Sign Up
Log In
Log In
or
Sign Up
Places
All Projects
Status Monitor
Collapse sidebar
openSUSE
keepalived.11715
keepalived.changes
Overview
Repositories
Revisions
Requests
Users
Attributes
Meta
File keepalived.changes of Package keepalived.11715
------------------------------------------------------------------- Thu Nov 7 02:20:31 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - new BR pkgconfig(libnftnl) to fix nftables support ------------------------------------------------------------------- Thu Nov 7 02:03:15 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.19 Fix minor IPVS features support. Extend BFD to support more than one BFD instance with a neighnour. Extend nftable support. Script timeout extension. Properly filter IGMP/MLD packets on VMAC interface. Refer to ChangeLog for more infos. https://keepalived.org/changelog.html - changes from 2.0.18 Add support to IPVS new GUE tunnel type. New feature 'weight reverse' available in all trackers. Resolve all outstanding coverity issues. Some fixes and performance extensions. Refer to ChangeLog for more infos. https://keepalived.org/changelog.html ------------------------------------------------------------------- Tue Jul 23 17:30:04 UTC 2019 - chris@computersalat.de - Update to 2.0.17 (2019-06-25) * https://www.keepalived.org/changelog.html - remove obsolete patch * systemd-after-snmp.patch - rebase patch * linux-4.15.patch ------------------------------------------------------------------- Tue Apr 16 19:04:13 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - added systemd-after-snmp.patch: If you want to use the snmp support the masterx socket needs be available otherwise the snmp support is broken strictly speaking we would need to use BindsTo= here but that would require that add a Requires for net-snmp to the keepalived package. to be discussed. ------------------------------------------------------------------- Tue Apr 16 19:01:38 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.15 - Fix uninitialised variable. - Fix rpmbuild on CentOS7, and rely on auto-requires. - Add option to flush lvs on shutdown. Currently all known virtual servers and their real servers are removed one at a time at shutdown. With large configurations on a busy system, this can take some time. Add an option just like the existing 'lvs_flush' which operates on shutdown. Typical environments with a single keepalived instance can take advantage of this option to achieve a faster shutdown or restart cycle. - Make alpha mode checkers on new real servers start down on reload. Patch #1180 identified that new real servers with alpha mode checkers were being added online immediately, and if the checker then failed were being removed. This commit makes real servers that didn't exist before the reload start in down state if they have alpha mode checkers. - Remove duplicate config dump entry. - Make new real servers at reload start down if have alpha mode checkers. - Close checker and smtp_alert sockets on reload. Issue #1177 identified that sockets were being left open (lost) after a reload. It transpired that these were sockets opened by TCP_CHECK, HTTP_GET, SSL_GET, DNS_CHECK and SMTP_CHECK checkers, and by smtp_alerts in the process of being sent. This commit adds an extra parameter to thread_add_read() and thread_add_write() to allow indicating that the scheduler should close the socket when destroying threads. - Send vrrp group backup notifies at startup. - Make inhibit_on_failure be inherited by real server from virtual server. - Allow real and sorry servers to be configured with port 0 This is to maintain backwards compatibility with keepalived prior to commit d87f07c - "Ensure always check return from inet_stosockaddr when parsing config". The proper way to configure this is to omit the port, which requires the next commit. - Don't setup IPVS config with real and virtual servers ports different. If the real server is using DR or TUN, the port of the real server must be the same as the port of the virtual server. This commit uses the virtual server port for the real server when configuring IPVS. - Log warnings if real server and virtual server ports don't match This commit adds logging warnings if virtual and real server ports, when using TUN or DR, don't match. It also sets the real server ports to be the same as the virtual server ports. Although listing the IPVS configuration with ipvsadm will look different, the kernel ignored the port of a real server when using DR or TUN, so the behaviour isn't changed, but when looking at the configuration it now shows what is actually happening. - Fix warning when protocol specified for virtual server with fwmark. - Add log message that nb_get_retry is