From patchwork Sun Apr 5 12:03:45 2026 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Adolf Belka X-Patchwork-Id: 9601 Return-Path: Received: from mail01.ipfire.org (mail01.haj.ipfire.org [172.28.1.202]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519) (Client CN "mail01.haj.ipfire.org", Issuer "R12" (not verified)) by web04.haj.ipfire.org (Postfix) with ESMTPS id 4fpWNl2D0Zz3x95 for ; Sun, 05 Apr 2026 12:04:55 +0000 (UTC) Received: from mail02.haj.ipfire.org (mail02.haj.ipfire.org [172.28.1.201]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519) (Client CN "mail02.haj.ipfire.org", Issuer "E7" (not verified)) by mail01.ipfire.org (Postfix) with ESMTPS id 4fpWNl08MYz5mj for ; Sun, 05 Apr 2026 12:04:55 +0000 (UTC) Received: from mail02.haj.ipfire.org (localhost [IPv6:::1]) by mail02.haj.ipfire.org (Postfix) with ESMTP id 4fpWN24cDdz37DC for ; Sun, 05 Apr 2026 12:04:18 +0000 (UTC) X-Original-To: development@lists.ipfire.org Received: from mail01.ipfire.org (mail01.haj.ipfire.org [172.28.1.202]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519) (Client CN "mail01.haj.ipfire.org", Issuer "R12" (not verified)) by mail02.haj.ipfire.org (Postfix) with ESMTPS id 4fpWMs3cxxz36V2 for ; Sun, 05 Apr 2026 12:04:09 +0000 (UTC) Received: from [127.0.0.1] (localhost [127.0.0.1]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange x25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (No client certificate requested) by mail01.ipfire.org (Postfix) with ESMTPSA id 4fpWMr0d2Sz5hQ; Sun, 05 Apr 2026 12:04:08 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=ipfire.org; s=202003rsa; t=1775390648; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=wLbze7aVKJp6PgDIYLin4oIJZN0Du370/Q9Hv6ANqfI=; b=Iz3JgIRRqE3cHHcyO8e7Sdu+OPBJzXd3iWCX3x40Vt3oAUtiVOm6WXZuMGCRlX8R2RqFGg zoAXKVF6YsC72y+u2RLDN8bYmsB8AGRVVWlShwFEfmbte6sK6CmjE71vccg+7VWQW4cfHA X8ZscQlHKdnuHp16CmN5N0PEaYw1xzBMLoi9zR7Kbmsn55ZiErPKKVe9gvLvJkdZKqVx6L ybOtbCngJ0HmbW4MibZ+tbNNoIrP9WXJlhU6asRLT1obqO/hF0ECkK1UYVjpw+PYRZINRZ QMkS5VhY5XqqfA+1MH7EoVyTL32qQTritFtKNFWbHBUZ/TAqDFlO0T4wF4ZvqQ== DKIM-Signature: v=1; a=ed25519-sha256; c=relaxed/relaxed; d=ipfire.org; s=202003ed25519; t=1775390648; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=wLbze7aVKJp6PgDIYLin4oIJZN0Du370/Q9Hv6ANqfI=; b=3AOWKcN+l5GvgPn6NwjABT6n7Pw00W/netixUAci6fIz3tI2cDYdfpPSKw9rjJ6FOfmc3/ xFIp9r/6LWtHDmDQ== From: Adolf Belka To: development@lists.ipfire.org Cc: Adolf Belka Subject: [PATCH] keepalived: Update to version 2.3.4 Date: Sun, 5 Apr 2026 14:03:45 +0200 Message-ID: <20260405120354.137211-28-adolf.belka@ipfire.org> In-Reply-To: <20260405120354.137211-1-adolf.belka@ipfire.org> References: <20260405120354.137211-1-adolf.belka@ipfire.org> Precedence: list List-Id: List-Subscribe: , List-Unsubscribe: , List-Post: List-Help: Sender: Mail-Followup-To: MIME-Version: 1.0 - Update from version 2.3.1 to 2.3.4 - No change to rootfile - Changelog 2.3.4 Improvements core: properly restore process priorities after a reload. core: allow specifying iproute_usr_dir even if no iproute2 support. core: include network namespace name when error opening namespace fds. core: resolve lang warning when comparing ordering of function addresses. The only reason function addresses are compared is with a red-black tree to quickly convert a function address into its name. It clearlt isn’t a sandard thing to do to compare ordering of function addresses, but it this case it is quite valid. core: stop repeatedly calling getpid(). We only need to call getpid() once per process, and can then save the value. A PID of a given process is never going to change! core: add code to calculate maximum stack usage and use it for no_swap. When a process has no_Swap specified, if we want to ensure that the stack is resident in memory we need to know the maximum size that it is likely to grow to. This commit adds diagnostic code (usually disabled) to report maximum usage, so that the code can be updated to know, in advance, the maximum likely stack usage. core: set CLOEXEC flag on all file descriptors except stdin/stdout/stderr. core: set CLOEXEC flag on streams (fopen/popen). snmp: set CLOEXEC on file descriptors opened by snmp. snmp: use close_range() if available for closing snmp file descriptors. core: call close_range() if available before exec’ing scripts. Fixes build: fix snap build process. vrrp: fix segfault at reload when DBus re-enabled. If dbus as enabled, then a reload disabled it, and another reload re-enabled it, then keepalived would often segfault. This