da74-4e03-a0c5-a532c635bad9 13011 47470c02-4862-4c33-80e7-a952899570e5 13014 86332123-539a-47bf-853f-8c8ea8b2a2b5 13013 8f9efa99-3143-47b2-83cf-d618c8dea711 13012 3cc0f5c7-d790-496b-8d39-bec77647af5b 13015 3ec64915-8f95-4374-9e66-e777dc8791e0 13009 0f9bf894-dcde-408c-b094-6e0bb3255452 13011 49c7bfde-3ffb-40e9-9a8d-8084d650ed8f 13014 e23f6f2e-35c5-4433-a294-b790ab902653 13012 There are multiple benefits to using this method. First, it doesn't depend anymore on a non-portable API. Second it's thread safe. Third it is fast and more proven than any hack we could attempt to try to work around the deficiencies of the various implementations around. This commit depends on previous patches "MINOR: tools: add 64-bit rotate operators" and "BUG/MEDIUM: random: initialize the random pool a bit better", all of which will need to be backported at least as far as version 2.0. It doesn't require to backport the build fixes for circular include files dependecy anymore.
2020-03-07 18:42:37 -05:00
*
* Before starting threads, it's still possible to call random() as srandom()
* is initialized from this, but after threads and/or processes are started,
* only ha_random() is expected to be used to guarantee distinct sequences.
*/
static void ha_random_boot(char *const *argv)
{
unsigned char message[256];
unsigned char *m = message;
struct timeval tv;
blk_SHA_CTX ctx;
unsigned long l;
int fd;
int i;
/* start with current time as pseudo-random seed */
gettimeofday(&tv, NULL);
write_u32(m, tv.tv_sec); m += 4;
write_u32(m, tv.tv_usec); m += 4;
/* PID and PPID add some OS-based randomness */
write_u16(m, getpid()); m += 2;
write_u16(m, getppid()); m += 2;
/* take up to 160 bits bytes from /dev/urandom if available (non-blocking) */
fd = open("/dev/urandom", O_RDONLY);
if (fd >= 0) {
i = read(fd, m, 20);
if (i > 0)
m += i;
close(fd);
}
/* take up to 160 bits bytes from openssl (non-blocking) */
#ifdef USE_OPENSSL
if (RAND_bytes(m, 20) == 1)
m += 20;
#endif
/* take 160 bits from existing random in case it was already initialized */
for (i = 0; i < 5; i++) {
write_u32(m, random());
m += 4;
}
/* stack address (benefit form operating system's ASLR) */
l = (unsigned long)&m;
memcpy(m, &l, sizeof(l)); m += sizeof(l);
/* argv address (benefit form operating system's ASLR) */
l = (unsigned long)&argv;
memcpy(m, &l, sizeof(l)); m += sizeof(l);
/* use tv_usec again after all the operations above */
gettimeofday(&tv, NULL);
write_u32(m, tv.tv_usec); m += 4;
/*
* At this point, ~84-92 bytes have been used
*/
/* finish with the hostname */
strncpy((char *)m, hostname, message + sizeof(message) - m);
m += strlen(hostname);
/* total message length */
l = m - message;
memset(&ctx, 0, sizeof(ctx));
blk_SHA1_Init(&ctx);
blk_SHA1_Update(&ctx, message, l);
blk_SHA1_Final(boot_seed, &ctx);
srandom(read_u32(boot_seed));
BUG/MEDIUM: random: implement a thread-safe and process-safe PRNG This is the replacement of failed attempt to add thread safety and per-process sequences of random numbers initally tried with commit 1c306aa84d ("BUG/MEDIUM: random: implement per-thread and per-process random sequences"). This new version takes a completely different approach and doesn't try to work around the horrible OS-specific and non-portable random API anymore. Instead it implements "xoroshiro128**", a reputedly high quality random number generator, which is one of the many variants of xorshift, which passes all quality tests and which is described here: http://prng.di.unimi.it/ While not cryptographically secure, it is fast and features a 2^128-1 period. It supports fast jumps allowing to cut the period into smaller non-overlapping sequences, which we use here to support up to 2^32 processes each having their own, non-overlapping sequence of 2^96 numbers (~7*10^28). This is enough to provide 1 billion randoms per second and per process for 2200 billion years. The implementation was made thread-safe either by using a double 64-bit CAS on platforms supporting it (x86_64, aarch64) or by using a local lock for the time needed to perform the shift operations. This ensures that all threads pick numbers from the same pool so that it is not needed to assign per-thread ranges. For processes we use the fast jump method to advance the sequence by 2^96 for each process. Before this patch, the following config: global nbproc 8 frontend f bind :4445 mode http log stdout format raw daemon log-format "%[uuid] %pid" redirect location / Would produce this output: a4d0ad64-2645-4b74-b894-48acce0669af 12987 a4d0ad64-2645-4b74-b894-48acce0669af 12992 a4d0ad64-2645-4b74-b894-48acce0669af 12986 a4d0ad64-2645-4b74-b894-48acce0669af 12988 a4d0ad64-2645-4b74-b894-48acce0669af 12991 a4d0ad64-2645-4b74-b894-48acce0669af 12989 a4d0ad64-2645-4b74-b894-48acce0669af 12990 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12987 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12992 82d5f6cd-f6c1-4f85-a89c-36ae85d26fb9 12986 (...) And now produces: f94b29b3-da74-4e03-a0c5-a532c635bad9 13011 47470c02-4862-4c33-80e7-a952899570e5 13014 86332123-539a-47bf-853f-8c8ea8b2a2b5 13013 8f9efa99-3143-47b2-83cf-d618c8dea711 13012 3cc0f5c7-d790-496b-8d39-bec77647af5b 13015 3ec64915-8f95-4374-9e66-e777dc8791e0 13009 0f9bf894-dcde-408c-b094-6e0bb3255452 13011 49c7bfde-3ffb-40e9-9a8d-8084d650ed8f 13014 e23f6f2e-35c5-4433-a294-b790ab902653 13012 There are multiple benefits to using this method. First, it doesn't depend anymore on a non-portable API. Second it's thread safe. Third it is fast and more proven than any hack we could attempt to try to work around the deficiencies of the various implementations around. This commit depends on previous patches "MINOR: tools: add 64-bit rotate operators" and "BUG/MEDIUM: random: initialize the random pool a bit better", all of which will need to be backported at least as far as version 2.0. It doesn't require to backport the build fixes for circular include files dependecy anymore.
2020-03-07 18:42:37 -05:00
ha_random_seed(boot_seed, sizeof(boot_seed));
}
/* considers splicing proxies' maxconn, computes the ideal global.maxpipes
* setting, and returns it. It may return -1 meaning "unlimited" if some
* unlimited proxies have been found and the global.maxconn value is not yet
* set. It may also return a value greater than maxconn if it's not yet set.
* Note that a value of zero means there is no need for pipes. -1 is never
* returned if global.maxconn is valid.
*/
static int compute_ideal_maxpipes()
{
struct proxy *cur;
int nbfe = 0, nbbe = 0;
int unlimited = 0;
int pipes;
int max;
for (cur = proxies_list; cur; cur = cur->next) {
if (cur->options2 & (PR_O2_SPLIC_ANY)) {
if (cur->cap & PR_CAP_FE) {
max = cur->maxconn;
nbfe += max;
if (!max) {
unlimited = 1;
break;
}
}
if (cur->cap & PR_CAP_BE) {
max = cur->fullconn ? cur->fullconn : global.maxconn;
nbbe += max;
if (!max) {
unlimited = 1;
break;
}
}
}
}
pipes = MAX(nbfe, nbbe);
if (global.maxconn) {
if (pipes > global.maxconn || unlimited)
pipes = global.maxconn;
} else if (unlimited) {
pipes = -1;
}
return pipes >= 4 ? pipes / 4 : pipes;
}
/* considers global.maxsocks, global.maxpipes, async engines, SSL frontends and
* rlimits and computes an ideal maxconn. It's meant to be called only when
* maxsock contains the sum of listening FDs, before it is updated based on
* maxconn and pipes. If there are not enough FDs left, DEFAULT_MAXCONN (by
* default 100) is returned as it is expected that it will even run on tight
* environments, and will maintain compatibility with previous packages that
* used to rely on this value as the default one. The system will emit a
* warning indicating how many FDs are missing anyway if needed.
*/
static int compute_ideal_maxconn()
{
int ssl_sides = !!global.ssl_used_frontend + !!global.ssl_used_backend;
int engine_fds = global.ssl_used_async_engines * ssl_sides;
int pipes = compute_ideal_maxpipes();
BUG/MINOR: init: make the automatic maxconn consider the max of soft/hard limits James Stroehmann reported something working as documented but that can be considered as a regression in the way the automatic maxconn is calculated from the process' limits : https://www.mail-archive.com/haproxy@formilux.org/msg36523.html The purpose of the changes in 2.0 was to have maxconn default to the highest possible value permitted to the user based on the ulimit -n setting, however the calculation starts from the soft limit, which can be lower than what users were allowed to with previous versions where the default value of 2000 would force a higher ulimit -n as long as it fitted in the hard limit. Usually this is not noticeable if the user changes the limits, because quite commonly setting a new value restricts both the soft and hard values. Let's instead always use the max between the hard and soft limits, as we know these values are permitted. This was tried on the following setup: $ cat ulimit-n.cfg global stats socket /tmp/sock1 level admin $ ulimit -n 1024 Before the change the limits would show like this: $ socat - /tmp/sock1 <<< "show info" | grep -im2 ^Max Maxsock: 1023 Maxconn: 489 After the change the limits are now much better and more in line with the default settings in earlier versions: $ socat - /tmp/sock1 <<< "show info" | grep -im2 ^Max Maxsock: 4095 Maxconn: 2025 The difference becomes even more obvious when running moderately large configs with hundreds of checked servers and hundreds of listeners: $ cat ulimit-n.cfg global stats socket /tmp/sock1 level admin listen l bind :10000-10300 server-template srv- 300 0.0.0.0 check disabled Before After Maxsock 1024 4096 Maxconn 189 1725 This issue is tagged as minor since a trivial config change fixes it, but it would help new users to have it backported as far as 2.0.
2020-03-06 04:25:31 -05:00
int remain = MAX(rlim_fd_cur_at_boot, rlim_fd_max_at_boot);
int maxconn;
/* we have to take into account these elements :
* - number of engine_fds, which inflates the number of FD needed per
* connection by this number.
* - number of pipes per connection on average : for the unlimited
* case, this is 0.5 pipe FDs per connection, otherwise it's a
* fixed value of 2*pipes.
* - two FDs per connection
*/
if (global.fd_hard_limit && remain > global.fd_hard_limit)
remain = global.fd_hard_limit;
/* subtract listeners and checks */
remain -= global.maxsock;
/* one epoll_fd/kqueue_fd per thread */
remain -= global.nbthread;
/* one wake-up pipe (2 fd) per thread */
remain -= 2 * global.nbthread;
/* Fixed pipes values : we only subtract them if they're not larger
* than the remaining FDs because pipes are optional.
*/
if (pipes >= 0 && pipes * 2 < remain)
remain -= pipes * 2;
if (pipes < 0) {
/* maxsock = maxconn * 2 + maxconn/4 * 2 + maxconn * engine_fds.
* = maxconn * (2 + 0.5 + engine_fds)
* = maxconn * (4 + 1 + 2*engine_fds) / 2
*/
maxconn = 2 * remain / (5 + 2 * engine_fds);
} else {
/* maxsock = maxconn * 2 + maxconn * engine_fds.
* = maxconn * (2 + engine_fds)
*/
maxconn = remain / (2 + engine_fds);
}
return MAX(maxconn, DEFAULT_MAXCONN);
}
/* computes the estimated maxsock value for the given maxconn based on the
* possibly set global.maxpipes and existing partial global.maxsock. It may
* temporarily change global.maxconn for the time needed to propagate the
* computations, and will reset it.
*/
static int compute_ideal_maxsock(int maxconn)
{
int maxpipes = global.maxpipes;
int maxsock = global.maxsock;
if (!maxpipes) {
int old_maxconn = global.maxconn;
global.maxconn = maxconn;
maxpipes = compute_ideal_maxpipes();
global.maxconn = old_maxconn;
}
maxsock += maxconn * 2; /* each connection needs two sockets */
maxsock += maxpipes * 2; /* each pipe needs two FDs */
maxsock += global.nbthread; /* one epoll_fd/kqueue_fd per thread */
maxsock += 2 * global.nbthread; /* one wake-up pipe (2 fd) per thread */
/* compute fd used by async engines */
if (global.ssl_used_async_engines) {
int sides = !!global.ssl_used_frontend + !!global.ssl_used_backend;
maxsock += maxconn * sides * global.ssl_used_async_engines;
}
return maxsock;
}
/* Tests if it is possible to set the current process's RLIMIT_NOFILE to
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
* <maxsock>, then sets it back to the previous value. Returns non-zero if the
* value is accepted, non-zero otherwise. This is used to determine if an
* automatic limit may be applied or not. When it is not, the caller knows that
* the highest we can do is the rlim_max at boot. In case of error, we return
* that the setting is possible, so that we defer the error processing to the
* final stage in charge of enforcing this.
*/
static int check_if_maxsock_permitted(int maxsock)
{
struct rlimit orig_limit, test_limit;
int ret;
if (global.fd_hard_limit && maxsock > global.fd_hard_limit)
return 0;
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
if (getrlimit(RLIMIT_NOFILE, &orig_limit) != 0)
return 1;
/* don't go further if we can't even set to what we have */
if (raise_rlim_nofile(NULL, &orig_limit) != 0)
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
return 1;
test_limit.rlim_max = MAX(maxsock, orig_limit.rlim_max);
test_limit.rlim_cur = test_limit.rlim_max;
ret = raise_rlim_nofile(NULL, &test_limit);
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
if (raise_rlim_nofile(NULL, &orig_limit) != 0)
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
return 1;
return ret == 0;
}
/* This performs th every basic early initialization at the end of the PREPARE
* init stage. It may only assume that list heads are initialized, but not that
* anything else is correct. It will initialize a number of variables that
* depend on command line and will pre-parse the command line. If it fails, it
* directly exits.
*/
static void init_early(int argc, char **argv)
{
char *progname;
char *tmp;
int len;
setenv("HAPROXY_STARTUP_VERSION", HAPROXY_VERSION, 0);
/* First, let's initialize most global variables */
totalconn = actconn = listeners = stopping = 0;
killed = pid = 0;
global.maxsock = 10; /* reserve 10 fds ; will be incremented by socket eaters */
global.rlimit_memmax_all = HAPROXY_MEMMAX;
global.mode = MODE_STARTING;
/* if we were in mworker mode, we should restart in mworker mode */
if (getenv("HAPROXY_MWORKER_REEXEC") != NULL)
global.mode |= MODE_MWORKER;
/* initialize date, time, and pid */
tzset();
clock_init_process_date();
2023-02-07 09:52:14 -05:00
start_date = date;
start_time_ns = now_ns;
pid = getpid();
/* Set local host name and adjust some environment variables.
* NB: POSIX does not make it mandatory for gethostname() to
* NULL-terminate the string in case of truncation, and at least
* FreeBSD appears not to do it.
*/
memset(hostname, 0, sizeof(hostname));
gethostname(hostname, sizeof(hostname) - 1);
/* preset some environment variables */
localpeer = strdup(hostname);
if (!localpeer || setenv("HAPROXY_LOCALPEER", localpeer, 1) < 0) {
ha_alert("Cannot allocate memory for local peer.\n");
exit(EXIT_FAILURE);
}
/* extract the program name from argv[0], it will be used for the logs
* and error messages.
*/
progname = *argv;
while ((tmp = strchr(progname, '/')) != NULL)
progname = tmp + 1;
len = strlen(progname);
progname = strdup(progname);
if (!progname) {
ha_alert("Cannot allocate memory for log_tag.\n");
exit(EXIT_FAILURE);
}
chunk_initlen(&global.log_tag, progname, len, len);
}
/* handles program arguments. Very minimal parsing is performed, variables are
* fed with some values, and lists are completed with other ones. In case of
* error, it will exit.
*/
static void init_args(int argc, char **argv)
{
char *progname = global.log_tag.area;
char *err_msg = NULL;
/* pre-fill in the global tuning options before we let the cmdline
* change them.