deprecated. - Fix whitespace in configure.ac. - Fix configure error when systemd not installed configure was trying to execute pkg-config --variable=systemdsystemunitdir systemd even if systemd was not available. This commit makes configure only execute the above if it has determined that systemd is the correct init package to use. - Correct references to RFC6527 (VRRPv3 SNMP RFC). - nsure checker->has_run is always set once a checker has run. - Fix some indentation in configure.ac. - Update fopen_safe() to open temporary file in destination directory rename() in fopen_safe() was failing if the file being created was not on the same filesystem as /tmp. - Add ${_RANDOM} configuration keyword. It might seem strange to introduce random elements to configuration files, but it can be useful for testing. - Fix using ~SEQ() in multiline configuration definitions. - Make blank lines terminate a multiline definition. - Minor updates for lvs_flush_on_stop. - Add option to skip deleting real servers on shutdown or reload If a virtual server is removed, the kernel will remove its real servers, so keepalived doesn't explicitly need to do so. The lvs_flush_onstop option removes all LVS configuration, whereas this new option will only remove the virtual servers managed by keepalived. - Correct error message re checker_log_all_failures. - Fix syntax error in configure.ac. - Fix track_process initialisation for processes with PIDs starting 9. - Remove debugging log message. - Remove inappropriate function const attributes They were causing iptables/ipsets not to be initialised. - Stop warning: function might be candidate for attribute ‘const’ Depending on what configure options are selected, gcc can output the above warning for initialise_debug_options(). This commit ensures that the warning is not produced. - Enable strict-config-checks option in keepalived.spec RPM file. - vrrp: relax attribute 'const' warning at iptables helpers. - Propagate libm to KA_LIBS. - Fix building on Alpine Linux. Alpine (musl) doesn't have a definition of __GNU_PREREQ, so create a dummy definition. ------------------------------------------------------------------- Wed Apr 3 13:52:51 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - add buildrequires for file-devel - used in the checker to verify scripts ------------------------------------------------------------------- Wed Apr 3 13:46:22 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.14 - Add compiler warning -Wfloat-conversion and fix new warnings. It was discovered that passing 0.000001 as a parameter specified as uint32_t to a function did not generate any warning of type mismatch, or loss of precision. This commit adds -Wfloat-conversion and fixes 3 instances of new warnings that were generated. - For non systemd enviroment, it occurs syntax error 'fi'. To avoid syntax error, modify keepalived.spec.in. - When uninstall keepalived with init upstart, stop keepalived process. - Fix type re LOG_INGO should be LOG_INFO - 6git stash --cached. The code was actualy in a #ifdef INCLUDE_UNUSED_CODE block, and so isn't currently compiled. - Register missing thread function for thread debugging. - Fix reutrn value of notify_script_compare misusing issue. - Fix typo in keepalived.conf man page re BFD min_rx. - Fix segfault when bfd process reloads config. Issue #1145 reported the bdf process was segfaulting when reloading. The bfd process was freeing and allocating a new thread_master_t when reloading, which doesn't work. This commit changes the bfd process to clean and reinitialise the thread_master_t. - Fix segfault in handle_proc_ev(). On Linux 3.10 the ack bit can be set in a connector message, and the CPU number is set to UINT32_MAX. This commit skips acks, and also checks that CPU number is within range of the number of CPUs on the system. - Fix OpenSSL init failure with OpenSSL v1.1.1. OpenSSL v1.1.1, but not v1.1.0h or v1.1.1b failed in SSL_CTX_new() if OPENSSL_init_crypto(OPENSSL_INIT_NO_LOAD_CONFIG) had previously been called. This commit doesn't call OPENSSL_init_crypto() if doing so causes SSL_CTX_new() to fail. - Remove all references to libnfnetlink. Commit 2899da6 (Stop using linbl for mcast group membership and setting rx buf sizes) stopped using libnfnetlink, but INSTALL and keepalived.spec.in were not updated accordingly. - Fix genhash re OPENSSL_init_crypto bug and improve configure.ac. Commit fe6d6ac (Fix OpenSSL init failure with OpenSSL v1.1.1) didn't update the identical code in genhash/ssl.c. Also, an improvement for the test in configure.ac was suggested. - Fix log output when real server removed. FMT_VS() and FMT_RS() both call inet_sockaddrtotrio which uses a static buffer to return the formatted