is resolved by clearing dbus_startup_completed when dbus is stopped. vrrp: fix track_process warn identified by -Wflex-array-member-not-at-end. notify: fix resolving group name to gid for scripts. If a group had a large number of members, the memory allocated for getgrnam_r() could be insufficient and the call fail. This commit now allocates as much memory as the size of the group file, which should be sufficient. ipvs: resolve infinity loop when SMTP_CHECKers have ‘host’ config. core: fix keepalived not coredumping after a reload. This made it very difficult to resolve segfaults occurring due to a reload. vrrp: document and fix specifying iproute_etc_dir and iproute_usr_dir. build: fix some RHEL 7 and friends compilation problems. core: fix memory leak in track_file. 2.3.3 New vrrp: don’t allow unicast instance without interface to have a VMAC. If the interface is not configured, we can’t know what interface to add the VMAC to. vrrp: Add setting IP_FREEBIND/IPV6_FREEBIND socket option. This allows creating and configuring unicast sockets before the configured source address is added to the system. core: add O_CLOEXEC flag to pidfiles. vrrp: Support logging rate-limiting specified by RFC 9568 vrrp: add option for address owner to drop received VRRP packets. RFC 9568 (and RFC 5798 and RFC 3768) state that an address owner must drop any received VRRP packets. The consequence of this is that if there is more than one VRRP instance configured with priority 255 then they will all be in master state simultaneously. It seems more sensible for such received packets to be processed normally, and all but the VRRP instance with the primary IP address will revert to backup state. RFC 9568 appears to allow more than one instance to have priority 255, since section 8.3.2 was changed from “No more than one router on the link is to be configured with priority 255, especially if preemption is set” (note the contradiction here) to “only a single VRRP Router on the link SHOULD be configured with priority 255” and then descibes the situation if there is more than one such router. keepalived defaults to processing received packets when the local priority is 255, but the option added by this patch allows working in accordance with the RFC, i.e. to drop any received packets. Improvements core: Allow building on very old systems with kernels < 3.15. Open file descriptor locks were introduced in Linux 3.15, so we cannot use that type of locking on systems with older kernels (e.g. CentOS 7, which of course is no longer supported). Since this problem only occurs on kernels no longer supported by keepalived this commit simply removes the file locking, rather than implementing a more comprehensive solution. It is expected that at some point, in order to simplify the code, support for kernels no longer supported by any of the main distros will be removed from keepalived. doc: add oldest distro versions with their EOL dates and kernel versions. docker: Install linux-headers pkg to build in Docker. vrrp: handle checking ip utility version properly with BusyBox. snap: Misc snap improvements. build: make default _FORTIFY_SOURCE setting 3. Various distros already use _FORTIFY_SOURCE=3 by default, so we should do so too. vrrp: check the iproute2 directories exist when read first file vrrp: create /etc/iproute2 directory if it doesn’t exist. vrrp: Restore priority 255 if duplicate address owner detected. The VRRP RFCs assume that only one device is configured as the address owned for any VRID. keepalived has extended functionality which detects if two (or more) systems are configured as the address owner (this is completely invalid configuration). To avoid multiple systems acting as address owner, and hence all of them remaining in master mode, keepalived will reduce an address owner’s priority to 254 if the other device configured as address owner does not go away. This commit restores the priority of a vrrp instance to 255 if it had reduced it to 254 to avoid multiple VRRP instances simultaneously advertising that they are the address owner. vrrp: Only reduce address owner priority if primary ip address lower. If a VRRP instance is configured as address owner and it detects another device also advertising it is the address owner, only initially reduce our priority if our primary IP address is lower than the other device’s primary IP address. vrrp: if duplicate address owners, reduce priority if other won’t. If a VRRP instance is configured as address owner and it detects another