*/
global.tune.options |= GTUNE_USE_SELECT; /* select() is always available */
#if defined(USE_POLL)
global.tune.options |= GTUNE_USE_POLL;
#endif
#if defined(USE_EPOLL)
global.tune.options |= GTUNE_USE_EPOLL;
#endif
#if defined(USE_KQUEUE)
global.tune.options |= GTUNE_USE_KQUEUE;
#endif
#if defined(USE_EVPORTS)
global.tune.options |= GTUNE_USE_EVPORTS;
#endif
#if defined(USE_LINUX_SPLICE)
global.tune.options |= GTUNE_USE_SPLICE;
#endif
#if defined(USE_GETADDRINFO)
global.tune.options |= GTUNE_USE_GAI;
#endif
#ifdef USE_THREAD
global.tune.options |= GTUNE_IDLE_POOL_SHARED;
#endif
#ifdef USE_QUIC
global.tune.options |= GTUNE_QUIC_SOCK_PER_CONN;
#endif
global.tune.options |= GTUNE_STRICT_LIMITS;
global.tune.options |= GTUNE_USE_FAST_FWD; /* Use fast-forward by default */
/* Use zero-copy forwarding by default */
global.tune.no_zero_copy_fwd = 0;
/* keep a copy of original arguments for the master process */
old_argv = copy_argv(argc, argv);
if (!old_argv) {
ha_alert("failed to copy argv.\n");
exit(EXIT_FAILURE);
}
/* skip program name and start */
argc--; argv++;
while (argc > 0) {
char *flag;
if (**argv == '-') {
flag = *argv+1;
/* 1 arg */
if (*flag == 'v') {
display_version();
if (flag[1] == 'v') /* -vv */
display_build_opts();
deinit_and_exit(0);
}
#if defined(USE_EPOLL)
else if (*flag == 'd' && flag[1] == 'e')
global.tune.options &= ~GTUNE_USE_EPOLL;
#endif
#if defined(USE_POLL)
else if (*flag == 'd' && flag[1] == 'p')
global.tune.options &= ~GTUNE_USE_POLL;
#endif
#if defined(USE_KQUEUE)
else if (*flag == 'd' && flag[1] == 'k')
global.tune.options &= ~GTUNE_USE_KQUEUE;
#endif
#if defined(USE_EVPORTS)
else if (*flag == 'd' && flag[1] == 'v')
global.tune.options &= ~GTUNE_USE_EVPORTS;
#endif
#if defined(USE_LINUX_SPLICE)
else if (*flag == 'd' && flag[1] == 'S')
global.tune.options &= ~GTUNE_USE_SPLICE;
#endif
#if defined(USE_GETADDRINFO)
else if (*flag == 'd' && flag[1] == 'G')
global.tune.options &= ~GTUNE_USE_GAI;
#endif
#if defined(SO_REUSEPORT)
else if (*flag == 'd' && flag[1] == 'R')
protocol_clrf_all(PROTO_F_REUSEPORT_SUPPORTED);
#endif
else if (*flag == 'd' && flag[1] == 'F')
global.tune.options &= ~GTUNE_USE_FAST_FWD;
else if (*flag == 'd' && flag[1] == 'V')
global.ssl_server_verify = SSL_SERVER_VERIFY_NONE;
else if (*flag == 'd' && flag[1] == 'Z')
global.tune.no_zero_copy_fwd |= NO_ZERO_COPY_FWD;
else if (*flag == 'V')
arg_mode |= MODE_VERBOSE;
else if (*flag == 'd' && flag[1] == 'C') {
char *end;
char *key;
key = flag + 2;
for (;key && *key; key = end) {
end = strchr(key, ',');
if (end)
*(end++) = 0;
if (strcmp(key, "line") == 0)
arg_mode |= MODE_DUMP_NB_L;
}
arg_mode |= MODE_DUMP_CFG;
HA_ATOMIC_STORE(&global.anon_key, atoll(flag + 2));
}
else if (*flag == 'd' && flag[1] == 'b')
arg_mode |= MODE_FOREGROUND;
else if (*flag == 'd' && flag[1] == 'D')
arg_mode |= MODE_DIAG;
else if (*flag == 'd' && flag[1] == 'W')
arg_mode |= MODE_ZERO_WARNING;
else if (*flag == 'd' && flag[1] == 'M') {
int ret = pool_parse_debugging(flag + 2, &err_msg);
if (ret <= -1) {
if (ret < -1)
ha_alert("-dM: %s\n", err_msg);
else
printf("%s\n", err_msg);
ha_free(&err_msg);
exit(ret < -1 ? EXIT_FAILURE : 0);
} else if (ret == 0) {
ha_warning("-dM: %s\n", err_msg);
ha_free(&err_msg);
}
}
else if (*flag == 'd' && flag[1] == 'r')
global.tune.options |= GTUNE_RESOLVE_DONTFAIL;
#if defined(HA_HAVE_DUMP_LIBS)
else if (*flag == 'd' && flag[1] == 'L')
arg_mode |= MODE_DUMP_LIBS;
#endif
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
else if (*flag == 'd' && flag[1] == 'K') {
arg_mode |= MODE_DUMP_KWD;
kwd_dump = flag + 2;
}
else if (*flag == 'd' && flag[1] == 't') {
if (argc > 1 && argv[1][0] != '-') {
if (trace_parse_cmd(argv[1], &err_msg)) {
ha_alert("-dt: %s.\n", err_msg);
ha_free(&err_msg);
exit(EXIT_FAILURE);
}
argc--; argv++;
}
else {
trace_parse_cmd(NULL, NULL);
}
}
else if (*flag == 'd')
arg_mode |= MODE_DEBUG;
else if (*flag == 'c' && flag[1] == 'c') {
arg_mode |= MODE_CHECK_CONDITION;
argv++;
argc--;
check_condition = *argv;
}
else if (*flag == 'c')
arg_mode |= MODE_CHECK;
else if (*flag == 'D')
arg_mode |= MODE_DAEMON;
MEDIUM: mworker: Add systemd `Type=notify` support This patch adds support for `Type=notify` to the systemd unit. Supporting `Type=notify` improves both starting as well as reloading of the unit, because systemd will be let known when the action completed. See this quote from `systemd.service(5)`: > Note however that reloading a daemon by sending a signal (as with the > example line above) is usually not a good choice, because this is an > asynchronous operation and hence not suitable to order reloads of > multiple services against each other. It is strongly recommended to > set ExecReload= to a command that not only triggers a configuration > reload of the daemon, but also synchronously waits for it to complete. By making systemd aware of a reload in progress it is able to wait until the reload actually succeeded. This patch introduces both a new `USE_SYSTEMD` build option which controls including the sd-daemon library as well as a `-Ws` runtime option which runs haproxy in master-worker mode with systemd support. When haproxy is running in master-worker mode with systemd support it will send status messages to systemd using `sd_notify(3)` in the following cases: - The master process forked off the worker processes (READY=1) - The master process entered the `mworker_reload()` function (RELOADING=1) - The master process received the SIGUSR1 or SIGTERM signal (STOPPING=1) Change the unit file to specify `Type=notify` and replace master-worker mode (`-W`) with master-worker mode with systemd support (`-Ws`). Future evolutions of this feature could include making use of the `STATUS` feature of `sd_notify()` to send information about the number of active connections to systemd. This would require bidirectional communication between the master and the workers and thus is left for future work.
2017-11-20 09:58:35 -05:00
else if (*flag == 'W' && flag[1] == 's') {
arg_mode |= MODE_MWORKER | MODE_FOREGROUND;
MEDIUM: mworker: Add systemd `Type=notify` support This patch adds support for `Type=notify` to the systemd unit. Supporting `Type=notify` improves both starting as well as reloading of the unit, because systemd will be let known when the action completed. See this quote from `systemd.service(5)`: > Note however that reloading a daemon by sending a signal (as with the > example line above) is usually not a good choice, because this is an > asynchronous operation and hence not suitable to order reloads of > multiple services against each other. It is strongly recommended to > set ExecReload= to a command that not only triggers a configuration > reload of the daemon, but also synchronously waits for it to complete. By making systemd aware of a reload in progress it is able to wait until the reload actually succeeded. This patch introduces both a new `USE_SYSTEMD` build option which controls including the sd-daemon library as well as a `-Ws` runtime option which runs haproxy in master-worker mode with systemd support. When haproxy is running in master-worker mode with systemd support it will send status messages to systemd using `sd_notify(3)` in the following cases: - The master process forked off the worker processes (READY=1) - The master process entered the `mworker_reload()` function (RELOADING=1) - The master process received the SIGUSR1 or SIGTERM signal (STOPPING=1) Change the unit file to specify `Type=notify` and replace master-worker mode (`-W`) with master-worker mode with systemd support (`-Ws`). Future evolutions of this feature could include making use of the `STATUS` feature of `sd_notify()` to send information about the number of active connections to systemd. This would require bidirectional communication between the master and the workers and thus is left for future work.
2017-11-20 09:58:35 -05:00
#if defined(USE_SYSTEMD)
global.tune.options |= GTUNE_USE_SYSTEMD;
#else
ha_alert("master-worker mode with systemd support (-Ws) requested, but not compiled. Use master-worker mode (-W) if you are not using Type=notify in your unit file or recompile with USE_SYSTEMD=1.\n\n");
MEDIUM: mworker: Add systemd `Type=notify` support This patch adds support for `Type=notify` to the systemd unit. Supporting `Type=notify` improves both starting as well as reloading of the unit, because systemd will be let known when the action completed. See this quote from `systemd.service(5)`: > Note however that reloading a daemon by sending a signal (as with the > example line above) is usually not a good choice, because this is an > asynchronous operation and hence not suitable to order reloads of > multiple services against each other. It is strongly recommended to > set ExecReload= to a command that not only triggers a configuration > reload of the daemon, but also synchronously waits for it to complete. By making systemd aware of a reload in progress it is able to wait until the reload actually succeeded. This patch introduces both a new `USE_SYSTEMD` build option which controls including the sd-daemon library as well as a `-Ws` runtime option which runs haproxy in master-worker mode with systemd support. When haproxy is running in master-worker mode with systemd support it will send status messages to systemd using `sd_notify(3)` in the following cases: - The master process forked off the worker processes (READY=1) - The master process entered the `mworker_reload()` function (RELOADING=1) - The master process received the SIGUSR1 or SIGTERM signal (STOPPING=1) Change the unit file to specify `Type=notify` and replace master-worker mode (`-W`) with master-worker mode with systemd support (`-Ws`). Future evolutions of this feature could include making use of the `STATUS` feature of `sd_notify()` to send information about the number of active connections to systemd. This would require bidirectional communication between the master and the workers and thus is left for future work.
2017-11-20 09:58:35 -05:00
usage(progname);
#endif
}
else if (*flag == 'W')
arg_mode |= MODE_MWORKER;
else if (*flag == 'q')
arg_mode |= MODE_QUIET;
else if (*flag == 'x') {
if (argc <= 1) {
ha_alert("Unix socket path expected with the -x flag\n\n");
usage(progname);
}
if (old_unixsocket)
ha_warning("-x option already set, overwriting the value\n");
old_unixsocket = argv[1];
argv++;
argc--;
}
else if (*flag == 'S') {
struct wordlist *c;
if (argc <= 1) {
ha_alert("Socket and optional bind parameters expected with the -S flag\n");
usage(progname);
}
if ((c = malloc(sizeof(*c))) == NULL || (c->s = strdup(argv[1])) == NULL) {
ha_alert("Cannot allocate memory\n");
exit(EXIT_FAILURE);
}
LIST_INSERT(&mworker_cli_conf, &c->list);
argv++;
argc--;
}
else if (*flag == 's' && (flag[1] == 'f' || flag[1] == 't')) {
/* list of pids to finish ('f') or terminate ('t') */
if (flag[1] == 'f')
oldpids_sig = SIGUSR1; /* finish then exit */
else
oldpids_sig = SIGTERM; /* terminate immediately */
while (argc > 1 && argv[1][0] != '-') {
char * endptr = NULL;
oldpids = realloc(oldpids, (nb_oldpids + 1) * sizeof(int));
if (!oldpids) {
ha_alert("Cannot allocate old pid : out of memory.\n");
exit(1);
}
argc--; argv++;
errno = 0;
oldpids[nb_oldpids] = strtol(*argv, &endptr, 10);
if (errno) {
ha_alert("-%2s option: failed to parse {%s}: %s\n",
flag,
*argv, strerror(errno));
exit(1);
} else if (endptr && strlen(endptr)) {
while (isspace((unsigned char)*endptr)) endptr++;
if (*endptr != 0) {
ha_alert("-%2s option: some bytes unconsumed in PID list {%s}\n",
flag, endptr);
exit(1);
}
}
if (oldpids[nb_oldpids] <= 0)
usage(progname);
nb_oldpids++;
}
}
else if (flag[0] == '-' && flag[1] == 0) { /* "--" */
/* now that's a cfgfile list */
argv++; argc--;
while (argc > 0) {
if (!list_append_word(&cfg_cfgfiles, *argv, &err_msg)) {
ha_alert("Cannot load configuration file/directory %s : %s\n",
*argv,
err_msg);
exit(1);
}
argv++; argc--;
}
break;
}
else { /* >=2 args */
argv++; argc--;
if (argc == 0)
usage(progname);
switch (*flag) {
case 'C' : change_dir = *argv; break;
case 'n' : cfg_maxconn = atol(*argv); break;
case 'm' : global.rlimit_memmax_all = atol(*argv); break;
case 'N' : cfg_maxpconn = atol(*argv); break;
case 'L' :
free(localpeer);
if ((localpeer = strdup(*argv)) == NULL) {
ha_alert("Cannot allocate memory for local peer.\n");
exit(EXIT_FAILURE);
}
setenv("HAPROXY_LOCALPEER", localpeer, 1);
global.localpeer_cmdline = 1;
break;
case 'f' :
if (!list_append_word(&cfg_cfgfiles, *argv, &err_msg)) {
ha_alert("Cannot load configuration file/directory %s : %s\n",
*argv,
err_msg);
exit(1);
}
break;
case 'p' :
free(global.pidfile);
if ((global.pidfile = strdup(*argv)) == NULL) {
ha_alert("Cannot allocate memory for pidfile.\n");
exit(EXIT_FAILURE);
}
break;
default: usage(progname);
}
}
}
else
usage(progname);
argv++; argc--;
}
free(err_msg);
}
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
/* call the various keyword dump functions based on the comma-delimited list of
* classes in kwd_dump.
*/
static void dump_registered_keywords(void)
{
char *end;
int all __maybe_unused = 0;
for (; kwd_dump && *kwd_dump; kwd_dump = end) {
end = strchr(kwd_dump, ',');
if (end)
*(end++) = 0;
if (strcmp(kwd_dump, "help") == 0) {
printf("# List of supported keyword classes:\n");
printf("all: list all keywords\n");
printf("acl: ACL keywords\n");
printf("cfg: configuration keywords\n");
printf("cli: CLI keywords\n");
printf("cnv: sample converter keywords\n");
printf("flt: filter names\n");
printf("smp: sample fetch functions\n");
printf("svc: service names\n");
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
continue;
}
else if (strcmp(kwd_dump, "all") == 0) {
all = 1;
}
if (all || strcmp(kwd_dump, "acl") == 0) {
printf("# List of registered ACL keywords:\n");
acl_dump_kwd();
}
if (all || strcmp(kwd_dump, "cfg") == 0) {
printf("# List of registered configuration keywords:\n");
cfg_dump_registered_keywords();
}
if (all || strcmp(kwd_dump, "cli") == 0) {
printf("# List of registered CLI keywords:\n");
cli_list_keywords();
}
if (all || strcmp(kwd_dump, "cnv") == 0) {
printf("# List of registered sample converter functions:\n");
smp_dump_conv_kw();
}
if (all || strcmp(kwd_dump, "flt") == 0) {
printf("# List of registered filter names:\n");
flt_dump_kws(NULL);
}
if (all || strcmp(kwd_dump, "smp") == 0) {
printf("# List of registered sample fetch functions:\n");
smp_dump_fetch_kw();
}
if (all || strcmp(kwd_dump, "svc") == 0) {
printf("# List of registered service names:\n");
list_services(NULL);
}
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
}
}
/* Generate a random cluster-secret in case the setting is not provided in the
* configuration. This allows to use features which rely on it albeit with some
* limitations.
*/
static void generate_random_cluster_secret()
{
/* used as a default random cluster-secret if none defined. */
uint64_t rand;
/* The caller must not overwrite an already defined secret. */
BUG_ON(cluster_secret_isset);
rand = ha_random64();
memcpy(global.cluster_secret, &rand, sizeof(rand));
rand = ha_random64();
memcpy(global.cluster_secret + sizeof(rand), &rand, sizeof(rand));
cluster_secret_isset = 1;
}
/*
* This function initializes all the necessary variables. It only returns
* if everything is OK. If something fails, it exits.
*/
static void init(int argc, char **argv)
{
char *progname = global.log_tag.area;
int err_code = 0;
struct wordlist *wl;
struct proxy *px;
struct post_check_fct *pcf;
struct pre_check_fct *prcf;
int ideal_maxconn;
const char *cc, *cflags, *opts;
#ifdef USE_OPENSSL
#ifdef USE_OPENSSL_WOLFSSL
wolfSSL_Init();
wolfSSL_Debugging_ON();
#endif
#ifdef USE_OPENSSL_AWSLC
const char *version_str = OpenSSL_version(OPENSSL_VERSION);
if (strncmp(version_str, "AWS-LC", 6) != 0) {
ha_alert("HAPRoxy built with AWS-LC but running with %s.\n", version_str);
exit(1);
}
#endif
#if (HA_OPENSSL_VERSION_NUMBER < 0x1010000fL)
/* Initialize the error strings of OpenSSL
* It only needs to be done explicitly with older versions of the SSL
* library. On newer versions, errors strings are loaded during start
* up. */
SSL_load_error_strings();
#endif
#endif
startup_logs_init();
if (init_acl() != 0)
exit(1);
/* Initialise lua. */
hlua_init();
global.mode |= (arg_mode & (MODE_DAEMON | MODE_MWORKER | MODE_FOREGROUND | MODE_VERBOSE
| MODE_QUIET | MODE_CHECK | MODE_DEBUG | MODE_ZERO_WARNING
| MODE_DIAG | MODE_CHECK_CONDITION | MODE_DUMP_LIBS | MODE_DUMP_KWD
| MODE_DUMP_CFG | MODE_DUMP_NB_L));
if (getenv("HAPROXY_MWORKER_WAIT_ONLY")) {
unsetenv("HAPROXY_MWORKER_WAIT_ONLY");
global.mode |= MODE_MWORKER_WAIT;
global.mode &= ~MODE_MWORKER;
}
/* set the atexit functions when not doing configuration check */
if (!(global.mode & (MODE_CHECK | MODE_CHECK_CONDITION))
&& (getenv("HAPROXY_MWORKER_REEXEC") != NULL)) {
if (global.mode & MODE_MWORKER) {
atexit_flag = 1;
atexit(reexec_on_failure);
} else if (global.mode & MODE_MWORKER_WAIT) {
atexit_flag = 1;
atexit(exit_on_waitmode_failure);
}
}
if (change_dir && chdir(change_dir) < 0) {
ha_alert("Could not change to directory %s : %s\n", change_dir, strerror(errno));
exit(1);
}
usermsgs_clr("config");
if (global.mode & MODE_CHECK_CONDITION) {
int result;
uint32_t err;
const char *errptr;
char *errmsg = NULL;
char *args[MAX_LINE_ARGS+1];
int arg = sizeof(args) / sizeof(*args);
size_t outlen;
char *w;
if (!check_condition)
usage(progname);
outlen = strlen(check_condition) + 1;
err = parse_line(check_condition, check_condition, &outlen, args, &arg,
PARSE_OPT_ENV | PARSE_OPT_WORD_EXPAND | PARSE_OPT_DQUOTE | PARSE_OPT_SQUOTE | PARSE_OPT_BKSLASH,
&errptr);
if (err & PARSE_ERR_QUOTE) {
ha_alert("Syntax Error in condition: Unmatched quote.\n");
exit(2);
}
if (err & PARSE_ERR_HEX) {
ha_alert("Syntax Error in condition: Truncated or invalid hexadecimal sequence.\n");
exit(2);
}
if (err & (PARSE_ERR_TOOLARGE|PARSE_ERR_OVERLAP)) {
ha_alert("Error in condition: Line too long.\n");
exit(2);
}
if (err & PARSE_ERR_TOOMANY) {
ha_alert("Error in condition: Too many words.\n");
exit(2);
}
if (err) {
ha_alert("Unhandled error in condition, please report this to the developers.\n");
exit(2);
}
/* remerge all words into a single expression */
for (w = *args; (w += strlen(w)) < check_condition + outlen - 1; *w = ' ')
;
result = cfg_eval_condition(args, &errmsg, &errptr);
if (result < 0) {
if (errmsg)
ha_alert("Failed to evaluate condition: %s\n", errmsg);
exit(2);
}
exit(result ? 0 : 1);
}
/* in wait mode, we don't try to read the configuration files */
if (!(global.mode & MODE_MWORKER_WAIT)) {
char *env_cfgfiles = NULL;
int env_err = 0;
/* handle cfgfiles that are actually directories */
cfgfiles_expand_directories();
if (LIST_ISEMPTY(&cfg_cfgfiles))
usage(progname);
/* temporary create environment variables with default
* values to ease user configuration. Do not forget to
* unset them after the list_for_each_entry loop.