string, but since FMT_VS(), wheich simply calls format_vs() copies the returned string to its own static buffer, if FMT_VS() was called before FMT_RS() then the returned strings from both could be used. The problem occurs when both FMT_VS() and FMT_RS() are used as parameters to log_message() (or printf etc). It appeared to work fine on x86_64, but was writing the same IP address for both the real server and virtual server on ARM architectures. This is due to the compiler evaluating parameters to the log_message() function call in a different order on the different architectures. This commit adds inet_sockaddrtotrio_r() which allows the output to be in a buffer specified by the caller, and so FMT_VS() and FMT_RS() can now be called in either order without one overwriting a buffer used by the other. - Streamline some string formatting with FMT_RS() and FMR_VS(). Following commit 9fe353d (Fix log output when real server removed) some code can be streamlined now that the order of calling FMT_VS() and FMT_RS() does not matter. - Replace FMT_HTTP_RS(), FMT_TCP_RS() and FMT_DNS_RS() with FMT_CHK(). They were all simply defined to be FMT_CHK() so just replace them with that. This made it much simpler to find all used of FMT_CHK(). - Fix building with gcc 4.4.7 (Centos 6.5). gcc v4.4.7 doesn't support -Wfloat-conversion, so check for it at configure time. - Add dumping checker config/status when receive SIGUSR1. - Don't put alpha mode checkers into failed state at reload If a new checker is added at a reload, unless the real server aleady has failed checkers, then ignore the alpha mode of the checker. This means that the real server, if up, won't be taken down and then brought back up again almost straight away. If the real server already has failed checkers, then setting an alpha mode checker down initially won't take down the real server, so we can allow the alpha mode setting to apply. - Handle alpha mode checkers initial failure at startup better. - Fix compile failure discovered by Travis-CI. - Fix calling syslog when not using signalfd(). Pull request #1149 identified that syslog is AS-Unsafe (see signal-safety man page), and that therefore signals should be blocked when calling it. This commit blocks signals when calling syslog()/vsyslog() when signalfd() is not being used. - Rationalise function attributes. - Fix enable-optimise configure option. - Use AS_HELP_STRING for all options in configure.ac. - Streamline genhash -h option. - Make genhash -v version match keepalived. - Fix config check of virtual server quorum against weights of real servers. - Fix some configure tested checks for OPENSSL_init_crypto. - Add infrastructure for adding additional compiler warnings. - Add standard and extra compiler warnings. - Add and resolve missing-declarations and missing-prototypes warnings Approximately 16 additional functions are now declared static. - Add and resolve old-style-definitions warnings - Add and resolve redundant-decls warnings - Add and resolve jump-misses-init warnings - Add and resolve shadow warnings - Add and resolve unsuffixed-float-constants warnings - Add and resolve suggest-attribute=const warnings - Add and resolve suggest-attribute=format warnings - Add and resolve suggest-attribute=malloc warnings - Add and resolve suggest-attribute=noreturn warnings - Add and resolve suggest-attribute=pure warnings - Add and resolve unused-macros warnings - Add and resolve null-dereference warnings - Add and resolve float-equal warnings - Add and resolve stack-protector warnings - Add and resolve strict-overflow=4 warnings - Add and resolve pointer-arith warnings This particularly includes adding a number of bytes to a void -. - Add and resolve cast-qual warnings - Resolve additional warnings identified on Centos 6.5/gcc 4.4.7 - Remove static from zalloc() - Fix some compiler warnings on Ubuntu Xenial, and add comments re others. - Rename LIST parameters to lst in list_head.h to avoid upper case. - Fix real server checkers moving from failed to OK on reload. - add rs judgement in migrate_checkers. - Detect connection failure in genhash and exit rather than loop. - Add another function pure attribute. - Fix sending notifies for vrrp instances at startup when in sync group Issue #1155 idenfified that notify scripts for vrrp instance transition to backup state when keepalived started up were not being sent if the vrrp instance was in a sync group. It was also the case that SNMP traps, SMTP alerts and FIFO notifies were not being sent either. This commit make keepalived send the initial notifies when the vrrp instance is in a sync group. - Fix building keepalived RPM on Fedora 26. For some reason -fPIC is needed when testing for the presence