device also advertising it is the address owner, we don’t reduce our priority if our primary IP address is higher than the other device’s primary IP address. However, if the other system, with a lower primary IP address, won’t reduce its priority (e.g. it is not a keepalived implementation), then we will reduce our priority after a suitable time. vrrp: add more helpful log messages if duplicate address owner. vrrp: log rate-limited message if advert has no VIPs. vrrp: log rate-limited warning if VRRPv3 advert interval mismatch. vrrp: it is not an error if VIPs in advert do not match configured. We should accept a VRRP advert if the VIPs in an advert do not match our configuration, but just log a rate-limited warning. vrrp: update saved master address when receive high priority advert. If we are in master state and receive a higher priority advert, saving the new master address saves checking VIPs twice. vrrp: include source address in log after receiving a bad advert. vrrp: check that VIPs are not duplicated. vrrp: check TTL/HL and unicast source ip even when not checking VIPs. The checking of TTL/HL and unicast source ip was only being done if the VIPs were being checked, whereas they should be checked even if the VIPs are not being checked. vrrp: change rx_ttl_hop_limit to rx_ttl_hl. The name was confusing since it suggested the value was a limit. vrrp: identify unicast peer in unicast_peer block configuration errors. vrrp: detect and reject duplicate unicast_peers in configuration. vrrp: add logging a change of master when detailed logging enabled. vrrp: handle a reload with no more startup_delay. During the vrrp_startup_delay time, if keepalived is reloaded with no more startup_delay, the startup_delay is never timed out and all received adverts is discarded. The commit causes the startup_delay timer to be reinstated after a reload with no more startup_delay if the timer has not yet expired. vrrp: Skip running not idle vrrp scripts. When a vrrp script is to be run (initially or after specified interval), first it is checked if it’s in IDLE state. If not a log message is printed informing about skipping run due to script being either running or timed out. However despite not being idle the code continues to run new script process. In heavily loaded systems this caused running multiple instances of vrrp script at the same time. This patch brings back missing return, which was lost during refactoring. codeQL: update codeQL.yml vrrp: add checks that interface fault flags not inconsistent. When a fault is added in down_instance() or cleared in try_up_instance() check that the flag that is being modified is not already set or cleared, as appropriate. This check is enabled by configure option –enable-fault-flags-check. vrrp: use a fault flag if num_track_faults is non zero. It simplifies the code to set a fault flag is num_track_faults is non-zero and clear the flag if num_track_faults is zero. vrrp: don’t attempt to send advert if socket is closed. This avoids an unnecessary log message. vrrp: don’t have multiple tracking objects for a VRRP instance. The code did have separate tracking objects for dynamic and non dynamic tracking objects for a VRRP instance. It also would add an addition dynamic tracking object every time a tracked interface was created, causing down_instance() to be called multiple times when an interface was deleted and previous creations of the interface. Prior to the patch to add fault flag bits this resulted in the vrrp instance not coming back up after the interface was recreated. Ths issue of vrrp instances remaining in fault state after after deletion and re-creation of interfaces is now resolved. vrrp: delay deleting VMACs are parent interface is deleted. The interface structure needs to have the ifindex set for the first pass through the VRRP instances, but it must be unset when the VMACs are cleaned up. vrrp: don’t change link local IPv6 address when extra added to base if. If an additional link local address was added to the base interface of a VMAC, keepalived was changing the source address of adverts to be the new address. The commit makes keepalived change the source address if the one it is using is deleted. track: don’t overwrite track file at startup unless configured to. vrrp: allow interface up debounce timer to exceed 2 * advert interval. There was no need to limit the up debounce timer in the same way that the down debounce timer has to be limited, so this commit removes the 2 * advert interval upper limit. vrrp: update