*/
setenv("HAPROXY_HTTP_LOG_FMT", default_http_log_format, 1);
setenv("HAPROXY_HTTPS_LOG_FMT", default_https_log_format, 1);
setenv("HAPROXY_TCP_LOG_FMT", default_tcp_log_format, 1);
setenv("HAPROXY_BRANCH", PRODUCT_BRANCH, 1);
list_for_each_entry(wl, &cfg_cfgfiles, list) {
int ret;
if (env_err == 0) {
if (!memprintf(&env_cfgfiles, "%s%s%s",
(env_cfgfiles ? env_cfgfiles : ""),
(env_cfgfiles ? ";" : ""), wl->s))
env_err = 1;
}
ret = readcfgfile(wl->s);
if (ret == -1) {
ha_alert("Could not open configuration file %s : %s\n",
wl->s, strerror(errno));
free(env_cfgfiles);
exit(1);
}
if (ret & (ERR_ABORT|ERR_FATAL))
ha_alert("Error(s) found in configuration file : %s\n", wl->s);
err_code |= ret;
if (err_code & ERR_ABORT) {
free(env_cfgfiles);
exit(1);
}
}
/* remove temporary environment variables. */
unsetenv("HAPROXY_BRANCH");
unsetenv("HAPROXY_HTTP_LOG_FMT");
unsetenv("HAPROXY_HTTPS_LOG_FMT");
unsetenv("HAPROXY_TCP_LOG_FMT");
/* do not try to resolve arguments nor to spot inconsistencies when
* the configuration contains fatal errors caused by files not found
* or failed memory allocations.
*/
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Fatal errors found in configuration.\n");
free(env_cfgfiles);
exit(1);
}
if (env_err) {
ha_alert("Could not allocate memory for HAPROXY_CFGFILES env variable\n");
exit(1);
}
setenv("HAPROXY_CFGFILES", env_cfgfiles, 1);
free(env_cfgfiles);
MEDIUM: config: don't check config validity when there are fatal errors Overall we do have an issue with the severity of a number of errors. Most fatal errors are reported with ERR_FATAL (which prevents startup) and not ERR_ABORT (which stops parsing ASAP), but check_config_validity() is still called on ERR_FATAL, and will most of the time report bogus errors. This is what caused smp_resolve_args() to be called on a number of unparsable ACLs, and it also is what reports incorrect ordering or unresolvable section names when certain entries could not be properly parsed. This patch stops this domino effect by simply aborting before trying to further check and resolve the configuration when it's already know that there are fatal errors. A concrete example comes from this config : userlist users : user foo insecure-password bar listen foo bind :1234 mode htttp timeout client 10S timeout server 10s timeout connect 10s stats uri /stats stats http-request auth unless { http_auth(users) } http-request redirect location /index.html if { path / } It contains a colon after the userlist name, a typo in the client timeout value, another one in "mode http" which cause some other configuration elements not to be properly handled. Previously it would confusingly report : [ALERT] 108/114851 (20224) : parsing [err-report.cfg:1] : 'userlist' cannot handle unexpected argument ':'. [ALERT] 108/114851 (20224) : parsing [err-report.cfg:6] : unknown proxy mode 'htttp'. [ALERT] 108/114851 (20224) : parsing [err-report.cfg:7] : unexpected character 'S' in 'timeout client' [ALERT] 108/114851 (20224) : Error(s) found in configuration file : err-report.cfg [ALERT] 108/114851 (20224) : parsing [err-report.cfg:11] : unable to find userlist 'users' referenced in arg 1 of ACL keyword 'http_auth' in proxy 'foo'. [WARNING] 108/114851 (20224) : config : missing timeouts for proxy 'foo'. | While not properly invalid, you will certainly encounter various problems | with such a configuration. To fix this, please ensure that all following | timeouts are set to a non-zero value: 'client', 'connect', 'server'. [WARNING] 108/114851 (20224) : config : 'stats' statement ignored for proxy 'foo' as it requires HTTP mode. [WARNING] 108/114851 (20224) : config : 'http-request' rules ignored for proxy 'foo' as they require HTTP mode. [ALERT] 108/114851 (20224) : Fatal errors found in configuration. The "requires HTTP mode" errors are just pollution resulting from the improper spelling of this mode earlier. The unresolved reference to the userlist is caused by the extra colon on the declaration, and the warning regarding the missing timeouts is caused by the wrong character. Now it more accurately reports : [ALERT] 108/114900 (20225) : parsing [err-report.cfg:1] : 'userlist' cannot handle unexpected argument ':'. [ALERT] 108/114900 (20225) : parsing [err-report.cfg:6] : unknown proxy mode 'htttp'. [ALERT] 108/114900 (20225) : parsing [err-report.cfg:7] : unexpected character 'S' in 'timeout client' [ALERT] 108/114900 (20225) : Error(s) found in configuration file : err-report.cfg [ALERT] 108/114900 (20225) : Fatal errors found in configuration. Despite not really a fix, this patch should be backported at least to 1.7, possibly even 1.6, and 1.5 since it hardens the config parser against certain bad situations like the recently reported use-after-free and the last null dereference.
2017-04-19 05:24:07 -04:00
}
if (global.mode & MODE_MWORKER) {
struct mworker_proc *tmproc;
setenv("HAPROXY_MWORKER", "1", 1);
if (getenv("HAPROXY_MWORKER_REEXEC") == NULL) {
tmproc = mworker_proc_new();
if (!tmproc) {
ha_alert("Cannot allocate process structures.\n");
exit(EXIT_FAILURE);
}
tmproc->options |= PROC_O_TYPE_MASTER; /* master */
tmproc->pid = pid;
tmproc->timestamp = start_date.tv_sec;
proc_self = tmproc;
LIST_APPEND(&proc_list, &tmproc->list);
}
tmproc = mworker_proc_new();
if (!tmproc) {
ha_alert("Cannot allocate process structures.\n");
exit(EXIT_FAILURE);
}
tmproc->options |= PROC_O_TYPE_WORKER; /* worker */
if (mworker_cli_sockpair_new(tmproc, 0) < 0) {
exit(EXIT_FAILURE);
}
LIST_APPEND(&proc_list, &tmproc->list);
}
BUG/MEDIUM: master: force the thread count earlier Christopher bisected that recent commit d0b73bca71 ("MEDIUM: listener: switch bind_thread from global to group-local") broke the master socket in that only the first out of the Nth initial connections would work, where N is the number of threads, after which they all work. The cause is that the master socket was bound to multiple threads, despite global.nbthread being 1 there, so the incoming connection load balancing would try to send incoming connections to non-existing threads, however the bind_thread mask would nonetheless include multiple threads. What happened is that in 1.9 we forced "nbthread" to 1 in the master's poll loop with commit b3f2be338b ("MEDIUM: mworker: use the haproxy poll loop"). In 2.0, nbthread detection was enabled by default in commit 149ab779cc ("MAJOR: threads: enable one thread per CPU by default"). From this point on, the operation above is unsafe because everything during startup is performed with nbthread corresponding to the default value, then it changes to one when starting the polling loop. But by then we weren't using the wait mode except for reload errors, so even if it would have happened nobody would have noticed. In 2.5 with commit fab0fdce9 ("MEDIUM: mworker: reexec in waitpid mode after successful loading") we started to rexecute all the time, not just for errors, so as to release precious resources and to possibly spot bugs that were rarely exposed in this mode. By then the incoming connection LB was enforcing all_threads_mask on the listener's thread mask so that the incorrect value was being corrected while using it. Finally in 2.7 commit d0b73bca71 ("MEDIUM: listener: switch bind_thread from global to group-local") replaces the all_threads_mask there with the listener's bind_thread, but that one was never adjusted by the starting master, whose thread group was filled to N threads by the automatic detection during early setup. The best approach here is to set nbthread to 1 very early in init() when we're in the master in wait mode, so that we don't try to guess the best value and don't end up with incorrect bindings anymore. This patch does this and also sets nbtgroups to 1 in preparation for a possible future where this will also be automatically calculated. There is no need to backport this patch since no other versions were affected, but if it were to be discovered that the incorrect bind mask on some of the master's FDs could be responsible for any trouble in older versions, then the backport should be safe (provided that nbtgroups is dropped of course).
2022-07-22 11:35:49 -04:00
if (global.mode & MODE_MWORKER_WAIT) {
/* in exec mode, there's always exactly one thread. Failure to
* set these ones now will result in nbthread being detected
* automatically.
*/
global.nbtgroups = 1;
global.nbthread = 1;
}
if (global.mode & (MODE_MWORKER|MODE_MWORKER_WAIT)) {
struct wordlist *it, *c;
master = 1;
/* get the info of the children in the env */
if (mworker_env_to_proc_list() < 0) {
exit(EXIT_FAILURE);
}
if (!LIST_ISEMPTY(&mworker_cli_conf)) {
char *path = NULL;
if (mworker_cli_proxy_create() < 0) {
ha_alert("Can't create the master's CLI.\n");
exit(EXIT_FAILURE);
}
list_for_each_entry_safe(c, it, &mworker_cli_conf, list) {
if (mworker_cli_proxy_new_listener(c->s) == NULL) {
ha_alert("Can't create the master's CLI.\n");
exit(EXIT_FAILURE);
}
LIST_DELETE(&c->list);
free(c->s);
free(c);
}
/* Creates the mcli_reload listener, which is the listener used
* to retrieve the master CLI session which asked for the reload.
*
* ipc_fd[1] will be used as a listener, and ipc_fd[0]
* will be used to send the FD of the session.
*
* Both FDs will be kept in the master. The sockets are
* created only if they weren't inherited.
*/
if ((proc_self->ipc_fd[1] == -1) &&
socketpair(AF_UNIX, SOCK_STREAM, 0, proc_self->ipc_fd) < 0) {
ha_alert("cannot create the mcli_reload socketpair.\n");
exit(EXIT_FAILURE);
}
/* Create the mcli_reload listener from the proc_self struct */
memprintf(&path, "sockpair@%d", proc_self->ipc_fd[1]);
mcli_reload_bind_conf = mworker_cli_proxy_new_listener(path);
if (mcli_reload_bind_conf == NULL) {
ha_alert("Cannot create the mcli_reload listener.\n");
exit(EXIT_FAILURE);
}
ha_free(&path);
}
}
if (!LIST_ISEMPTY(&mworker_cli_conf) && !(arg_mode & MODE_MWORKER)) {
ha_alert("a master CLI socket was defined, but master-worker mode (-W) is not enabled.\n");
exit(EXIT_FAILURE);
}
/* destroy unreferenced defaults proxies */
proxy_destroy_all_unref_defaults();
list_for_each_entry(prcf, &pre_check_list, list)
err_code |= prcf->fct();
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Fatal errors found in configuration.\n");
exit(1);
}
/* update the ready date that will be used to count the startup time
* during config checks (e.g. to schedule certain tasks if needed)
*/
clock_update_date(0, 1);
clock_adjust_now_offset();
ready_date = date;
/* Note: global.nbthread will be initialized as part of this call */
err_code |= check_config_validity();
/* update the ready date to also account for the check time */
clock_update_date(0, 1);
clock_adjust_now_offset();
ready_date = date;
for (px = proxies_list; px; px = px->next) {
struct server *srv;
struct post_proxy_check_fct *ppcf;
struct post_server_check_fct *pscf;
if (px->flags & (PR_FL_DISABLED|PR_FL_STOPPED))
continue;
list_for_each_entry(pscf, &post_server_check_list, list) {
for (srv = px->srv; srv; srv = srv->next)
err_code |= pscf->fct(srv);
}
list_for_each_entry(ppcf, &post_proxy_check_list, list)
err_code |= ppcf->fct(px);
}
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Fatal errors found in configuration.\n");
exit(1);
}
err_code |= pattern_finalize_config();
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Failed to finalize pattern config.\n");
exit(1);
}
if (global.rlimit_memmax_all)
global.rlimit_memmax = global.rlimit_memmax_all;
#ifdef USE_NS
MAJOR: namespace: add Linux network namespace support This patch makes it possible to create binds and servers in separate namespaces. This can be used to proxy between multiple completely independent virtual networks (with possibly overlapping IP addresses) and a non-namespace-aware proxy implementation that supports the proxy protocol (v2). The setup is something like this: net1 on VLAN 1 (namespace 1) -\ net2 on VLAN 2 (namespace 2) -- haproxy ==== proxy (namespace 0) net3 on VLAN 3 (namespace 3) -/ The proxy is configured to make server connections through haproxy and sending the expected source/target addresses to haproxy using the proxy protocol. The network namespace setup on the haproxy node is something like this: = 8< = $ cat setup.sh ip netns add 1 ip link add link eth1 type vlan id 1 ip link set eth1.1 netns 1 ip netns exec 1 ip addr add 192.168.91.2/24 dev eth1.1 ip netns exec 1 ip link set eth1.$id up ... = 8< = = 8< = $ cat haproxy.cfg frontend clients bind 127.0.0.1:50022 namespace 1 transparent default_backend scb backend server mode tcp server server1 192.168.122.4:2222 namespace 2 send-proxy-v2 = 8< = A bind line creates the listener in the specified namespace, and connections originating from that listener also have their network namespace set to that of the listener. A server line either forces the connection to be made in a specified namespace or may use the namespace from the client-side connection if that was set. For more documentation please read the documentation included in the patch itself. Signed-off-by: KOVACS Tamas <ktamas@balabit.com> Signed-off-by: Sarkozi Laszlo <laszlo.sarkozi@balabit.com> Signed-off-by: KOVACS Krisztian <hidden@balabit.com>
2014-11-17 09:11:45 -05:00
err_code |= netns_init();
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Failed to initialize namespace support.\n");
MAJOR: namespace: add Linux network namespace support This patch makes it possible to create binds and servers in separate namespaces. This can be used to proxy between multiple completely independent virtual networks (with possibly overlapping IP addresses) and a non-namespace-aware proxy implementation that supports the proxy protocol (v2). The setup is something like this: net1 on VLAN 1 (namespace 1) -\ net2 on VLAN 2 (namespace 2) -- haproxy ==== proxy (namespace 0) net3 on VLAN 3 (namespace 3) -/ The proxy is configured to make server connections through haproxy and sending the expected source/target addresses to haproxy using the proxy protocol. The network namespace setup on the haproxy node is something like this: = 8< = $ cat setup.sh ip netns add 1 ip link add link eth1 type vlan id 1 ip link set eth1.1 netns 1 ip netns exec 1 ip addr add 192.168.91.2/24 dev eth1.1 ip netns exec 1 ip link set eth1.$id up ... = 8< = = 8< = $ cat haproxy.cfg frontend clients bind 127.0.0.1:50022 namespace 1 transparent default_backend scb backend server mode tcp server server1 192.168.122.4:2222 namespace 2 send-proxy-v2 = 8< = A bind line creates the listener in the specified namespace, and connections originating from that listener also have their network namespace set to that of the listener. A server line either forces the connection to be made in a specified namespace or may use the namespace from the client-side connection if that was set. For more documentation please read the documentation included in the patch itself. Signed-off-by: KOVACS Tamas <ktamas@balabit.com> Signed-off-by: Sarkozi Laszlo <laszlo.sarkozi@balabit.com> Signed-off-by: KOVACS Krisztian <hidden@balabit.com>
2014-11-17 09:11:45 -05:00
exit(1);
}
#endif
thread_detect_binding_discrepancies();
thread_detect_more_than_cpus();
/* Apply server states */
apply_server_state();
for (px = proxies_list; px; px = px->next)
srv_compute_all_admin_states(px);
/* Apply servers' configured address */
err_code |= srv_init_addr();
if (err_code & (ERR_ABORT|ERR_FATAL)) {
ha_alert("Failed to initialize server(s) addr.\n");
exit(1);
}
if (warned & WARN_ANY && global.mode & MODE_ZERO_WARNING) {
ha_alert("Some warnings were found and 'zero-warning' is set. Aborting.\n");
exit(1);
}
#if defined(HA_HAVE_DUMP_LIBS)
if (global.mode & MODE_DUMP_LIBS) {
qfprintf(stdout, "List of loaded object files:\n");
chunk_reset(&trash);
if (dump_libs(&trash, ((arg_mode & (MODE_QUIET|MODE_VERBOSE)) == MODE_VERBOSE)))
printf("%s", trash.area);
}
#endif
MINOR: management: add some basic keyword dump infrastructure It's difficult from outside haproxy to detect the supported keywords and syntax. Interestingly, many of our modern keywords are enumerated since they're registered from constructors, so it's not very hard to enumerate most of them. This patch creates some basic infrastructure to support dumping existing keywords from different classes on stdout. The format will differ depending on the classes, but the idea is that the output could easily be passed to a script that generates some simple syntax highlighting rules, completion rules for editors, syntax checkers or config parsers. The principle chosen here is that if "-dK" is passed on the command-line, at the end of the parsing the registered keywords will be dumped for the requested classes passed after "-dK". Special name "help" will show known classes, while "all" will execute all of them. The reason for doing that after the end of the config processor is that it will also enumerate internally-generated keywords, Lua or even those loaded from external code (e.g. if an add-on is loaded using LD_PRELOAD). A typical way to call this with a valid config would be: ./haproxy -dKall -q -c -f /path/to/config If there's no config available, feeding /dev/null will also do the job, though it will not be able to detect dynamically created keywords, of course. This patch also updates the management doc. For now nothing but the help is listed, various subsystems will follow in subsequent patches.