of setns(). - Add vrrp_startup_delay configuration option. Some systems that start keepalived at boot time need to delay the startup of the vrrp instances, due to network interfaces taking time to properly come up. This commit adds a global configuration option vrrp_startup_delay that delays the vrrp instances starting up, for the specified number of seconds. - Handle checkers properly when reload immediately after startup. - Streamline some of the SMTP checker code. - Create separate checker for each host in SMTP_CHECK block Having multiple host entries in an SMTP_CHECK block is deprecated. This commit streamlines the SMTP_CHECK code by creating a separate SMTP checker for each host declared in the SMTP_CHECK block, so that apart from parsing the configuration, the code no longer handles multiple hosts per checker. The support for parsing configuration with multiple hosts is only enabled if WITH_HOST_ENTRIES is defined in check_smtp.c. It is currently enabled, but when support for multiple hosts in the SMTP_CHECK block is finally removed, it will simply be a matter of deleting all code in the WITH_HOST_ENTRIES conditional blocks. - Make checker fail if ENETUNREACH returned by connect(). The connect() call can return some immediate errors such as ENETUNREACH. These were not being treated as a failure of the checker, since the code used to assume that any non success return by connect() meant that the connection was in progress. keepalived will now treat ENETUNREACH, EHOSTUNREACH, ECONNREFUSED, EHOSTDOWN, ENETDOWN, ECONNRESET, ECONNABORTED, ETIMEDOUT, when returned by connect(), as meaning that the checker has failed. - Don't set SO_LINGER with a timeout of 0 SO_LINGER with a timeout of 0 causes a TCP connection to be reset rather than cleanly closed. Instead of specifying a timeout of 0, use 5 seconds, so that there is an orderly shutdown of the TCP connection, but the close socket doesn't remain in TIMED_WAIT state for more than a short time. - nftables: fix build with kernel lower than 4.1. - Remove dead code and cosmectics. Remove code marked as UNUSED where things simply go nowhere even if define is set. We keep for the moment UNUSED code related to debug helpers used during coding process. ------------------------------------------------------------------- Wed Mar 20 23:31:55 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.13 - Add BFD build option to keepalived.spec rpm file Issue #1114 identified that the keepalived.spec file was not being generated to build BFD support even if keepalived had been configured to support it. - Copy tarball to rpmbuild/SOURCES when building in place It seems that even when building in place, rpmbuild expects the tarball to be in the rpmbuild/SOURCES directory. - Fix configure check for __always_inline - Handle interface MAC addresses changing When an interface is added to a bond interface, if it is the first interface added, the MAC address of the bond interface is changed to the MAC address of the added interface. When subsequent interfaces are added, their MAC addresses are changed to that of the bond interface. Issue #1112 identified that if a bond interface is deleted and recreated, the gratuitous ARPs were sent with the wrong source MAC address. This commit now updates interface MAC addresses from the netlink RTM_NEWLINK messages, so that the correct MAC address is always used. - Minor tidying up of opening gratuitous ARP socket. - Streamline setting SOCK_NONBLOCK on vrrp sockets. - Use netlink reported hardware address length for unsolicited NAs ETH_ALEN is correct for Ethernet type interaces, but is not right for Infiniband interfaces. - Minor tidying up of opening gratuitous NA socket. - Make gratuitous ARP/NA sockets non blocking keepalived shouldn't block when sending gratutious ARP/NA messages. It is better to lose the messages than for keepalived to block, so set the sockets non blocking. - Use netlink provided broadcast address for gratuitous ARP If an interface has a non-standard broadcast address, we should honour it. - Fix building on pre 3.10 kernels re track_process Issue #1119 reported that keepalived wouldn't build on CentOS 6. Various PROC_EVENT_- declarations were assumed to exist, some of which were not introduced until Linux v3.10. Most of them are not needed, but PROC_EVENT_COMM is used by the track_process code. This commit now checks for the existence of the PROC_EVENT_- declarations, but since keepalived uses PROC_EVENT_COMM, track_process is not supported prior to Linux v3.2. - Make track_process work prior to Linux 3.2, but with limitations Prior to Linux 3.2 the PROC_EVENT_COMM event did not exist, which means that keepalived is unable to detect changes to process