delayed start time on reload if vrrp_startup_delay changed. vrrp: ignore IPv6 tentative addresses. We can’t do anything with them, and they are not usable, so we now wait until we are notified that the address is no longer tentative before we consider using it. Fixes parser: Fix error handling for HEX_STR parsing in UDP_CHECK. Fixes an issue where HEX_STR values with a trailing 0xff were incorrectly treated as errors. This HEX_STR is used in UDP_CHECK configuration, particularly in the payload and require_reply fields. ipvs: Fix segfault when using track_file checker. ipvs: Fix delay_loop for TCP_CHECK. scheduler: Fix segfault caused double erase from child_pid rbtree. In a situation when a child was timed out, but not yet processed, the thread is THREAD_CHILD_TIMEOUT type and remains on ready queue. If it gets terminated in this state, it needs to be removed from rb tree child_pid and transitioned to THREAD_CHILD_TERMINATED, but without additional moving it to ready queue as it is already there. The erase from child_pid tree is required to clean up pid from not terminated childs tree, but it needs to be done exactly once as rb tree implementation is not guarded against double removal. Erasing or adding same element multiple times, leads to malformed red-black tree and segmentation faults. This patch removes double erase in described scenario. build: fix compilation failure if building without VMACs. vrrp: fix reading of iproute2 conf files when directories don’t exist. vrrp: fix segfault when instance has no interface configured. If a vrrp instance has no interface configured (so it is unicast), processing SIGUSR1 resulted ina segfault. vrrp: Don’t segfault if open_sockpool_socket() fails to open sockets. If a unicast VRRP instance is configured and the unicast_src_ip does not exist on the system, then the bind() fails and the sockets are not opened. This commit ensures that in that case vrrp->sockets is not dereferenced. This is not a real fix to the problem. We need to track the addition and removal of unicast_src_ip addresses, and enter fault state if the address in not configured, or when it is removed. vrrp: interface add should call setup_interface(). When an interface is (re-)added, setup_interface() should be called even if vrrp->flags is set (eg VRRP_FLAG_NOPREEMPT). vrrp: fix recreating a VMAC interface with IPv6. The sin6_scope_id was not being updated if a VMACs underlying interface were deleted and recreated, causing the bind() call to fail. This commit now correctly updates the sin6_scope_id field in mcast_daddr. vrrp: fix persistent FAULT state with use_vmac when interfaces renamed. If an existing base interface of a VMAC is renamed, delete the VMAC since the configure base interface no longer exists. When an existing interface is renamed to match the base interface for a VMAC for a VRRP instance, for IPv6 when the VMAC interface is created a link local address is added, so clear the NO_ADDRESS fault flag by calling try_up_instance(). For IPv4 we do not add an address to the VMAC when it is created, so will wait for notlink notifications of addresses on the base interface, which can then be added to be VMAC. vrrp: fix keepalived warning of ipsets specified without iptables. keepalived was warning that using ipsets had been specified but iptables had not been specified, even if ipsets had not been specified. 2.3.2 New all: add –ignore-sigint option. This is needed for running keepalived under GDB (see https://bugzilla.kernel.org/show_bug.cgi?id=9039#c8). vrrp: allow specifing interval amd timeout to milli-second resolution. Although running track_scripts too rapidly can have use cause heavy system load, there are use cases for being able to run scripts more frequently than 1 second, and also at intervals not in whole seconds. This commit adds the option to be able to specify the interval and timeout timers to a resolution in milli-seconds. Improvements vrrp: remove need for route to have configured interface to track it. If a virtual route did not have an interface configured, keepalived would log a warning saying that it could not track the route, and then would disable tracking of that route. It appears that it is not necessary to know the interface in order to track the route, and in any event the netlink message received after adding the route identifies the interface for the route if it is appropriate. So this commit removes the requirement to specify an interface in order to track a route. intall: Update INSTALL instructions - add