2022-03-08 10:01:40 -05:00
if (global.mode & MODE_DUMP_KWD)
dump_registered_keywords();
if (global.mode & MODE_CHECK) {
struct peers *pr;
struct proxy *px;
if (warned & WARN_ANY)
qfprintf(stdout, "Warnings were found.\n");
for (pr = cfg_peers; pr; pr = pr->next)
if (pr->peers_fe)
break;
for (px = proxies_list; px; px = px->next)
if (!(px->flags & (PR_FL_DISABLED|PR_FL_STOPPED)) && px->li_all)
break;
if (!px) {
/* We may only have log-forward section */
for (px = cfg_log_forward; px; px = px->next)
if (!(px->flags & (PR_FL_DISABLED|PR_FL_STOPPED)) && px->li_all)
break;
}
if (pr || px) {
/* At least one peer or one listener has been found */
if (global.mode & MODE_VERBOSE)
qfprintf(stdout, "Configuration file is valid\n");
MINOR: haproxy: Make use of deinit_and_exit() for clean exits Particularly cleanly deinit() after a configuration check to clean up the output of valgrind which reports "possible losses" without a deinit() and does not with a deinit(), converting actual losses into proper hard losses which makes the whole stuff easier to analyze. As an example, given an example configuration of the following: frontend foo bind *:8080 mode http Running `haproxy -c -f cfg` within valgrind will report 4 possible losses: $ valgrind --leak-check=full ./haproxy -c -f ./example.cfg ==21219== Memcheck, a memory error detector ==21219== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al. ==21219== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info ==21219== Command: ./haproxy -c -f ./example.cfg ==21219== [WARNING] 165/001100 (21219) : config : missing timeouts for frontend 'foo'. | While not properly invalid, you will certainly encounter various problems | with such a configuration. To fix this, please ensure that all following | timeouts are set to a non-zero value: 'client', 'connect', 'server'. Warnings were found. Configuration file is valid ==21219== ==21219== HEAP SUMMARY: ==21219== in use at exit: 1,436,631 bytes in 130 blocks ==21219== total heap usage: 153 allocs, 23 frees, 1,447,758 bytes allocated ==21219== ==21219== 7 bytes in 1 blocks are possibly lost in loss record 5 of 54 ==21219== at 0x4C2DB8F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x5726489: strdup (strdup.c:42) ==21219== by 0x468FD9: bind_conf_alloc (listener.h:158) ==21219== by 0x468FD9: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 14 bytes in 1 blocks are possibly lost in loss record 9 of 54 ==21219== at 0x4C2DB8F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x5726489: strdup (strdup.c:42) ==21219== by 0x468F9B: bind_conf_alloc (listener.h:154) ==21219== by 0x468F9B: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 128 bytes in 1 blocks are possibly lost in loss record 35 of 54 ==21219== at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x468F90: bind_conf_alloc (listener.h:152) ==21219== by 0x468F90: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 608 bytes in 1 blocks are possibly lost in loss record 46 of 54 ==21219== at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x4B953A: create_listeners (listener.c:576) ==21219== by 0x4578F6: str2listener (cfgparse.c:192) ==21219== by 0x469039: cfg_parse_listen (cfgparse-listen.c:568) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== LEAK SUMMARY: ==21219== definitely lost: 0 bytes in 0 blocks ==21219== indirectly lost: 0 bytes in 0 blocks ==21219== possibly lost: 757 bytes in 4 blocks ==21219== still reachable: 1,435,874 bytes in 126 blocks ==21219== suppressed: 0 bytes in 0 blocks ==21219== Reachable blocks (those to which a pointer was found) are not shown. ==21219== To see them, rerun with: --leak-check=full --show-leak-kinds=all ==21219== ==21219== For counts of detected and suppressed errors, rerun with: -v ==21219== ERROR SUMMARY: 4 errors from 4 contexts (suppressed: 0 from 0) Re-running the same command with the patch applied will not report any losses any more: $ valgrind --leak-check=full ./haproxy -c -f ./example.cfg ==22124== Memcheck, a memory error detector ==22124== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al. ==22124== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info ==22124== Command: ./haproxy -c -f ./example.cfg ==22124== [WARNING] 165/001503 (22124) : config : missing timeouts for frontend 'foo'. | While not properly invalid, you will certainly encounter various problems | with such a configuration. To fix this, please ensure that all following | timeouts are set to a non-zero value: 'client', 'connect', 'server'. Warnings were found. Configuration file is valid ==22124== ==22124== HEAP SUMMARY: ==22124== in use at exit: 313,864 bytes in 82 blocks ==22124== total heap usage: 153 allocs, 71 frees, 1,447,758 bytes allocated ==22124== ==22124== LEAK SUMMARY: ==22124== definitely lost: 0 bytes in 0 blocks ==22124== indirectly lost: 0 bytes in 0 blocks ==22124== possibly lost: 0 bytes in 0 blocks ==22124== still reachable: 313,864 bytes in 82 blocks ==22124== suppressed: 0 bytes in 0 blocks ==22124== Reachable blocks (those to which a pointer was found) are not shown. ==22124== To see them, rerun with: --leak-check=full --show-leak-kinds=all ==22124== ==22124== For counts of detected and suppressed errors, rerun with: -v ==22124== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0) It might be worth investigating what exactly HAProxy does to lose pointers to the start of those 4 memory areas and then to be able to still free them during deinit(). If HAProxy is able to free them, they ideally should be "still reachable" and not "possibly lost".
2020-06-13 18:37:42 -04:00
deinit_and_exit(0);
}
qfprintf(stdout, "Configuration file has no error but will not start (no listener) => exit(2).\n");
exit(2);
}
if (global.mode & MODE_DUMP_CFG)
deinit_and_exit(0);
if (global.mode & MODE_DIAG) {
cfg_run_diagnostics();
}
#ifdef USE_OPENSSL
/* Initialize SSL random generator. Must be called before chroot for
* access to /dev/urandom, and before ha_random_boot() which may use
* RAND_bytes().
*/
if (!ssl_initialize_random()) {
ha_alert("OpenSSL random data generator initialization failed.\n");
exit(EXIT_FAILURE);
}
#endif
ha_random_boot(argv); // the argv pointer brings some kernel-fed entropy
CLEANUP: channel: use "channel" instead of "buffer" in function names This is a massive rename of most functions which should make use of the word "channel" instead of the word "buffer" in their names. In concerns the following ones (new names) : unsigned long long channel_forward(struct channel *buf, unsigned long long bytes); static inline void channel_init(struct channel *buf) static inline int channel_input_closed(struct channel *buf) static inline int channel_output_closed(struct channel *buf) static inline void channel_check_timeouts(struct channel *b) static inline void channel_erase(struct channel *buf) static inline void channel_shutr_now(struct channel *buf) static inline void channel_shutw_now(struct channel *buf) static inline void channel_abort(struct channel *buf) static inline void channel_stop_hijacker(struct channel *buf) static inline void channel_auto_connect(struct channel *buf) static inline void channel_dont_connect(struct channel *buf) static inline void channel_auto_close(struct channel *buf) static inline void channel_dont_close(struct channel *buf) static inline void channel_auto_read(struct channel *buf) static inline void channel_dont_read(struct channel *buf) unsigned long long channel_forward(struct channel *buf, unsigned long long bytes) Some functions provided by channel.[ch] have kept their "buffer" name because they are really designed to act on the buffer according to some information gathered from the channel. They have been moved together to the same place in the file for better readability but they were not changed at all. The "buffer" memory pool was also renamed "channel".
2012-08-27 18:06:31 -04:00
/* now we know the buffer size, we can initialize the channels and buffers */
init_buffer();
list_for_each_entry(pcf, &post_check_list, list) {
err_code |= pcf->fct();
if (err_code & (ERR_ABORT|ERR_FATAL))
exit(1);
}
/* set the default maxconn in the master, but let it be rewritable with -n */
if (global.mode & MODE_MWORKER_WAIT)
global.maxconn = MASTER_MAXCONN;
if (cfg_maxconn > 0)
global.maxconn = cfg_maxconn;
if (global.cli_fe)
global.maxsock += global.cli_fe->maxconn;
if (cfg_peers) {
/* peers also need to bypass global maxconn */
struct peers *p = cfg_peers;
for (p = cfg_peers; p; p = p->next)
if (p->peers_fe)
global.maxsock += p->peers_fe->maxconn;
}
/* Now we want to compute the maxconn and possibly maxsslconn values.
* It's a bit tricky. Maxconn defaults to the pre-computed value based
* on rlim_fd_cur and the number of FDs in use due to the configuration,
* and maxsslconn defaults to DEFAULT_MAXSSLCONN. On top of that we can
* enforce a lower limit based on memmax.
*
* If memmax is set, then it depends on which values are set. If
* maxsslconn is set, we use memmax to determine how many cleartext
* connections may be added, and set maxconn to the sum of the two.
* If maxconn is set and not maxsslconn, maxsslconn is computed from
* the remaining amount of memory between memmax and the cleartext
* connections. If neither are set, then it is considered that all
* connections are SSL-capable, and maxconn is computed based on this,
* then maxsslconn accordingly. We need to know if SSL is used on the
* frontends, backends, or both, because when it's used on both sides,
* we need twice the value for maxsslconn, but we only count the
* handshake once since it is not performed on the two sides at the
* same time (frontend-side is terminated before backend-side begins).
* The SSL stack is supposed to have filled ssl_session_cost and
* ssl_handshake_cost during its initialization. In any case, if
* SYSTEM_MAXCONN is set, we still enforce it as an upper limit for
* maxconn in order to protect the system.
*/
ideal_maxconn = compute_ideal_maxconn();
if (!global.rlimit_memmax) {
if (global.maxconn == 0) {
global.maxconn = ideal_maxconn;
if (global.mode & (MODE_VERBOSE|MODE_DEBUG))
fprintf(stderr, "Note: setting global.maxconn to %d.\n", global.maxconn);
}
}
#ifdef USE_OPENSSL
else if (!global.maxconn && !global.maxsslconn &&
(global.ssl_used_frontend || global.ssl_used_backend)) {
/* memmax is set, compute everything automatically. Here we want
* to ensure that all SSL connections will be served. We take
* care of the number of sides where SSL is used, and consider
* the worst case : SSL used on both sides and doing a handshake
* simultaneously. Note that we can't have more than maxconn
* handshakes at a time by definition, so for the worst case of
* two SSL conns per connection, we count a single handshake.
*/
int sides = !!global.ssl_used_frontend + !!global.ssl_used_backend;
int64_t mem = global.rlimit_memmax * 1048576ULL;
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
int retried = 0;
mem -= global.tune.sslcachesize * 200ULL; // about 200 bytes per SSL cache entry
mem -= global.maxzlibmem;
mem = mem * MEM_USABLE_RATIO;
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
/* Principle: we test once to set maxconn according to the free
* memory. If it results in values the system rejects, we try a
* second time by respecting rlim_fd_max. If it fails again, we
* go back to the initial value and will let the final code
* dealing with rlimit report the error. That's up to 3 attempts.
*/
do {
global.maxconn = mem /
((STREAM_MAX_COST + 2 * global.tune.bufsize) + // stream + 2 buffers per stream
sides * global.ssl_session_max_cost + // SSL buffers, one per side
global.ssl_handshake_max_cost); // 1 handshake per connection max
if (retried == 1)
global.maxconn = MIN(global.maxconn, ideal_maxconn);
global.maxconn = round_2dig(global.maxconn);
#ifdef SYSTEM_MAXCONN
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
if (global.maxconn > SYSTEM_MAXCONN)
global.maxconn = SYSTEM_MAXCONN;
#endif /* SYSTEM_MAXCONN */
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
global.maxsslconn = sides * global.maxconn;
if (check_if_maxsock_permitted(compute_ideal_maxsock(global.maxconn)))
break;
} while (retried++ < 2);
if (global.mode & (MODE_VERBOSE|MODE_DEBUG))
fprintf(stderr, "Note: setting global.maxconn to %d and global.maxsslconn to %d.\n",
global.maxconn, global.maxsslconn);
}
else if (!global.maxsslconn &&
(global.ssl_used_frontend || global.ssl_used_backend)) {
/* memmax and maxconn are known, compute maxsslconn automatically.
* maxsslconn being forced, we don't know how many of it will be
* on each side if both sides are being used. The worst case is
* when all connections use only one SSL instance because
* handshakes may be on two sides at the same time.
*/
int sides = !!global.ssl_used_frontend + !!global.ssl_used_backend;
int64_t mem = global.rlimit_memmax * 1048576ULL;
int64_t sslmem;
mem -= global.tune.sslcachesize * 200ULL; // about 200 bytes per SSL cache entry
mem -= global.maxzlibmem;
mem = mem * MEM_USABLE_RATIO;
REORG/MAJOR: session: rename the "session" entity to "stream" With HTTP/2, we'll have to support multiplexed streams. A stream is in fact the largest part of what we currently call a session, it has buffers, logs, etc. In order to catch any error, this commit removes any reference to the struct session and tries to rename most "session" occurrences in function names to "stream" and "sess" to "strm" when that's related to a session. The files stream.{c,h} were added and session.{c,h} removed. The session will be reintroduced later and a few parts of the stream will progressively be moved overthere. It will more or less contain only what we need in an embryonic session. Sample fetch functions and converters will have to change a bit so that they'll use an L5 (session) instead of what's currently called "L4" which is in fact L6 for now. Once all changes are completed, we should see approximately this : L7 - http_txn L6 - stream L5 - session L4 - connection | applet There will be at most one http_txn per stream, and a same session will possibly be referenced by multiple streams. A connection will point to a session and to a stream. The session will hold all the information we need to keep even when we don't yet have a stream. Some more cleanup is needed because some code was already far from being clean. The server queue management still refers to sessions at many places while comments talk about connections. This will have to be cleaned up once we have a server-side connection pool manager. Stream flags "SN_*" still need to be renamed, it doesn't seem like any of them will need to move to the session.
2015-04-02 18:22:06 -04:00
sslmem = mem - global.maxconn * (int64_t)(STREAM_MAX_COST + 2 * global.tune.bufsize);
global.maxsslconn = sslmem / (global.ssl_session_max_cost + global.ssl_handshake_max_cost);
global.maxsslconn = round_2dig(global.maxsslconn);
if (sslmem <= 0 || global.maxsslconn < sides) {
ha_alert("Cannot compute the automatic maxsslconn because global.maxconn is already too "
"high for the global.memmax value (%d MB). The absolute maximum possible value "
"without SSL is %d, but %d was found and SSL is in use.\n",
global.rlimit_memmax,
(int)(mem / (STREAM_MAX_COST + 2 * global.tune.bufsize)),
global.maxconn);
exit(1);
}
if (global.maxsslconn > sides * global.maxconn)
global.maxsslconn = sides * global.maxconn;
if (global.mode & (MODE_VERBOSE|MODE_DEBUG))
fprintf(stderr, "Note: setting global.maxsslconn to %d\n", global.maxsslconn);
}
#endif
else if (!global.maxconn) {
/* memmax and maxsslconn are known/unused, compute maxconn automatically */
int sides = !!global.ssl_used_frontend + !!global.ssl_used_backend;
int64_t mem = global.rlimit_memmax * 1048576ULL;
int64_t clearmem;
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
int retried = 0;
if (global.ssl_used_frontend || global.ssl_used_backend)
mem -= global.tune.sslcachesize * 200ULL; // about 200 bytes per SSL cache entry
mem -= global.maxzlibmem;
mem = mem * MEM_USABLE_RATIO;
clearmem = mem;
if (sides)
clearmem -= (global.ssl_session_max_cost + global.ssl_handshake_max_cost) * (int64_t)global.maxsslconn;
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
/* Principle: we test once to set maxconn according to the free
* memory. If it results in values the system rejects, we try a
* second time by respecting rlim_fd_max. If it fails again, we
* go back to the initial value and will let the final code
* dealing with rlimit report the error. That's up to 3 attempts.
*/
do {
global.maxconn = clearmem / (STREAM_MAX_COST + 2 * global.tune.bufsize);
if (retried == 1)
global.maxconn = MIN(global.maxconn, ideal_maxconn);
global.maxconn = round_2dig(global.maxconn);
#ifdef SYSTEM_MAXCONN
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
if (global.maxconn > SYSTEM_MAXCONN)
global.maxconn = SYSTEM_MAXCONN;
#endif /* SYSTEM_MAXCONN */
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
if (clearmem <= 0 || !global.maxconn) {
ha_alert("Cannot compute the automatic maxconn because global.maxsslconn is already too "
"high for the global.memmax value (%d MB). The absolute maximum possible value "
"is %d, but %d was found.\n",
global.rlimit_memmax,
(int)(mem / (global.ssl_session_max_cost + global.ssl_handshake_max_cost)),
MEDIUM: init: always try to push the FD limit when maxconn is set from -m When a maximum memory setting is passed to haproxy and maxconn is not set and ulimit-n is not set, it is expected that maxconn will be set to the highest value permitted by this memory setting, possibly affecting the FD limit. When maxconn was changed to be deduced from the current process's FD limit, the automatic setting above was partially lost because it now remains limited to the current FD limit in addition to being limited to the memory usage. For unprivileged processes it does not change anything, but for privileged processes the difference is important. Indeed, the previous behavior ensured that the new FD limit could be enforced on the process as long as the user had the privilege to do so. Now this does not happen anymore, and some people rely on this for automatic sizing in VM environments. This patch implements the ability to verify if the setting will be enforceable on the process or not. First it computes maxconn based on the memory limits alone, then checks if the process is willing to accept them, otherwise tries again by respecting the process' hard limit. Thanks to this we now have the best of the pre-2.0 behavior and the current one, in that privileged users will be able to get as high a maxconn as they need just based on the memory limit, while unprivileged users will still get as high a setting as permitted by the intersection of the memory limit and the process' FD limit. Ideally, after some observation period, this patch along with the previous one "MINOR: init: move the maxsock calculation code to compute_ideal_maxsock()" should be backported to 2.1 and 2.0. Thanks to Baptiste for raising the issue.
2020-03-10 12:54:54 -04:00
global.maxsslconn);
exit(1);
}
if (check_if_maxsock_permitted(compute_ideal_maxsock(global.maxconn)))
break;
} while (retried++ < 2);
if (global.mode & (MODE_VERBOSE|MODE_DEBUG)) {
if (sides && global.maxsslconn > sides * global.maxconn) {
fprintf(stderr, "Note: global.maxsslconn is forced to %d which causes global.maxconn "
"to be limited to %d. Better reduce global.maxsslconn to get more "
"room for extra connections.\n", global.maxsslconn, global.maxconn);
}
fprintf(stderr, "Note: setting global.maxconn to %d\n", global.maxconn);
}
}
global.maxsock = compute_ideal_maxsock(global.maxconn);
global.hardmaxconn = global.maxconn;
if (!global.maxpipes)
global.maxpipes = compute_ideal_maxpipes();
/* update connection pool thresholds */
global.tune.pool_low_count = ((long long)global.maxsock * global.tune.pool_low_ratio + 99) / 100;
global.tune.pool_high_count = ((long long)global.maxsock * global.tune.pool_high_ratio + 99) / 100;
proxy_adjust_all_maxconn();
if (global.tune.maxpollevents <= 0)
global.tune.maxpollevents = MAX_POLL_EVENTS;
if (global.tune.runqueue_depth <= 0) {
/* tests on various thread counts from 1 to 64 have shown an
* optimal queue depth following roughly 1/sqrt(threads).
*/
int s = my_flsl(global.nbthread);
s += (global.nbthread / s); // roughly twice the sqrt.
global.tune.runqueue_depth = RUNQUEUE_DEPTH * 2 / s;
}
if (global.tune.recv_enough == 0)
global.tune.recv_enough = MIN_RECV_AT_ONCE_ENOUGH;
if (global.tune.maxrewrite >= global.tune.bufsize / 2)
global.tune.maxrewrite = global.tune.bufsize / 2;
usermsgs_clr(NULL);
if (arg_mode & (MODE_DEBUG | MODE_FOREGROUND)) {
/* command line debug mode inhibits configuration mode */
global.mode &= ~(MODE_DAEMON | MODE_QUIET);
global.mode |= (arg_mode & (MODE_DEBUG | MODE_FOREGROUND));
}
if (arg_mode & MODE_DAEMON) {
/* command line daemon mode inhibits foreground and debug modes mode */
global.mode &= ~(MODE_DEBUG | MODE_FOREGROUND);
global.mode |= arg_mode & MODE_DAEMON;
}
global.mode |= (arg_mode & (MODE_QUIET | MODE_VERBOSE));
if ((global.mode & MODE_DEBUG) && (global.mode & (MODE_DAEMON | MODE_QUIET))) {
ha_warning("<debug> mode incompatible with <quiet> and <daemon>. Keeping <debug> only.\n");
global.mode &= ~(MODE_DAEMON | MODE_QUIET);
}
/* Realloc trash buffers because global.tune.bufsize may have changed */
if (!init_trash_buffers(0)) {
ha_alert("failed to initialize trash buffers.\n");
exit(1);
}
if (!init_log_buffers()) {
ha_alert("failed to initialize log buffers.\n");
exit(1);
}
if (!cluster_secret_isset)
generate_random_cluster_secret();
/*
* Note: we could register external pollers here.
* Built-in pollers have been registered before main().
*/
if (!(global.tune.options & GTUNE_USE_KQUEUE))
disable_poller("kqueue");
if (!(global.tune.options & GTUNE_USE_EVPORTS))
disable_poller("evports");
if (!(global.tune.options & GTUNE_USE_EPOLL))
disable_poller("epoll");
if (!(global.tune.options & GTUNE_USE_POLL))
disable_poller("poll");
if (!(global.tune.options & GTUNE_USE_SELECT))
disable_poller("select");
/* Note: we could disable any poller by name here */
if (global.mode & (MODE_VERBOSE|MODE_DEBUG)) {
list_pollers(stderr);
fprintf(stderr, "\n");
list_filters(stderr);
}
if (!init_pollers()) {
ha_alert("No polling mechanism available.\n"
" This may happen when using thread-groups with old pollers (poll/select), or\n"
" it is possible that haproxy was built with TARGET=generic and that FD_SETSIZE\n"
" is too low on this platform to support maxconn and the number of listeners\n"
" and servers. You should rebuild haproxy specifying your system using TARGET=\n"
" in order to support other polling systems (poll, epoll, kqueue) or reduce the\n"
" global maxconn setting to accommodate the system's limitation. For reference,\n"
" FD_SETSIZE=%d on this system, global.maxconn=%d resulting in a maximum of\n"
" %d file descriptors. You should thus reduce global.maxconn by %d. Also,\n"
" check build settings using 'haproxy -vv'.\n\n",
FD_SETSIZE, global.maxconn, global.maxsock, (global.maxsock + 1 - FD_SETSIZE) / 2);
exit(1);
}
if (global.mode & (MODE_VERBOSE|MODE_DEBUG)) {
printf("Using %s() as the polling mechanism.\n", cur_poller.name);
}
if (!global.node)
global.node = strdup(hostname);
/* stop disabled proxies */
for (px = proxies_list; px; px = px->next) {
if (px->flags & (PR_FL_DISABLED|PR_FL_STOPPED))
stop_proxy(px);
}
if (!hlua_post_init())
exit(1);
/* Set the per-thread pool cache size to the default value if not set.
* This is the right place to decide to automatically adjust it (e.g.
* check L2 cache size, thread counts or take into account certain
* expensive pools).
*/
if (!global.tune.pool_cache_size)
global.tune.pool_cache_size = CONFIG_HAP_POOL_CACHE_SIZE;
/* fill in a few info about our version and build options */
chunk_reset(&trash);
/* toolchain */
cc = chunk_newstr(&trash);
#if defined(__clang_version__)
chunk_appendf(&trash, "clang-" __clang_version__);
#elif defined(__VERSION__)
chunk_appendf(&trash, "gcc-" __VERSION__);
#endif
#if __has_feature(address_sanitizer) || defined(__SANITIZE_ADDRESS__)
chunk_appendf(&trash, "+asan");
#endif
/* toolchain opts */
cflags = chunk_newstr(&trash);
#ifdef BUILD_CC
chunk_appendf(&trash, "%s", BUILD_CC);
#endif
#ifdef BUILD_CFLAGS
chunk_appendf(&trash, " %s", BUILD_CFLAGS);
#endif
#ifdef BUILD_DEBUG
chunk_appendf(&trash, " %s", BUILD_DEBUG);
#endif
/* settings */
opts = chunk_newstr(&trash);
#ifdef BUILD_TARGET
chunk_appendf(&trash, "TARGET='%s'", BUILD_TARGET);
#endif
#ifdef BUILD_CPU
chunk_appendf(&trash, " CPU='%s'", BUILD_CPU);
#endif
#ifdef BUILD_OPTIONS
chunk_appendf(&trash, " %s", BUILD_OPTIONS);
#endif
post_mortem_add_component("haproxy", haproxy_version, cc, cflags, opts, argv[0]);
}
void deinit(void)
{
struct proxy *p = proxies_list, *p0;
struct wordlist *wl, *wlb;
struct uri_auth *uap, *ua = NULL;
struct logger *log, *logb;
struct build_opts_str *bol, *bolb;
struct post_deinit_fct *pdf, *pdfb;
struct proxy_deinit_fct *pxdf, *pxdfb;
struct server_deinit_fct *srvdf, *srvdfb;
struct per_thread_init_fct *tif, *tifb;
struct per_thread_deinit_fct *tdf, *tdfb;
struct per_thread_alloc_fct *taf, *tafb;
struct per_thread_free_fct *tff, *tffb;
struct post_server_check_fct *pscf, *pscfb;
struct post_check_fct *pcf, *pcfb;
struct post_proxy_check_fct *ppcf, *ppcfb;
struct pre_check_fct *prcf, *prcfb;
struct cfg_postparser *pprs, *pprsb;
int cur_fd;
/* the user may want to skip this phase */
if (global.tune.options & GTUNE_QUICK_EXIT)
return;
/* At this point the listeners state is weird:
* - most listeners are still bound and referenced in their protocol
* - some might be zombies that are not in their proto anymore, but
* still appear in their proxy's listeners with a valid FD.
* - some might be stopped and still appear in their proxy as FD #-1
* - among all of them, some might be inherited hence shared and we're
* not allowed to pause them or whatever, we must just close them.
* - finally some are not listeners (pipes, logs, stdout, etc) and
* must be left intact.
*
* The safe way to proceed is to unbind (and close) whatever is not yet
* unbound so that no more receiver/listener remains alive. Then close
* remaining listener FDs, which correspond to zombie listeners (those
* belonging to disabled proxies that were in another process).
* objt_listener() would be cleaner here but not converted yet.
*/
protocol_unbind_all();
for (cur_fd = 0; cur_fd < global.maxsock; cur_fd++) {
if (!fdtab || !fdtab[cur_fd].owner)
continue;
if (fdtab[cur_fd].iocb == &sock_accept_iocb) {
struct listener *l = fdtab[cur_fd].owner;
BUG_ON(l->state != LI_INIT);
unbind_listener(l);
}
}
deinit_signals();
while (p) {
/* build a list of unique uri_auths */
if (!ua)
ua = p->uri_auth;
else {
/* check if p->uri_auth is unique */
for (uap = ua; uap; uap=uap->next)
if (uap == p->uri_auth)
break;
if (!uap && p->uri_auth) {
/* add it, if it is */
p->uri_auth->next = ua;
ua = p->uri_auth;
}
}
p0 = p;
p = p->next;
free_proxy(p0);
}/* end while(p) */
/* we don't need to free sink_proxies_list nor cfg_log_forward proxies since
* they are respectively cleaned up in sink_deinit() and deinit_log_forward()
*/
/* destroy all referenced defaults proxies */
proxy_destroy_all_unref_defaults();
while (ua) {
struct stat_scope *scope, *scopep;
uap = ua;
ua = ua->next;
free(uap->uri_prefix);
free(uap->auth_realm);
free(uap->node);
free(uap->desc);
userlist_free(uap->userlist);
free_act_rules(&uap->http_req_rules);
scope = uap->scope;
while (scope) {
scopep = scope;
scope = scope->next;
free(scopep->px_id);
free(scopep);
}
free(uap);
}
userlist_free(userlist);
cfg_unregister_sections();
deinit_log_buffers();
list_for_each_entry(pdf, &post_deinit_list, list)
pdf->fct();
ha_free(&global.log_send_hostname);
chunk_destroy(&global.log_tag);
ha_free(&global.chroot);
ha_free(&global.pidfile);
ha_free(&global.node);
ha_free(&global.desc);
ha_free(&oldpids);
ha_free(&old_argv);
ha_free(&localpeer);
ha_free(&global.server_state_base);
ha_free(&global.server_state_file);
task_destroy(idle_conn_task);
idle_conn_task = NULL;
list_for_each_entry_safe(log, logb, &global.loggers, list) {
LIST_DEL_INIT(&log->list);
free_logger(log);
}
list_for_each_entry_safe(wl, wlb, &cfg_cfgfiles, list) {
free(wl->s);
LIST_DELETE(&wl->list);
free(wl);
}
list_for_each_entry_safe(bol, bolb, &build_opts_list, list) {
if (bol->must_free)
free((void *)bol->str);
LIST_DELETE(&bol->list);
free(bol);
}
list_for_each_entry_safe(pxdf, pxdfb, &proxy_deinit_list, list) {
LIST_DELETE(&pxdf->list);
free(pxdf);
}
list_for_each_entry_safe(pdf, pdfb, &post_deinit_list, list) {
LIST_DELETE(&pdf->list);
free(pdf);
}
list_for_each_entry_safe(srvdf, srvdfb, &server_deinit_list, list) {
LIST_DELETE(&srvdf->list);
free(srvdf);
}
list_for_each_entry_safe(pcf, pcfb, &post_check_list, list) {
LIST_DELETE(&pcf->list);
free(pcf);
}
list_for_each_entry_safe(pscf, pscfb, &post_server_check_list, list) {
LIST_DELETE(&pscf->list);
free(pscf);
}
list_for_each_entry_safe(ppcf, ppcfb, &post_proxy_check_list, list) {
LIST_DELETE(&ppcf->list);
free(ppcf);
}
list_for_each_entry_safe(prcf, prcfb, &pre_check_list, list) {
LIST_DELETE(&prcf->list);
free(prcf);
}
list_for_each_entry_safe(tif, tifb, &per_thread_init_list, list) {
LIST_DELETE(&tif->list);
free(tif);
}
list_for_each_entry_safe(tdf, tdfb, &per_thread_deinit_list, list) {
LIST_DELETE(&tdf->list);
free(tdf);
}
list_for_each_entry_safe(taf, tafb, &per_thread_alloc_list, list) {
LIST_DELETE(&taf->list);
free(taf);
}
list_for_each_entry_safe(tff, tffb, &per_thread_free_list, list) {
LIST_DELETE(&tff->list);
free(tff);
}
list_for_each_entry_safe(pprs, pprsb, &postparsers, list) {
LIST_DELETE(&pprs->list);
free(pprs);
}
vars_prune(&proc_vars, NULL, NULL);
pool_destroy_all();
[MEDIUM] Fix memory freeing at exit New functions implemented: - deinit_pollers: called at the end of deinit()) - prune_acl: called via list_for_each_entry_safe Add missing pool_destroy2 calls: - p->hdr_idx_pool - pool2_tree64 Implement all task stopping: - health-check: needs new "struct task" in the struct server - queue processing: queue_mgt - appsess_refresh: appsession_refresh before (idle system): ==6079== LEAK SUMMARY: ==6079== definitely lost: 1,112 bytes in 75 blocks. ==6079== indirectly lost: 53,356 bytes in 2,090 blocks. ==6079== possibly lost: 52 bytes in 1 blocks. ==6079== still reachable: 150,996 bytes in 504 blocks. ==6079== suppressed: 0 bytes in 0 blocks. after (idle system): ==6945== LEAK SUMMARY: ==6945== definitely lost: 7,644 bytes in 137 blocks. ==6945== indirectly lost: 9,913 bytes in 587 blocks. ==6945== possibly lost: 0 bytes in 0 blocks. ==6945== still reachable: 0 bytes in 0 blocks. ==6945== suppressed: 0 bytes in 0 blocks. before (running system for ~2m): ==9343== LEAK SUMMARY: ==9343== definitely lost: 1,112 bytes in 75 blocks. ==9343== indirectly lost: 54,199 bytes in 2,122 blocks. ==9343== possibly lost: 52 bytes in 1 blocks. ==9343== still reachable: 151,128 bytes in 509 blocks. ==9343== suppressed: 0 bytes in 0 blocks. after (running system for ~2m): ==11616== LEAK SUMMARY: ==11616== definitely lost: 7,644 bytes in 137 blocks. ==11616== indirectly lost: 9,981 bytes in 591 blocks. ==11616== possibly lost: 0 bytes in 0 blocks. ==11616== still reachable: 4 bytes in 1 blocks. ==11616== suppressed: 0 bytes in 0 blocks. Still not perfect but significant improvement.
2008-05-29 17:53:44 -04:00
deinit_pollers();
} /* end deinit() */
__attribute__((noreturn)) void deinit_and_exit(int status)
{
global.mode |= MODE_STOPPING;
deinit();
exit(status);
}
/* Runs the polling loop */
void run_poll_loop()
{
int next, wake;
_HA_ATOMIC_OR(&th_ctx->flags, TH_FL_IN_LOOP);
clock_update_date(0,1);
while (1) {
MINOR: tasks: split wake_expired_tasks() in two parts to avoid useless wakeups We used to have wake_expired_tasks() wake up tasks and return the next expiration delay. The problem this causes is that we have to call it just before poll() in order to consider latest timers, but this also means that we don't wake up all newly expired tasks upon return from poll(), which thus systematically requires a second poll() round. This is visible when running any scheduled task like a health check, as there are systematically two poll() calls, one with the interval, nothing is done after it, and another one with a zero delay, and the task is called: listen test bind *:8001 server s1 127.0.0.1:1111 check 09:37:38.200959 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8696843}) = 0 09:37:38.200967 epoll_wait(3, [], 200, 1000) = 0 09:37:39.202459 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8712467}) = 0 >> nothing run here, as the expired task was not woken up yet. 09:37:39.202497 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8715766}) = 0 09:37:39.202505 epoll_wait(3, [], 200, 0) = 0 09:37:39.202513 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8719064}) = 0 >> now the expired task was woken up 09:37:39.202522 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 7 09:37:39.202537 fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 09:37:39.202565 setsockopt(7, SOL_TCP, TCP_NODELAY, [1], 4) = 0 09:37:39.202577 setsockopt(7, SOL_TCP, TCP_QUICKACK, [0], 4) = 0 09:37:39.202585 connect(7, {sa_family=AF_INET, sin_port=htons(1111), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress) 09:37:39.202659 epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLOUT, {u32=7, u64=7}}) = 0 09:37:39.202673 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8814713}) = 0 09:37:39.202683 epoll_wait(3, [{EPOLLOUT|EPOLLERR|EPOLLHUP, {u32=7, u64=7}}], 200, 1000) = 1 09:37:39.202693 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8818617}) = 0 09:37:39.202701 getsockopt(7, SOL_SOCKET, SO_ERROR, [111], [4]) = 0 09:37:39.202715 close(7) = 0 Let's instead split the function in two parts: - the first part, wake_expired_tasks(), called just before process_runnable_tasks(), wakes up all expired tasks; it doesn't compute any timeout. - the second part, next_timer_expiry(), called just before poll(), only computes the next timeout for the current thread. Thanks to this, all expired tasks are properly woken up when leaving poll, and each poll call's timeout remains up to date: 09:41:16.270449 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10223556}) = 0 09:41:16.270457 epoll_wait(3, [], 200, 999) = 0 09:41:17.270130 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10238572}) = 0 09:41:17.270157 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 7 09:41:17.270194 fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 09:41:17.270204 setsockopt(7, SOL_TCP, TCP_NODELAY, [1], 4) = 0 09:41:17.270216 setsockopt(7, SOL_TCP, TCP_QUICKACK, [0], 4) = 0 09:41:17.270224 connect(7, {sa_family=AF_INET, sin_port=htons(1111), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress) 09:41:17.270299 epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLOUT, {u32=7, u64=7}}) = 0 09:41:17.270314 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10337841}) = 0 09:41:17.270323 epoll_wait(3, [{EPOLLOUT|EPOLLERR|EPOLLHUP, {u32=7, u64=7}}], 200, 1000) = 1 09:41:17.270332 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10341860}) = 0 09:41:17.270340 getsockopt(7, SOL_SOCKET, SO_ERROR, [111], [4]) = 0 09:41:17.270367 close(7) = 0 This may be backported to 2.1 and 2.0 though it's unlikely to bring any user-visible improvement except to clarify debugging.