openSUSE. ipvs: Retry ipvs_nl_send_message() in ipvs_getinfo(). If we have to call keepalived_modprobe() for the ip_vs module, on some distros (e.g. RHEL based ones but not Fedora) we need to call ipvs_nl_send_message() twice in ipvs_getinfo(), since the first call fails. On most distros keepalived_modprobe() does not need to be called, since calling genl_ctrl_resolve(sock, IPVS_GENL_NAME) loads the ip_vs module. core: improve error message for process event listen. all: Properly handle an include file name ending with ‘’. vrrp: Allow for Ethernet frame padding for short packets. Some network interface cards do not strip Ethernet frame padding before passing a packet to userspace (recvmesg()). keepalived checks the received packet length but wasn’t allowing for extra bytes to be received that were added as frame padding. This commit allows for frame padding to be received and not report an incorrect packet length. vrrp: Remove duplicate dumping of master advert interval. Don’t write master advert interval in keepalived.data twice when using VRRPv3 and the VRRP instance is in backup state. vrrp: Handle empty ipset names with vrrp_ipsets keyword. We now handle empty ipset names and return a config error. vrrp: handle empty iptables chain names - vrrp_iptables keyword. We now return an error if a chain name is empty. vrrp-ipvs: handle empty nftables chain names. We now return an error if a chain name is empty. vrrp: use configured vrrp ipset names rather than ignore them. vrrp: check configured vrrp ipset names are all different. If a pair of configured ipset names are the same, there will be an error when using the ipsets. This commits checks and logs an error if two ipset names are the same. core: remove some duplicate include files. core: ensure only one instance of keepalived can run per config_id. There was a window when keepalived starts up when if two (or more) instances were starting at the same time, they might not detect the other instance is running. This commit add advisort file locking on the PID files to ensure that only one instance can run at a time. vrrp: Duplicate/drop MLDv1 listener reports on VMACs. MLDv2 listener reports were being handled, but not MLDv1. This commit now adds handling of MLDv1 listener reports as well. all: Ensure pid file exists when respawning child process. If a child process is respawned, the old pidfile may or may not still exist. If it doesn’t exist, we need to recreate it. If it still exists we need to reset our file offset and truncate the file before re-wrighting it. all: better pidfile handling after reload. vrrp: add thread_timer_expired keyword as a synonym of timer_expired_backup. The release notes referred to thread_timer_expired, so it is added for completeness but logs a message to change the keyword to timer_expired_backup. bfd: use time_t to avoid implicit ptr type casting. This fixes an incompatible pointer type [-Wincompatible-pointer-types] issue when compiling keepalived with GCC 14 [1] in 32-bit architectures where time_t size is 64 bits. vrrp-ipvs: Stop setting SO_LINGER on TCP sockets. Setting SO_LINGER causes the close() call to block until the first of: 1. the ACK of the FIN is received 2. the SO_LINGER timeout expires Since the SO_LINGER timeout was set to 5 seconds, if the FIN or the subsequent ACK were lost, then keepalived would block for 5 seconds, which must not be allowed to happen. The only TCP sockets that keepalived opens are for TCP_CHECK, HTTP_GET, SSL_GET, SMTP_CHECK and sending notify emails. For all of these, for any data that keepalived sends it receives data in response, and so there is no purpose in using SO_LINGER. Removing setting SO_LINGER will stop the occasional ‘A thread timer expired 5.1nnnnn seconds ago’, as reported in issue #2271. all: use correct format specifier for time fields. 32 bit Debian uses a 32 bit TIMESIZE, whereas 32 bit Ubuntu uses a 64 bit TIMESIZE. This means that on 32 bit Ubuntu some time types need to be printed using “%lld”, whereas on 32 bit Debian, and on 64 bit systems “%ld” is what is needed. Using the wrong format specifier was causing compilation warnings on 32 bit Debian. The issue impacts printing time_t, struct timeval tv_sec and tv_usec and struct timespec tv_sec fields. Peversely, on a 32 bit system when TIMESIZE is 64, struct timeval tv_usec is 64 bits, whereas struct timesec tv_nsec is 32 bits. The commit adds configure time checking of the right format specifiers to use, and adds definitions PRI_time_t, PRI_tv_sec, PRI_ts_sec etc. core: update addattr_l to match current iproute2 code - almost. The alignment calculations were not coerect, so this commit updates addattr_l to match the iproute2 version, EXCEPT there appears to be 1 issue in the iproute2 code when NLMSG_ALIGN is used when RTA_ALIGN should be used. The difference is entirely cosmetic (at the moment) since the functionality of the 2 macros is currently identical. lib: add micro-second timers to memory allocation debugging. Previously the time was logged for memory allocation/freeing operations in seconds. When comparing when memory was allocated/freed to debugging logging via a log file, it was helpful, in terms of being able to identify the sequence of events, to have the time of memory allocations etc logged in micro-seconds. vrrp: on reload only configured track_script name was checked. On a reload, only the configured name of a track_script was being checked to see if the new config track_script matched the old config track_script. If the script to be executed were changed, but the configured named of the script were kept the same, then the status of the old script would be transferred to the new script, despite the scripts being completely different. This commit now checks that the script really is the same, in terms of the path, parameters and user executing the script. vrrp: On reload with addresses added to VRRP instance send 2nd GARPs. If garp_master_delay is non zero, then after a reload when VIPs are added to a VRRP instance in master state, as well as the initial block of GARP messages that are sent, the messages need to be repeated after garp_master_delay seconds. This commit adds sending the second block. vrrp: merge vrrp instance garp_pending and gna_pending flags. Combine garp_pending and gna_pending flags into a single flags; that is all that is necessary and simplifies the code. vrrp: Use timer threads for delayed sending of GARPs/GNAs. Previously whenever a VRRP instance send an advert, it checked to see if any more GARPs/GNAs were due to be sent, either for garp_master_delay or garp_master_refresh. Using timer threads removes the checking every time an advert is sent, and the relevant code is only triggered when a timer expires. vrrp: stop using alloc_strvec() for parsing rttables files. It was a good idea at the time, but is not really appropriate. The parsing can be done just as simply without using alloc_strvec(). all: stop “unmatched quotes” warning for quoted strings. If a line with a quoted string has unbalanced quote characters when parsed as a standard (not quoted) string, an innapropriate warning was issued for unmatches quotes. This commit now stops the warning. This commit is not elegant, and it would be appreciated if a neater solution could be found. If anyone has a better solution, please submit a pull request or raise an issue explaining the solution. all: change checking process name at reload to include not NULL checks. The code was using the reload variable as an indicator that prev_global_data was not NULL, and this was causing some static code analysers to to flag up NULL pointer dereferences. The patch explicitly checks whether prev_global_data is NULL or not, since this is synonymous with testing the reload variable. all: clear pointers to old data structures freed after reload. This means that if that if there is a subsequent reference to the old data via thoe old_global_data, or old{bfd,check,vrrp} pointers, it should cause a segfault rather than undefined behaviour. It will also make it more straightford to debug any problem should it occur. vrrp: update location of iproute config files. Since iproute2 version 3.3 the location of the config files has been configurable, with the default being /etc/iproute2. Since version 4.4 there has been an rt_tables.d sub-directory. Version 4.10 added an rt_protos.d sub-directory, and version 6.5 added a second directory (/usr/lib/iproute2 or /usr/lib64/iproute) which 6.7 changed to /usr/share/iproute2 as the default. No major distro appears to change the default locations, and the only distro that used verion 6.5 or 6.6 was Fedora 40, but that has now upgraded to 6.7 so we are not bothered with the /usr/lib* options. The two directories have configure options, and if they are not specified, configure attempts to get the locations from the ip-route man page or the ip executable. vrrp: Specify protocol for IP addresses that keepalived adds. This is similar to being able to specify a protocol of ip routes and rules. vrrp: Add configure option