2019-12-11 02:12:23 -05:00
wake_expired_tasks();
/* check if we caught some signals and process them in the
first thread */
if (signal_queue_len && tid == 0) {
activity[tid].wake_signal++;
signal_process_queue();
}
/* Process a few tasks */
process_runnable_tasks();
/* also stop if we failed to cleanly stop all tasks */
if (killed > 1)
break;
/* expire immediately if events or signals are pending */
wake = 1;
if (thread_has_tasks())
activity[tid].wake_tasks++;
else {
_HA_ATOMIC_OR(&th_ctx->flags, TH_FL_SLEEPING);
_HA_ATOMIC_AND(&th_ctx->flags, ~TH_FL_NOTIFIED);
__ha_barrier_atomic_store();
if (thread_has_tasks()) {
activity[tid].wake_tasks++;
_HA_ATOMIC_AND(&th_ctx->flags, ~TH_FL_SLEEPING);
} else if (signal_queue_len) {
/* this check is required after setting TH_FL_SLEEPING to avoid
* a race with wakeup on signals using wake_threads() */
_HA_ATOMIC_AND(&th_ctx->flags, ~TH_FL_SLEEPING);
} else
wake = 0;
}
if (!wake) {
BUG/MINOR: soft-stop: always wake up waiting threads on stopping Currently the soft-stop can lead to old processes remaining alive for as long as two seconds after receiving a soft-stop signal. What happens is that when receiving SIGUSR1, one thread (usually the first one) wakes up, handles the signal, sets "stopping", goes into runn_poll_loop(), and discovers that stopping is set, so its also sets itself in the stopping_thread_mask bit mask. After this it sees that other threads are not yet willing to stop, so it continues to wait. From there, other threads which were waiting in poll() expire after one second on poll timeout and enter run_poll_loop() in turn. That's already one second of wait time. They discover each in turn that they're stopping and see that other threads are not yet stopping, so they go back waiting. After the end of the first second, all threads know they're stopping and have set their bit in stopping_thread_mask. It's only now that those who started to wait first wake up again on timeout to discover that all other ones are stopping, and can now quit. One second later all threads will have done it and the process will quit. This is effectively strictly larger than one second and up to two seconds. What the current patch does is simple, when the first thread stops, it sets its own bit into stopping_thread_mask then wakes up all other threads to do also set theirs. This kills the first second which corresponds to the time to discover the stopping state. Second, when a thread exists, it wakes all other ones again because some might have gone back sleeping waiting for "jobs" to go down to zero (i.e. closing the last connection). This kills the last second of wait time. Thanks to this, as SIGUSR1 now acts instantly again if there's no active connection, or it stops immediately after the last connection has left if one was still present. This should be backported as far as 2.0.
2020-05-13 07:51:01 -04:00
int i;
if (stopping) {
/* stop muxes/quic-conns before acknowledging stopping */
if (!(tg_ctx->stopping_threads & ti->ltid_bit)) {
task_wakeup(mux_stopping_data[tid].task, TASK_WOKEN_OTHER);
wake = 1;
}
if (_HA_ATOMIC_OR_FETCH(&tg_ctx->stopping_threads, ti->ltid_bit) == ti->ltid_bit &&
_HA_ATOMIC_OR_FETCH(&stopping_tgroup_mask, tg->tgid_bit) == tg->tgid_bit) {
/* first one to detect it, notify all threads that stopping was just set */
for (i = 0; i < global.nbthread; i++) {
if (_HA_ATOMIC_LOAD(&ha_thread_info[i].tg->threads_enabled) &
ha_thread_info[i].ltid_bit &
~_HA_ATOMIC_LOAD(&ha_thread_info[i].tg_ctx->stopping_threads))
wake_thread(i);
}
}
BUG/MINOR: soft-stop: always wake up waiting threads on stopping Currently the soft-stop can lead to old processes remaining alive for as long as two seconds after receiving a soft-stop signal. What happens is that when receiving SIGUSR1, one thread (usually the first one) wakes up, handles the signal, sets "stopping", goes into runn_poll_loop(), and discovers that stopping is set, so its also sets itself in the stopping_thread_mask bit mask. After this it sees that other threads are not yet willing to stop, so it continues to wait. From there, other threads which were waiting in poll() expire after one second on poll timeout and enter run_poll_loop() in turn. That's already one second of wait time. They discover each in turn that they're stopping and see that other threads are not yet stopping, so they go back waiting. After the end of the first second, all threads know they're stopping and have set their bit in stopping_thread_mask. It's only now that those who started to wait first wake up again on timeout to discover that all other ones are stopping, and can now quit. One second later all threads will have done it and the process will quit. This is effectively strictly larger than one second and up to two seconds. What the current patch does is simple, when the first thread stops, it sets its own bit into stopping_thread_mask then wakes up all other threads to do also set theirs. This kills the first second which corresponds to the time to discover the stopping state. Second, when a thread exists, it wakes all other ones again because some might have gone back sleeping waiting for "jobs" to go down to zero (i.e. closing the last connection). This kills the last second of wait time. Thanks to this, as SIGUSR1 now acts instantly again if there's no active connection, or it stops immediately after the last connection has left if one was still present. This should be backported as far as 2.0.
2020-05-13 07:51:01 -04:00
}
/* stop when there's nothing left to do */
if ((jobs - unstoppable_jobs) == 0 &&
(_HA_ATOMIC_LOAD(&stopping_tgroup_mask) & all_tgroups_mask) == all_tgroups_mask) {
/* check that all threads are aware of the stopping status */
for (i = 0; i < global.nbtgroups; i++)
if ((_HA_ATOMIC_LOAD(&ha_tgroup_ctx[i].stopping_threads) &
_HA_ATOMIC_LOAD(&ha_tgroup_info[i].threads_enabled)) !=
_HA_ATOMIC_LOAD(&ha_tgroup_info[i].threads_enabled))
break;
#ifdef USE_THREAD
if (i == global.nbtgroups) {
/* all are OK, let's wake them all and stop */
for (i = 0; i < global.nbthread; i++)
if (i != tid && _HA_ATOMIC_LOAD(&ha_thread_info[i].tg->threads_enabled) & ha_thread_info[i].ltid_bit)
wake_thread(i);
break;
}
#endif
BUG/MINOR: soft-stop: always wake up waiting threads on stopping Currently the soft-stop can lead to old processes remaining alive for as long as two seconds after receiving a soft-stop signal. What happens is that when receiving SIGUSR1, one thread (usually the first one) wakes up, handles the signal, sets "stopping", goes into runn_poll_loop(), and discovers that stopping is set, so its also sets itself in the stopping_thread_mask bit mask. After this it sees that other threads are not yet willing to stop, so it continues to wait. From there, other threads which were waiting in poll() expire after one second on poll timeout and enter run_poll_loop() in turn. That's already one second of wait time. They discover each in turn that they're stopping and see that other threads are not yet stopping, so they go back waiting. After the end of the first second, all threads know they're stopping and have set their bit in stopping_thread_mask. It's only now that those who started to wait first wake up again on timeout to discover that all other ones are stopping, and can now quit. One second later all threads will have done it and the process will quit. This is effectively strictly larger than one second and up to two seconds. What the current patch does is simple, when the first thread stops, it sets its own bit into stopping_thread_mask then wakes up all other threads to do also set theirs. This kills the first second which corresponds to the time to discover the stopping state. Second, when a thread exists, it wakes all other ones again because some might have gone back sleeping waiting for "jobs" to go down to zero (i.e. closing the last connection). This kills the last second of wait time. Thanks to this, as SIGUSR1 now acts instantly again if there's no active connection, or it stops immediately after the last connection has left if one was still present. This should be backported as far as 2.0.
2020-05-13 07:51:01 -04:00
}
}
MINOR: tasks: split wake_expired_tasks() in two parts to avoid useless wakeups We used to have wake_expired_tasks() wake up tasks and return the next expiration delay. The problem this causes is that we have to call it just before poll() in order to consider latest timers, but this also means that we don't wake up all newly expired tasks upon return from poll(), which thus systematically requires a second poll() round. This is visible when running any scheduled task like a health check, as there are systematically two poll() calls, one with the interval, nothing is done after it, and another one with a zero delay, and the task is called: listen test bind *:8001 server s1 127.0.0.1:1111 check 09:37:38.200959 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8696843}) = 0 09:37:38.200967 epoll_wait(3, [], 200, 1000) = 0 09:37:39.202459 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8712467}) = 0 >> nothing run here, as the expired task was not woken up yet. 09:37:39.202497 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8715766}) = 0 09:37:39.202505 epoll_wait(3, [], 200, 0) = 0 09:37:39.202513 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8719064}) = 0 >> now the expired task was woken up 09:37:39.202522 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 7 09:37:39.202537 fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 09:37:39.202565 setsockopt(7, SOL_TCP, TCP_NODELAY, [1], 4) = 0 09:37:39.202577 setsockopt(7, SOL_TCP, TCP_QUICKACK, [0], 4) = 0 09:37:39.202585 connect(7, {sa_family=AF_INET, sin_port=htons(1111), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress) 09:37:39.202659 epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLOUT, {u32=7, u64=7}}) = 0 09:37:39.202673 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8814713}) = 0 09:37:39.202683 epoll_wait(3, [{EPOLLOUT|EPOLLERR|EPOLLHUP, {u32=7, u64=7}}], 200, 1000) = 1 09:37:39.202693 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=8818617}) = 0 09:37:39.202701 getsockopt(7, SOL_SOCKET, SO_ERROR, [111], [4]) = 0 09:37:39.202715 close(7) = 0 Let's instead split the function in two parts: - the first part, wake_expired_tasks(), called just before process_runnable_tasks(), wakes up all expired tasks; it doesn't compute any timeout. - the second part, next_timer_expiry(), called just before poll(), only computes the next timeout for the current thread. Thanks to this, all expired tasks are properly woken up when leaving poll, and each poll call's timeout remains up to date: 09:41:16.270449 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10223556}) = 0 09:41:16.270457 epoll_wait(3, [], 200, 999) = 0 09:41:17.270130 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10238572}) = 0 09:41:17.270157 socket(AF_INET, SOCK_STREAM, IPPROTO_TCP) = 7 09:41:17.270194 fcntl(7, F_SETFL, O_RDONLY|O_NONBLOCK) = 0 09:41:17.270204 setsockopt(7, SOL_TCP, TCP_NODELAY, [1], 4) = 0 09:41:17.270216 setsockopt(7, SOL_TCP, TCP_QUICKACK, [0], 4) = 0 09:41:17.270224 connect(7, {sa_family=AF_INET, sin_port=htons(1111), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress) 09:41:17.270299 epoll_ctl(3, EPOLL_CTL_ADD, 7, {EPOLLOUT, {u32=7, u64=7}}) = 0 09:41:17.270314 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10337841}) = 0 09:41:17.270323 epoll_wait(3, [{EPOLLOUT|EPOLLERR|EPOLLHUP, {u32=7, u64=7}}], 200, 1000) = 1 09:41:17.270332 clock_gettime(CLOCK_THREAD_CPUTIME_ID, {tv_sec=0, tv_nsec=10341860}) = 0 09:41:17.270340 getsockopt(7, SOL_SOCKET, SO_ERROR, [111], [4]) = 0 09:41:17.270367 close(7) = 0 This may be backported to 2.1 and 2.0 though it's unlikely to bring any user-visible improvement except to clarify debugging.
2019-12-11 02:12:23 -05:00
/* If we have to sleep, measure how long */
next = wake ? TICK_ETERNITY : next_timer_expiry();
/* The poller will ensure it returns around <next> */
cur_poller.poll(&cur_poller, next, wake);
activity[tid].loops++;
}
_HA_ATOMIC_AND(&th_ctx->flags, ~TH_FL_IN_LOOP);
}
static void *run_thread_poll_loop(void *data)
{
struct per_thread_alloc_fct *ptaf;
struct per_thread_init_fct *ptif;
struct per_thread_deinit_fct *ptdf;
struct per_thread_free_fct *ptff;
static int init_left = 0;
__decl_thread(static pthread_mutex_t init_mutex = PTHREAD_MUTEX_INITIALIZER);
__decl_thread(static pthread_cond_t init_cond = PTHREAD_COND_INITIALIZER);
ha_set_thread(data);
set_thread_cpu_affinity();
clock_set_local_source();
#ifdef USE_THREAD
ha_thread_info[tid].pth_id = ha_get_pthread_id(tid);
#endif
ha_thread_info[tid].stack_top = __builtin_frame_address(0);
BUG/MEDIUM: thread: consider secondary threads as idle+harmless during boot idle and harmless bits in the tgroup_ctx structure were not explicitly set during boot. | struct tgroup_ctx ha_tgroup_ctx[MAX_TGROUPS] = { }; As the structure is first statically initialized, .threads_harmless and .threads_idle are automatically zero- initialized by the compiler. Unfortulately, this means that such threads are not considered idle nor harmless by thread_isolate(_full)() functions until they enter the polling loop (thread_harmless_now() and thread_idle_now() are respectively called before entering the polling loop) Because of this, any attempt to call thread_isolate() or thread_isolate_full() during a startup phase with nbthreads >= 2 will cause thread_isolate to loop until every secondary threads make it through their first polling loop. If the startup phase is aborted during boot (ie: "-c" option to check the configuration), secondary threads may be initialized but will never be started (ie: they won't enter the polling loop), thus thread_isolate() could would loop forever in such cases. We can easily reveal the bug with this patch reproducer: | diff --git a/src/haproxy.c b/src/haproxy.c | index e91691658..0b733f6ee 100644 | --- a/src/haproxy.c | +++ b/src/haproxy.c | @@ -2317,6 +2317,10 @@ static void init(int argc, char **argv) | if (pr || px) { | /* At least one peer or one listener has been found */ | qfprintf(stdout, "Configuration file is valid\n"); | + printf("haproxy will loop...\n"); | + thread_isolate(); | + printf("we will never reach this\n"); | + thread_release(); | deinit_and_exit(0); | } | qfprintf(stdout, "Configuration file has no error but will not start (no listener) => exit(2).\n"); Now we start haproxy with a valid config: $> haproxy -c -f valid.conf Configuration file is valid haproxy will loop... ^C ------------------------------------------------------------------------------ This did not cause any issue so far because no early deinit paths require full thread isolation. But this may change when new features or requirements are introduced, so we should fix this before it becomes a real issue. To fix this, we explicitly assign .threads_harmless and .threads_idle to .threads_enabled value in thread_map_to_groups() function during boot. This is the proper place to do this since as long as .threads_enabled is not explicitly set, its default value is also 0 (zero-initialized by the compiler) code snippet from thread_isolate() function: ulong te = _HA_ATOMIC_LOAD(&ha_tgroup_info[tgrp].threads_enabled); ulong th = _HA_ATOMIC_LOAD(&ha_tgroup_ctx[tgrp].threads_harmless); if ((th & te) == te) break; Thus thread_isolate(_full()) won't be looping forever in thread_isolate() even if it were to be used before thread_map_to_groups() is executed. No backport needed unless this is a requirement.
2023-01-27 09:13:28 -05:00
/* thread is started, from now on it is not idle nor harmless */
thread_harmless_end();
thread_idle_end();
_HA_ATOMIC_OR(&th_ctx->flags, TH_FL_STARTED);
/* Now, initialize one thread init at a time. This is better since
* some init code is a bit tricky and may release global resources
* after reallocating them locally. This will also ensure there is
* no race on file descriptors allocation.
*/
#ifdef USE_THREAD
pthread_mutex_lock(&init_mutex);
#endif
/* The first thread must set the number of threads left */
if (!init_left)
init_left = global.nbthread;
init_left--;
clock_init_thread_date();
/* per-thread alloc calls performed here are not allowed to snoop on
* other threads, so they are free to initialize at their own rhythm
* as long as they act as if they were alone. None of them may rely
* on resources initialized by the other ones.
*/
list_for_each_entry(ptaf, &per_thread_alloc_list, list) {
if (!ptaf->fct()) {
ha_alert("failed to allocate resources for thread %u.\n", tid);
#ifdef USE_THREAD
pthread_mutex_unlock(&init_mutex);
#endif
exit(1);
}
}
/* per-thread init calls performed here are not allowed to snoop on
* other threads, so they are free to initialize at their own rhythm
* as long as they act as if they were alone.
*/
list_for_each_entry(ptif, &per_thread_init_list, list) {
if (!ptif->fct()) {
ha_alert("failed to initialize thread %u.\n", tid);
#ifdef USE_THREAD
pthread_mutex_unlock(&init_mutex);
#endif
exit(1);
}
}
BUG/MEDIUM: init/threads: prevent initialized threads from starting before others Since commit 6ec902a ("MINOR: threads: serialize threads initialization") we now serialize threads initialization. But doing so has emphasized another race which is that some threads may actually start the loop before others are done initializing. As soon as all threads enter the first thread_release() call, their rdv bit is cleared and they're all waiting for all others' rdv to be cleared as well, with their harmless bit set. The first one to notice the cleared mask will progress through thread_isolate(), take rdv again preventing most others from noticing its short pass to zero, and this first one will be able to run all the way through the initialization till the last call to thread_release() which it happily crosses, being the only one with the rdv bit, leaving the room for one or a few others to do the same. This results in some threads entering the loop before others are done with their initialization, which is particularly bad. PiBa-NL reported that some regtests fail for him due to this (which was impossible to reproduce here, but races are racy by definition). However placing some printf() in the initialization code definitely shows this unsychronized startup. This patch takes a different approach in three steps : - first, we don't start with thread_release() anymore and we don't set the rdv mask anymore in the main call. This was initially done to let all threads start toghether, which we don't want. Instead we just start with thread_isolate(). Since all threads are harmful by default, they all wait for each other's readiness before starting. - second, we don't release with thread_release() but with thread_sync_release(), meaning that we don't leave the function until other ones have reached the point in the function where they decide to leave it as well. - third, it makes sure we don't start the listeners using protocol_enable_all() before all threads have allocated their local FD tables or have initialized their pollers, otherwise startup could be racy as well. It's worth noting that it is even possible to limit this call to thread #0 as it only needs to be performed once. This now guarantees that all thread init calls start only after all threads are ready, and that no thread enters the polling loop before all others have completed their initialization. Please check GH issues #111 and #117 for more context. No backport is needed, though if some new init races are reported in 1.9 (or even 1.8) which do not affect 2.0, then it may make sense to carefully backport this small series.
2019-06-10 03:51:04 -04:00
/* enabling protocols will result in fd_insert() calls to be performed,
* we want all threads to have already allocated their local fd tables
* before doing so, thus only the last thread does it.