to update /etc/rt_addrprotos. If there is no keepalived entry in rt_addrprotos create an entry which is removed when keepalived terminates. This will allow ip address show to display the protocol of an address as “keepalived” rather than 0x12. vrrp: always add a keepalived entry to rt_addrprotos is none exists. doc: Some updates. Fixes vrrp: Handle a reload before vrrp_delayed_start has expired. If keepalived reloaded its configuration before a specified vrrp_startup_delay had expired, the startup_delay was never being timed out, and so all received adverts would be discarded. The commit caused the startup_delay timer to be reinstated after a reload if the timer has not yet expired. ipvs: Update status code of misc checker if changes while in fault state. The exit code of a misc checker can be read via SNMP. The misc check code was not updating the last exit code if the checker was not dynamic, the checker was already down (i.e. returned a non 0 exit code), and the exit code changed from the previous exit code. This meant that the exit code reported via SNMP was not the latest exit code, but the exit code that caused the status of the checker to change. This commit now updates the last exit code, even if the checker is already down. vrrp: Ensure VRRPv3 advert interval strictly <= 40.95 seconds. If an advert interval of 40.958 seconds was configured, it was being round up to 40.96 after the check that the advert interval was less than 40.96. The consequence of this was that adverts were being sent at 40.96 second intervals, but worse, the advert interval in the VRRP packet was set to 0. This commit now ensures that after the rounding the advert interval is <= 40.95 seconds. vrrp: fix track process reinitialize fork delay timer. Github user Bbulatov identified that terminate_delay was being used when fork_delay should have been used. While investigating, it was also found, albeit in a debug message that fork_delay was used where terminate_delay should have been used. Further, the process state was being updated immediately even if the fork_delay was being invoked. vrrp: fix memory leak if error in vrrp_ipsets configuration. vrrp: stop memory leak when error in configuring vrrp_iptables. bfd: make alloc_bfd() return NULL rather than false on error. alloc_bfd() returns a bfd_t *, but in the case of errors it was returning false, which clearly should have been NULL. This issues was identified by compiling with -std=c23. vrrp: fix corruption of master-child_pid red black tree. Child process thread_t structures use two red-black trees, one for the timeout, and the other for pids. It is important to ensure that threads are removed from the child_pid RB tree at the correct time. This was not happening when reloads were occurring and there was a THREAD_CHILD_TIMEOUT thread on the ready list. A few other instances of the thread not being removed from the child_pid RB tree correctly, which are also resolved by this commit. all: Fix parsing of xNN in quoted strings. Following x keepalived processed all following hex digits, but only returned one byte. For example x20file would result in a byte 0x0f followed by the string “ile”. This commit limits the number of hex digits consumed to 2. all: fix parsing of escaped characters in quoted strings. core: fix error report in json version parser. Signed-off-by: Adolf Belka --- lfs/keepalived | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/lfs/keepalived b/lfs/keepalived index ed1ad87d3..36c918583 100644 --- a/lfs/keepalived +++ b/lfs/keepalived @@ -1,7 +1,7 @@ ############################################################################### # # # IPFire.org - A linux based firewall # -# Copyright (C) 2007-2024 IPFire Team # +# Copyright (C) 2007-2026 IPFire Team # # # # 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 # @@ -26,7 +26,7 @@ include Config SUMMARY = A keepalive facility for Linux -VER = 2.3.1 +VER = 2.3.4 THISAPP = keepalived-$(VER) DL_FILE = $(THISAPP).tar.gz @@ -34,7 +34,7 @@ DL_FROM = $(URL_IPFIRE) DIR_APP = $(DIR_SRC)/$(THISAPP) TARGET = $(DIR_INFO)/$(THISAPP) PROG = keepalived -PAK_VER = 16 +PAK_VER = 17 DEPS = @@ -48,7 +48,7 @@ objects = $(DL_FILE) $(DL_FILE) = $(DL_FROM)/$(DL_FILE) -$(DL_FILE)_BLAKE2 = f4e0de6147f0a33d2d87f12da4dda014fa820ba3502fd8f92eaaa4a8e56d26ea4bb1902df5c7ee13e4b81336e3b237d1ddc8247e443ea351646259c9e1ddb90b +$(DL_FILE)_BLAKE2 = 6468cbdd8a5ead443bdb0247ad2a2a7043e20f11c72afa870cded9b52ec23ff98493394aa3fee6e4ece953c85585edcfe76582a13532904cbc29b23f65de7850 install : $(TARGET)