BUG/MEDIUM: init/threads: prevent initialized threads from starting before others Since commit 6ec902a ("MINOR: threads: serialize threads initialization") we now serialize threads initialization. But doing so has emphasized another race which is that some threads may actually start the loop before others are done initializing. As soon as all threads enter the first thread_release() call, their rdv bit is cleared and they're all waiting for all others' rdv to be cleared as well, with their harmless bit set. The first one to notice the cleared mask will progress through thread_isolate(), take rdv again preventing most others from noticing its short pass to zero, and this first one will be able to run all the way through the initialization till the last call to thread_release() which it happily crosses, being the only one with the rdv bit, leaving the room for one or a few others to do the same. This results in some threads entering the loop before others are done with their initialization, which is particularly bad. PiBa-NL reported that some regtests fail for him due to this (which was impossible to reproduce here, but races are racy by definition). However placing some printf() in the initialization code definitely shows this unsychronized startup. This patch takes a different approach in three steps : - first, we don't start with thread_release() anymore and we don't set the rdv mask anymore in the main call. This was initially done to let all threads start toghether, which we don't want. Instead we just start with thread_isolate(). Since all threads are harmful by default, they all wait for each other's readiness before starting. - second, we don't release with thread_release() but with thread_sync_release(), meaning that we don't leave the function until other ones have reached the point in the function where they decide to leave it as well. - third, it makes sure we don't start the listeners using protocol_enable_all() before all threads have allocated their local FD tables or have initialized their pollers, otherwise startup could be racy as well. It's worth noting that it is even possible to limit this call to thread #0 as it only needs to be performed once. This now guarantees that all thread init calls start only after all threads are ready, and that no thread enters the polling loop before all others have completed their initialization. Please check GH issues #111 and #117 for more context. No backport is needed, though if some new init races are reported in 1.9 (or even 1.8) which do not affect 2.0, then it may make sense to carefully backport this small series.
2019-06-10 03:51:04 -04:00
*/
if (init_left == 0)
protocol_enable_all();
#ifdef USE_THREAD
pthread_cond_broadcast(&init_cond);
pthread_mutex_unlock(&init_mutex);
/* now wait for other threads to finish starting */
pthread_mutex_lock(&init_mutex);
while (init_left)
pthread_cond_wait(&init_cond, &init_mutex);
pthread_mutex_unlock(&init_mutex);
#endif
#if defined(PR_SET_NO_NEW_PRIVS) && defined(USE_PRCTL)
/* Let's refrain from using setuid executables. This way the impact of
* an eventual vulnerability in a library remains limited. It may
* impact external checks but who cares about them anyway ? In the
* worst case it's possible to disable the option. Obviously we do this
* in workers only. We can't hard-fail on this one as it really is
* implementation dependent though we're interested in feedback, hence
* the warning.
*/
if (!(global.tune.options & GTUNE_INSECURE_SETUID) && !master) {
static int warn_fail;
if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) == -1 && !_HA_ATOMIC_FETCH_ADD(&warn_fail, 1)) {
ha_warning("Failed to disable setuid, please report to developers with detailed "
"information about your operating system. You can silence this warning "
"by adding 'insecure-setuid-wanted' in the 'global' section.\n");
}
}
#endif
MEDIUM: init: prevent process and thread creation at runtime Some concerns are regularly raised about the risk to inherit some Lua files which make use of a fork (e.g. via os.execute()) as well as whether or not some of bugs we fix might or not be exploitable to run some code. Given that haproxy is event-driven, any foreground activity completely stops processing and is easy to detect, but background activity is a different story. A Lua script could very well discretely fork a sub-process connecting to a remote location and taking commands, and some injected code could also try to hide its activity by creating a process or a thread without blocking the rest of the processing. While such activities should be extremely limited when run in an empty chroot without any permission, it would be better to get a higher assurance they cannot happen. This patch introduces something very simple: it limits the number of processes and threads to zero in the workers after the last thread was created. By doing so, it effectively instructs the system to fail on any fork() or clone() syscall. Thus any undesired activity has to happen in the foreground and is way easier to detect. This will obviously break external checks (whose concept is already totally insecure), and for this reason a new option "insecure-fork-wanted" was added to disable this protection, and it is suggested in the fork() error report from the checks. It is obviously recommended not to use it and to reconsider the reasons leading to it being enabled in the first place. If for any reason we fail to disable forks, we still start because it could be imaginable that some operating systems refuse to set this limit to zero, but in this case we emit a warning, that may or may not be reported since we're after the fork point. Ideally over the long term it should be conditionned by strict-limits and cause a hard fail.
2019-12-03 01:07:36 -05:00
#if defined(RLIMIT_NPROC)
/* all threads have started, it's now time to prevent any new thread
* or process from starting. Obviously we do this in workers only. We
* can't hard-fail on this one as it really is implementation dependent
* though we're interested in feedback, hence the warning.
*/
if (!(global.tune.options & GTUNE_INSECURE_FORK) && !master) {
struct rlimit limit = { .rlim_cur = 0, .rlim_max = 0 };
static int warn_fail;
if (setrlimit(RLIMIT_NPROC, &limit) == -1 && !_HA_ATOMIC_FETCH_ADD(&warn_fail, 1)) {
MEDIUM: init: prevent process and thread creation at runtime Some concerns are regularly raised about the risk to inherit some Lua files which make use of a fork (e.g. via os.execute()) as well as whether or not some of bugs we fix might or not be exploitable to run some code. Given that haproxy is event-driven, any foreground activity completely stops processing and is easy to detect, but background activity is a different story. A Lua script could very well discretely fork a sub-process connecting to a remote location and taking commands, and some injected code could also try to hide its activity by creating a process or a thread without blocking the rest of the processing. While such activities should be extremely limited when run in an empty chroot without any permission, it would be better to get a higher assurance they cannot happen. This patch introduces something very simple: it limits the number of processes and threads to zero in the workers after the last thread was created. By doing so, it effectively instructs the system to fail on any fork() or clone() syscall. Thus any undesired activity has to happen in the foreground and is way easier to detect. This will obviously break external checks (whose concept is already totally insecure), and for this reason a new option "insecure-fork-wanted" was added to disable this protection, and it is suggested in the fork() error report from the checks. It is obviously recommended not to use it and to reconsider the reasons leading to it being enabled in the first place. If for any reason we fail to disable forks, we still start because it could be imaginable that some operating systems refuse to set this limit to zero, but in this case we emit a warning, that may or may not be reported since we're after the fork point. Ideally over the long term it should be conditionned by strict-limits and cause a hard fail.
2019-12-03 01:07:36 -05:00
ha_warning("Failed to disable forks, please report to developers with detailed "
"information about your operating system. You can silence this warning "
"by adding 'insecure-fork-wanted' in the 'global' section.\n");
}
}
#endif
run_poll_loop();
list_for_each_entry(ptdf, &per_thread_deinit_list, list)
ptdf->fct();
list_for_each_entry(ptff, &per_thread_free_list, list)
ptff->fct();
#ifdef USE_THREAD
if (!_HA_ATOMIC_AND_FETCH(&ha_tgroup_info[ti->tgid-1].threads_enabled, ~ti->ltid_bit))
_HA_ATOMIC_AND(&all_tgroups_mask, ~tg->tgid_bit);
if (!_HA_ATOMIC_AND_FETCH(&tg_ctx->stopping_threads, ~ti->ltid_bit))
_HA_ATOMIC_AND(&stopping_tgroup_mask, ~tg->tgid_bit);
if (tid > 0)
pthread_exit(NULL);
#endif
return NULL;
}
/* set uid/gid depending on global settings */
static void set_identity(const char *program_name)
{
int from_uid __maybe_unused = geteuid();
if (global.gid) {
if (getgroups(0, NULL) > 0 && setgroups(0, NULL) == -1)
ha_warning("[%s.main()] Failed to drop supplementary groups. Using 'gid'/'group'"
" without 'uid'/'user' is generally useless.\n", program_name);
if (setgid(global.gid) == -1) {
ha_alert("[%s.main()] Cannot set gid %d.\n", program_name, global.gid);
protocol_unbind_all();
exit(1);
}
}
#if defined(USE_LINUX_CAP)
if (prepare_caps_for_setuid(from_uid, global.uid) < 0) {
ha_alert("[%s.main()] Cannot switch uid to %d.\n", program_name, global.uid);
protocol_unbind_all();
exit(1);
}
#endif
if (global.uid && setuid(global.uid) == -1) {
ha_alert("[%s.main()] Cannot set uid %d.\n", program_name, global.uid);
protocol_unbind_all();
exit(1);
}
#if defined(USE_LINUX_CAP)
if (finalize_caps_after_setuid(from_uid, global.uid) < 0) {
ha_alert("[%s.main()] Cannot switch uid to %d.\n", program_name, global.uid);
protocol_unbind_all();
exit(1);
}
#endif
}
int main(int argc, char **argv)
{
int err, retry;
struct rlimit limit;
int pidfd = -1;
int intovf = (unsigned char)argc + 1; /* let the compiler know it's strictly positive */
/* Catch broken toolchains */
if (sizeof(long) != sizeof(void *) || (intovf + 0x7FFFFFFF >= intovf)) {
const char *msg;
if (sizeof(long) != sizeof(void *))
/* Apparently MingW64 was not made for us and can also break openssl */
msg = "The compiler this program was built with uses unsupported integral type sizes.\n"
"Most likely it follows the unsupported LLP64 model. Never try to link HAProxy\n"
"against libraries built with that compiler either! Please only use a compiler\n"
"producing ILP32 or LP64 programs for both programs and libraries.\n";
else if (intovf + 0x7FFFFFFF >= intovf)
/* Catch forced CFLAGS that miss 2-complement integer overflow */
msg = "The source code was miscompiled by the compiler, which usually indicates that\n"
"some of the CFLAGS needed to work around overzealous compiler optimizations\n"
"were overwritten at build time. Please do not force CFLAGS, and read Makefile\n"
"and INSTALL files to decide on the best way to pass your local build options.\n";
else
msg = "Bug in the compiler bug detection code, please report it to developers!\n";
fprintf(stderr,
"FATAL ERROR: invalid code detected -- cannot go further, please recompile!\n"
"%s"
"\nBuild options :"
#ifdef BUILD_TARGET
"\n TARGET = " BUILD_TARGET
#endif
#ifdef BUILD_CPU
"\n CPU = " BUILD_CPU
#endif
#ifdef BUILD_CC
"\n CC = " BUILD_CC
#endif
#ifdef BUILD_CFLAGS
"\n CFLAGS = " BUILD_CFLAGS
#endif
#ifdef BUILD_OPTIONS
"\n OPTIONS = " BUILD_OPTIONS
#endif
#ifdef BUILD_DEBUG
"\n DEBUG = " BUILD_DEBUG
#endif
"\n\n", msg);
return 1;
}
setvbuf(stdout, NULL, _IONBF, 0);
/* take a copy of initial limits before we possibly change them */
getrlimit(RLIMIT_NOFILE, &limit);
if (limit.rlim_max == RLIM_INFINITY)
limit.rlim_max = limit.rlim_cur;
rlim_fd_cur_at_boot = limit.rlim_cur;
rlim_fd_max_at_boot = limit.rlim_max;
/* process all initcalls in order of potential dependency */
RUN_INITCALLS(STG_PREPARE);
RUN_INITCALLS(STG_LOCK);
RUN_INITCALLS(STG_REGISTER);
/* now's time to initialize early boot variables */
init_early(argc, argv);
/* handles argument parsing */
init_args(argc, argv);
RUN_INITCALLS(STG_ALLOC);
RUN_INITCALLS(STG_POOL);
/* some code really needs to have the trash properly allocated */
if (!trash.area) {
ha_alert("failed to initialize trash buffers.\n");
exit(1);
}
RUN_INITCALLS(STG_INIT);
/* this is the late init where the config is parsed */
init(argc, argv);
signal_register_fct(SIGQUIT, dump, SIGQUIT);
signal_register_fct(SIGUSR1, sig_soft_stop, SIGUSR1);
signal_register_fct(SIGHUP, sig_dump_state, SIGHUP);
signal_register_fct(SIGUSR2, NULL, 0);
/* Always catch SIGPIPE even on platforms which define MSG_NOSIGNAL.
* Some recent FreeBSD setups report broken pipes, and MSG_NOSIGNAL
* was defined there, so let's stay on the safe side.
*/
signal_register_fct(SIGPIPE, NULL, 0);
/* ulimits */
if (!global.rlimit_nofile)
global.rlimit_nofile = global.maxsock;
if (global.rlimit_nofile) {
limit.rlim_cur = global.rlimit_nofile;
limit.rlim_max = MAX(rlim_fd_max_at_boot, limit.rlim_cur);
if ((global.fd_hard_limit && limit.rlim_cur > global.fd_hard_limit) ||
raise_rlim_nofile(NULL, &limit) != 0) {
getrlimit(RLIMIT_NOFILE, &limit);
if (global.fd_hard_limit && limit.rlim_cur > global.fd_hard_limit)
limit.rlim_cur = global.fd_hard_limit;
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] Cannot raise FD limit to %d, limit is %d.\n",
argv[0], global.rlimit_nofile, (int)limit.rlim_cur);
exit(1);
}
else {
/* try to set it to the max possible at least */
limit.rlim_cur = limit.rlim_max;
if (global.fd_hard_limit && limit.rlim_cur > global.fd_hard_limit)
limit.rlim_cur = global.fd_hard_limit;
if (raise_rlim_nofile(&limit, &limit) == 0)
getrlimit(RLIMIT_NOFILE, &limit);
ha_warning("[%s.main()] Cannot raise FD limit to %d, limit is %d.\n",
argv[0], global.rlimit_nofile, (int)limit.rlim_cur);
global.rlimit_nofile = limit.rlim_cur;
}
}
}
if (global.rlimit_memmax) {
limit.rlim_cur = limit.rlim_max =
global.rlimit_memmax * 1048576ULL;
#ifdef RLIMIT_AS
if (setrlimit(RLIMIT_AS, &limit) == -1) {
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] Cannot fix MEM limit to %d megs.\n",
argv[0], global.rlimit_memmax);
exit(1);
}
else
ha_warning("[%s.main()] Cannot fix MEM limit to %d megs.\n",
argv[0], global.rlimit_memmax);
}
#else
if (setrlimit(RLIMIT_DATA, &limit) == -1) {
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] Cannot fix MEM limit to %d megs.\n",
argv[0], global.rlimit_memmax);
exit(1);
}
else
ha_warning("[%s.main()] Cannot fix MEM limit to %d megs.\n",
argv[0], global.rlimit_memmax);
}
#endif
}
/* Try to get the listeners FD from the previous process using
* _getsocks on the stat socket, it must never been done in wait mode
* and check mode
*/
if (old_unixsocket &&
!(global.mode & (MODE_MWORKER_WAIT|MODE_CHECK|MODE_CHECK_CONDITION))) {
if (strcmp("/dev/null", old_unixsocket) != 0) {
if (sock_get_old_sockets(old_unixsocket) != 0) {
ha_alert("Failed to get the sockets from the old process!\n");
if (!(global.mode & MODE_MWORKER))
exit(1);
}
}
}
/* We will loop at most 100 times with 10 ms delay each time.
* That's at most 1 second. We only send a signal to old pids
* if we cannot grab at least one port.
*/
retry = MAX_START_RETRIES;
err = ERR_NONE;
while (retry >= 0) {
struct timeval w;
err = protocol_bind_all(retry == 0 || nb_oldpids == 0);
/* exit the loop on no error or fatal error */
if ((err & (ERR_RETRYABLE|ERR_FATAL)) != ERR_RETRYABLE)
break;
if (nb_oldpids == 0 || retry == 0)
break;
/* FIXME-20060514: Solaris and OpenBSD do not support shutdown() on
* listening sockets. So on those platforms, it would be wiser to
* simply send SIGUSR1, which will not be undoable.
*/
if (tell_old_pids(SIGTTOU) == 0) {
/* no need to wait if we can't contact old pids */
retry = 0;
continue;
}
/* give some time to old processes to stop listening */
w.tv_sec = 0;
w.tv_usec = 10*1000;
select(0, NULL, NULL, NULL, &w);
retry--;
}
/* Note: protocol_bind_all() sends an alert when it fails. */
if ((err & ~ERR_WARN) != ERR_NONE) {
ha_alert("[%s.main()] Some protocols failed to start their listeners! Exiting.\n", argv[0]);
if (retry != MAX_START_RETRIES && nb_oldpids)
tell_old_pids(SIGTTIN);
protocol_unbind_all(); /* cleanup everything we can */
exit(1);
}
if (!(global.mode & MODE_MWORKER_WAIT) && listeners == 0) {
ha_alert("[%s.main()] No enabled listener found (check for 'bind' directives) ! Exiting.\n", argv[0]);
/* Note: we don't have to send anything to the old pids because we
* never stopped them. */
exit(1);
}
/* Ok, all listeners should now be bound, close any leftover sockets
* the previous process gave us, we don't need them anymore
*/
sock_drop_unused_old_sockets();
/* prepare pause/play signals */
signal_register_fct(SIGTTOU, sig_pause, SIGTTOU);
signal_register_fct(SIGTTIN, sig_listen, SIGTTIN);
/* MODE_QUIET can inhibit alerts and warnings below this line */
if (getenv("HAPROXY_MWORKER_REEXEC") != NULL) {
/* either stdin/out/err are already closed or should stay as they are. */
if ((global.mode & MODE_DAEMON)) {
/* daemon mode re-executing, stdin/stdout/stderr are already closed so keep quiet */
global.mode &= ~MODE_VERBOSE;
global.mode |= MODE_QUIET; /* ensure that we won't say anything from now */
}
} else {
if ((global.mode & MODE_QUIET) && !(global.mode & MODE_VERBOSE)) {
/* detach from the tty */
stdio_quiet(-1);
}
}
/* open log & pid files before the chroot */
if ((global.mode & MODE_DAEMON || global.mode & MODE_MWORKER) &&
!(global.mode & MODE_MWORKER_WAIT) && global.pidfile != NULL) {
unlink(global.pidfile);
pidfd = open(global.pidfile, O_CREAT | O_WRONLY | O_TRUNC, 0644);
if (pidfd < 0) {
ha_alert("[%s.main()] Cannot create pidfile %s\n", argv[0], global.pidfile);
if (nb_oldpids)
tell_old_pids(SIGTTIN);
protocol_unbind_all();
exit(1);
}
}
if ((global.last_checks & LSTCHK_NETADM) && global.uid) {
ha_alert("[%s.main()] Some configuration options require full privileges, so global.uid cannot be changed.\n"
"", argv[0]);
protocol_unbind_all();
exit(1);
}
/* If the user is not root, we'll still let them try the configuration
* but we inform them that unexpected behaviour may occur.
*/
if ((global.last_checks & LSTCHK_NETADM) && getuid())
ha_warning("[%s.main()] Some options which require full privileges"
" might not work well.\n"
"", argv[0]);
if ((global.mode & (MODE_MWORKER|MODE_DAEMON)) == 0) {
/* chroot if needed */
if (global.chroot != NULL) {
if (chroot(global.chroot) == -1 || chdir("/") == -1) {
ha_alert("[%s.main()] Cannot chroot(%s).\n", argv[0], global.chroot);
if (nb_oldpids)
tell_old_pids(SIGTTIN);
protocol_unbind_all();
exit(1);
}
}
}
if (nb_oldpids && !(global.mode & MODE_MWORKER_WAIT))
nb_oldpids = tell_old_pids(oldpids_sig);
/* send a SIGTERM to workers who have a too high reloads number */
if ((global.mode & MODE_MWORKER) && !(global.mode & MODE_MWORKER_WAIT))
mworker_kill_max_reloads(SIGTERM);
/* Note that any error at this stage will be fatal because we will not
* be able to restart the old pids.
*/
if ((global.mode & (MODE_MWORKER | MODE_DAEMON)) == 0)
set_identity(argv[0]);
/* check ulimits */
limit.rlim_cur = limit.rlim_max = 0;
getrlimit(RLIMIT_NOFILE, &limit);
if (limit.rlim_cur < global.maxsock) {
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] FD limit (%d) too low for maxconn=%d/maxsock=%d. "
"Please raise 'ulimit-n' to %d or more to avoid any trouble.\n",
argv[0], (int)limit.rlim_cur, global.maxconn, global.maxsock,
global.maxsock);
exit(1);
}
else
ha_alert("[%s.main()] FD limit (%d) too low for maxconn=%d/maxsock=%d. "
"Please raise 'ulimit-n' to %d or more to avoid any trouble.\n",
argv[0], (int)limit.rlim_cur, global.maxconn, global.maxsock,
global.maxsock);
}
if (global.prealloc_fd && fcntl((int)limit.rlim_cur - 1, F_GETFD) == -1) {
if (dup2(0, (int)limit.rlim_cur - 1) == -1)
ha_warning("[%s.main()] Unable to preallocate file descriptor %d : %s",
argv[0], (int)limit.rlim_cur - 1, strerror(errno));
else
close((int)limit.rlim_cur - 1);
}
/* update the ready date a last time to also account for final setup time */
clock_update_date(0, 1);
clock_adjust_now_offset();
ready_date = date;
if (global.mode & (MODE_DAEMON | MODE_MWORKER | MODE_MWORKER_WAIT)) {
int ret = 0;
int in_parent = 0;
int devnullfd = -1;
/*
* if daemon + mworker: must fork here to let a master
* process live in background before forking children
*/
if ((getenv("HAPROXY_MWORKER_REEXEC") == NULL)
&& (global.mode & MODE_MWORKER)
&& (global.mode & MODE_DAEMON)) {
ret = fork();
if (ret < 0) {
ha_alert("[%s.main()] Cannot fork.\n", argv[0]);
protocol_unbind_all();
exit(1); /* there has been an error */
} else if (ret > 0) { /* parent leave to daemonize */
exit(0);
} else /* change the process group ID in the child (master process) */
setsid();
}
/* if in master-worker mode, write the PID of the father */
if (global.mode & MODE_MWORKER) {
char pidstr[100];
snprintf(pidstr, sizeof(pidstr), "%d\n", (int)getpid());
if (pidfd >= 0)
DISGUISE(write(pidfd, pidstr, strlen(pidstr)));
}
/* the father launches the required number of processes */
if (!(global.mode & MODE_MWORKER_WAIT)) {
struct ring *tmp_startup_logs = NULL;
if (global.mode & MODE_MWORKER)
mworker_ext_launch_all();
/* at this point the worker must have his own startup_logs buffer */
tmp_startup_logs = startup_logs_dup(startup_logs);
ret = fork();
if (ret < 0) {
ha_alert("[%s.main()] Cannot fork.\n", argv[0]);
protocol_unbind_all();
exit(1); /* there has been an error */
}
else if (ret == 0) { /* child breaks here */
startup_logs_free(startup_logs);
startup_logs = tmp_startup_logs;
/* This one must not be exported, it's internal! */
unsetenv("HAPROXY_MWORKER_REEXEC");
ha_random_jump96(1);
}
else { /* parent here */
in_parent = 1;
if (pidfd >= 0 && !(global.mode & MODE_MWORKER)) {
char pidstr[100];
snprintf(pidstr, sizeof(pidstr), "%d\n", ret);
DISGUISE(write(pidfd, pidstr, strlen(pidstr)));
}
if (global.mode & MODE_MWORKER) {
struct mworker_proc *child;
ha_notice("New worker (%d) forked\n", ret);
/* find the right mworker_proc */
list_for_each_entry(child, &proc_list, list) {
if (child->reloads == 0 &&
child->options & PROC_O_TYPE_WORKER &&
child->pid == -1) {
child->timestamp = date.tv_sec;
child->pid = ret;
child->version = strdup(haproxy_version);
/* at this step the fd is bound for the worker, set it to -1 so
* it could be close in case of errors in mworker_cleanup_proc() */
child->ipc_fd[1] = -1;
break;
}
}
}
}
} else {
/* wait mode */
in_parent = 1;
}
/* close the pidfile both in children and father */
if (pidfd >= 0) {
//lseek(pidfd, 0, SEEK_SET); /* debug: emulate eglibc bug */
close(pidfd);
}
/* We won't ever use this anymore */
ha_free(&global.pidfile);
if (in_parent) {
if (global.mode & (MODE_MWORKER|MODE_MWORKER_WAIT)) {
master = 1;
if ((!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) &&
(global.mode & MODE_DAEMON)) {
/* detach from the tty, this is required to properly daemonize. */
if ((getenv("HAPROXY_MWORKER_REEXEC") == NULL))
stdio_quiet(-1);
global.mode &= ~MODE_VERBOSE;
global.mode |= MODE_QUIET; /* ensure that we won't say anything from now */
}
if (global.mode & MODE_MWORKER_WAIT) {
/* only the wait mode handles the master CLI */
mworker_loop();
} else {
#if defined(USE_SYSTEMD)
if (global.tune.options & GTUNE_USE_SYSTEMD)
sd_notifyf(0, "READY=1\nMAINPID=%lu\nSTATUS=Ready.\n", (unsigned long)getpid());
#endif
/* if not in wait mode, reload in wait mode to free the memory */
setenv("HAPROXY_LOAD_SUCCESS", "1", 1);
ha_notice("Loading success.\n");
proc_self->failedreloads = 0; /* reset the number of failure */
mworker_reexec_waitmode();
}
/* should never get there */
exit(EXIT_FAILURE);
}
#if defined(USE_OPENSSL) && !defined(OPENSSL_NO_DH)
ssl_free_dh();
#endif
exit(0); /* parent must leave */
}
/* child must never use the atexit function */
atexit_flag = 0;
/* close useless master sockets */
if (global.mode & MODE_MWORKER) {
struct mworker_proc *child, *it;
master = 0;
mworker_cli_proxy_stop();
/* free proc struct of other processes */
list_for_each_entry_safe(child, it, &proc_list, list) {
/* close the FD of the master side for all
* workers, we don't need to close the worker
* side of other workers since it's done with
* the bind_proc */
if (child->ipc_fd[0] >= 0) {
close(child->ipc_fd[0]);
child->ipc_fd[0] = -1;
}
if (child->options & PROC_O_TYPE_WORKER &&
child->reloads == 0 &&
child->pid == -1) {
/* keep this struct if this is our pid */
proc_self = child;
continue;
}
LIST_DELETE(&child->list);
mworker_free_child(child);
child = NULL;
}
}
BUG/MEDIUM: threads/mworker: fix a race on startup Marc Fournier reported an interesting case when using threads with the master-worker mode : sometimes, a listener would have its FD closed during startup. Sometimes it could even be health checks seeing this. What happens is that after the threads are created, and the pollers enabled on each threads, the master-worker pipe is registered, and at the same time a close() is performed on the write side of this pipe since the children must not use it. But since this is replicated in every thread, what happens is that the first thread closes the pipe, thus releases the FD, and the next thread starting a listener in parallel gets this FD reassigned. Then another thread closes the FD again, which this time corresponds to the listener. It can also happen with the health check sockets if they're started early enough. This patch splits the mworker_pipe_register() function in two, so that the close() of the write side of the FD is performed very early after the fork() and long before threads are created (we don't need to delay it anyway). Only the pipe registration is done in the threaded code since it is important that the pollers are properly allocated for this. The mworker_pipe_register() function now takes care of registering the pipe only once, and this is guaranteed by a new surrounding lock. The call to protocol_enable_all() looks fragile in theory since it scans the list of proxies and their listeners, though in practice all threads scan the same list and take the same locks for each listener so it's not possible that any of them escapes the process and finishes before all listeners are started. And the operation is idempotent. This fix must be backported to 1.8. Thanks to Marc for providing very detailed traces clearly showing the problem.
2018-01-23 13:01:49 -05:00
if (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) {
devnullfd = open("/dev/null", O_RDWR, 0);
if (devnullfd < 0) {
ha_alert("Cannot open /dev/null\n");
exit(EXIT_FAILURE);
}
}
/* Must chroot and setgid/setuid in the children */
/* chroot if needed */
if (global.chroot != NULL) {
if (chroot(global.chroot) == -1 || chdir("/") == -1) {
ha_alert("[%s.main()] Cannot chroot(%s).\n", argv[0], global.chroot);
if (nb_oldpids)
tell_old_pids(SIGTTIN);
protocol_unbind_all();
exit(1);
}
}
ha_free(&global.chroot);
set_identity(argv[0]);
/*
* This is only done in daemon mode because we might want the
* logs on stdout in mworker mode. If we're NOT in QUIET mode,
* we should now close the 3 first FDs to ensure that we can
* detach from the TTY. We MUST NOT do it in other cases since
* it would have already be done, and 0-2 would have been
* affected to listening sockets
*/
if ((global.mode & MODE_DAEMON) &&
(!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE))) {
/* detach from the tty */
stdio_quiet(devnullfd);
global.mode &= ~MODE_VERBOSE;
global.mode |= MODE_QUIET; /* ensure that we won't say anything from now */
}
pid = getpid(); /* update child's pid */
if (!(global.mode & MODE_MWORKER)) /* in mworker mode we don't want a new pgid for the children */
setsid();
fork_poller();
}
/* pass through every cli socket, and check if it's bound to
* the current process and if it exposes listeners sockets.
* Caution: the GTUNE_SOCKET_TRANSFER is now set after the fork.
* */
if (global.cli_fe) {
struct bind_conf *bind_conf;
list_for_each_entry(bind_conf, &global.cli_fe->conf.bind, by_fe) {
if (bind_conf->level & ACCESS_FD_LISTENERS) {
global.tune.options |= GTUNE_SOCKET_TRANSFER;
break;
}
}
}
/* Note that here we can't be in the parent/master anymore */
#if !defined(USE_THREAD) && defined(USE_CPU_AFFINITY)
if (ha_cpuset_count(&cpu_map[0].thread[0])) { /* only do this if the process has a CPU map */
#if defined(CPUSET_USE_CPUSET) || defined(__DragonFly__)
struct hap_cpuset *set = &cpu_map[0].thread[0];
sched_setaffinity(0, sizeof(set->cpuset), &set->cpuset);
#elif defined(__FreeBSD__)
struct hap_cpuset *set = &cpu_map[0].thread[0];
ret = cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_PID, -1, sizeof(set->cpuset), &set->cpuset);
#endif
}
#endif
/* try our best to re-enable core dumps depending on system capabilities.
* What is addressed here :
* - remove file size limits
* - remove core size limits
* - mark the process dumpable again if it lost it due to user/group
*/
if (global.tune.options & GTUNE_SET_DUMPABLE) {
limit.rlim_cur = limit.rlim_max = RLIM_INFINITY;
#if defined(RLIMIT_FSIZE)
if (setrlimit(RLIMIT_FSIZE, &limit) == -1) {
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] Failed to set the raise the maximum "
"file size.\n", argv[0]);
exit(1);
}
else
ha_warning("[%s.main()] Failed to set the raise the maximum "
"file size.\n", argv[0]);
}
#endif
#if defined(RLIMIT_CORE)
if (setrlimit(RLIMIT_CORE, &limit) == -1) {
if (global.tune.options & GTUNE_STRICT_LIMITS) {
ha_alert("[%s.main()] Failed to set the raise the core "
"dump size.\n", argv[0]);
exit(1);
}
else
ha_warning("[%s.main()] Failed to set the raise the core "
"dump size.\n", argv[0]);
}
#endif
#if defined(USE_PRCTL)
if (prctl(PR_SET_DUMPABLE, 1, 0, 0, 0) == -1)
ha_warning("[%s.main()] Failed to set the dumpable flag, "
"no core will be dumped.\n", argv[0]);
#elif defined(USE_PROCCTL)
{
int traceable = PROC_TRACE_CTL_ENABLE;
if (procctl(P_PID, getpid(), PROC_TRACE_CTL, &traceable) == -1)
ha_warning("[%s.main()] Failed to set the traceable flag, "
"no core will be dumped.\n", argv[0]);
}
#endif
}
global.mode &= ~MODE_STARTING;
reset_usermsgs_ctx();
/* start threads 2 and above */
setup_extra_threads(&run_thread_poll_loop);
/* when multithreading we need to let only the thread 0 handle the signals */
haproxy_unblock_signals();
/* Finally, start the poll loop for the first thread */
run_thread_poll_loop(&ha_thread_info[0]);
/* wait for all threads to terminate */
wait_for_threads_completion();
MINOR: haproxy: Make use of deinit_and_exit() for clean exits Particularly cleanly deinit() after a configuration check to clean up the output of valgrind which reports "possible losses" without a deinit() and does not with a deinit(), converting actual losses into proper hard losses which makes the whole stuff easier to analyze. As an example, given an example configuration of the following: frontend foo bind *:8080 mode http Running `haproxy -c -f cfg` within valgrind will report 4 possible losses: $ valgrind --leak-check=full ./haproxy -c -f ./example.cfg ==21219== Memcheck, a memory error detector ==21219== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al. ==21219== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info ==21219== Command: ./haproxy -c -f ./example.cfg ==21219== [WARNING] 165/001100 (21219) : config : missing timeouts for frontend 'foo'. | While not properly invalid, you will certainly encounter various problems | with such a configuration. To fix this, please ensure that all following | timeouts are set to a non-zero value: 'client', 'connect', 'server'. Warnings were found. Configuration file is valid ==21219== ==21219== HEAP SUMMARY: ==21219== in use at exit: 1,436,631 bytes in 130 blocks ==21219== total heap usage: 153 allocs, 23 frees, 1,447,758 bytes allocated ==21219== ==21219== 7 bytes in 1 blocks are possibly lost in loss record 5 of 54 ==21219== at 0x4C2DB8F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x5726489: strdup (strdup.c:42) ==21219== by 0x468FD9: bind_conf_alloc (listener.h:158) ==21219== by 0x468FD9: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 14 bytes in 1 blocks are possibly lost in loss record 9 of 54 ==21219== at 0x4C2DB8F: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x5726489: strdup (strdup.c:42) ==21219== by 0x468F9B: bind_conf_alloc (listener.h:154) ==21219== by 0x468F9B: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 128 bytes in 1 blocks are possibly lost in loss record 35 of 54 ==21219== at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x468F90: bind_conf_alloc (listener.h:152) ==21219== by 0x468F90: cfg_parse_listen (cfgparse-listen.c:557) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== 608 bytes in 1 blocks are possibly lost in loss record 46 of 54 ==21219== at 0x4C2FB55: calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so) ==21219== by 0x4B953A: create_listeners (listener.c:576) ==21219== by 0x4578F6: str2listener (cfgparse.c:192) ==21219== by 0x469039: cfg_parse_listen (cfgparse-listen.c:568) ==21219== by 0x459DF3: readcfgfile (cfgparse.c:2167) ==21219== by 0x5056CD: init (haproxy.c:2021) ==21219== by 0x418232: main (haproxy.c:3121) ==21219== ==21219== LEAK SUMMARY: ==21219== definitely lost: 0 bytes in 0 blocks ==21219== indirectly lost: 0 bytes in 0 blocks ==21219== possibly lost: 757 bytes in 4 blocks ==21219== still reachable: 1,435,874 bytes in 126 blocks ==21219== suppressed: 0 bytes in 0 blocks ==21219== Reachable blocks (those to which a pointer was found) are not shown. ==21219== To see them, rerun with: --leak-check=full --show-leak-kinds=all ==21219== ==21219== For counts of detected and suppressed errors, rerun with: -v ==21219== ERROR SUMMARY: 4 errors from 4 contexts (suppressed: 0 from 0) Re-running the same command with the patch applied will not report any losses any more: $ valgrind --leak-check=full ./haproxy -c -f ./example.cfg ==22124== Memcheck, a memory error detector ==22124== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al. ==22124== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info ==22124== Command: ./haproxy -c -f ./example.cfg ==22124== [WARNING] 165/001503 (22124) : config : missing timeouts for frontend 'foo'. | While not properly invalid, you will certainly encounter various problems | with such a configuration. To fix this, please ensure that all following | timeouts are set to a non-zero value: 'client', 'connect', 'server'. Warnings were found. Configuration file is valid ==22124== ==22124== HEAP SUMMARY: ==22124== in use at exit: 313,864 bytes in 82 blocks ==22124== total heap usage: 153 allocs, 71 frees, 1,447,758 bytes allocated ==22124== ==22124== LEAK SUMMARY: ==22124== definitely lost: 0 bytes in 0 blocks ==22124== indirectly lost: 0 bytes in 0 blocks ==22124== possibly lost: 0 bytes in 0 blocks ==22124== still reachable: 313,864 bytes in 82 blocks ==22124== suppressed: 0 bytes in 0 blocks ==22124== Reachable blocks (those to which a pointer was found) are not shown. ==22124== To see them, rerun with: --leak-check=full --show-leak-kinds=all ==22124== ==22124== For counts of detected and suppressed errors, rerun with: -v ==22124== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0) It might be worth investigating what exactly HAProxy does to lose pointers to the start of those 4 memory areas and then to be able to still free them during deinit(). If HAProxy is able to free them, they ideally should be "still reachable" and not "possibly lost".
2020-06-13 18:37:42 -04:00
deinit_and_exit(0);
}
/*
* Local variables:
* c-indent-level: 8
* c-basic-offset: 8
* End:
*/