name (/proc/PID/comm) prior to Linux 3.2. most processes do not change their process name, and so using track_process prior to Linux 3.2 is safe so long as the monitored processes are known not to change their process name. - Stop configure failing when nftables is not supported. - Streamline socket use with linkbeat. Previously the socket used for ioctls was opened and closed twice per poll if using MII or ETHTOOL polling, and once per poll if using ioctl polling. This commit opens the socket once at startup, uses that socket for all linkbeat polls, and closes it on termination. - Enable linkbeat polling to work with dynamic interfaces. - Add linkbeat_interfaces configuration block It was not possible to indicate that an interface that wasn't used as the interface of a vrrp instance, but was used either as a track interface, or for virtual/static ip addresses or routes should use linkbeat. This commit adds that capability. - Add ability to specify linkbeat type in linkbeat_interfaces block. - Add --disable-linkbeat configure option Does anyone use linkbeat anymore? This commit enables keepalived to be build without the linkbeat code. - Don't remove link local IPv6 address from VMAC that isn't keepalived's If IFLA_INET6_ADDR_GEN_MODE isn't supported and a macvlan interface already had a (non-default) link local addresss and the link local address that matched the interface's MAC address was added, keepalived was removing it as soon as it was added. This commit stop keepalived removing the address when we shouldn't. - Set configure init type correctly in keepalived.spec file. - Fix handling of VMACs with multiple reloads If a configuration is loaded that has a VRRP instance using a VMAC, then the configuration is updated to remove that VRRP instance and keepalived reloads its configuration, then the configuration is updated again to reinstate the VRRP instance and the configuration is again reloaded, keepalived thought the VMAC interface still existed, whereas it was deleted following the first reload. This commit ensures that keepalived properly detects whether an interface exists following a reload. - Remember more than one interface local address per interface Keepalived needs a local address for each interface it sends adverts on. If the address keepalived is using is deleted and another address is configured on the interface, then keepalived should start using that address. To do this, a list of configured address on each interfaces needs to be maintained. - Don't consider VIPs as local addresses when restart after crash Keepalived maintains a list of addresses per interface that can be used as source adddresses for adverts. To build the list, keepalived reads the addresses configured on interfaces when it starts. However, if keepalived crashed it will have left VIPs configured on interfaces, and we don't want to use them as advert source addresses. This commit makes keepalived compare the addresses on interfaces to VIPs, and ignores any addresses that are VIPs. - Fix removing left over VIPs at startup. - Use read_timer() when parsing config where appropriate. - Allow fractional warmup, delay_loop and delay_before_retry for checkers To shorten the real server monitoring interval, make it possible to specify decimal value for following items: warmup delay_loop delay_before_retry - Update connect_timeout configuration options Based on the patch submitted by tamu.0.0.tamu@gmail.com this patch allows setting the connect_timeout to a resolution of micro-seconds. The patch also adds the ability to set a default value at the virtual server and real server levels. - Fix unused variable warning when building only with RFC compliant SNMP. - It enable to set zero value as mintime for delay_loop and connect_timeout. - Add option not to check for EINTR if using signalfd() If keepalived is using signalfd(), there are no asynchronous signal handlers, and therefore EINTR cannot be returned. Currently the check for EINTR is enabled by default, and configure option --disable-eintr-debug disables the check, while --enable-eintr-debug enables writing log entries if EINTR is returned. Once sufficient testing has been performed, the default will be changed not to test for EINTR if signalfd() is supported. - Make checking for EAGAIN/EWOULDBLOCK consistent The code in some places checked errno for EAGAIN and EWOULDBLOCK and in other places only checked EAGAIN. On Linux EAGAIN == EWOULDBLOCK, so the check is not necessary, but EAGAIN is not guaranteed to be the same value as EWOULDBLOCK, so define check_EAGAIN that only checks EAGAIN if they are the same value, but checks both if they are different. - Ensure default connection timeout for smtp checker hosts set. - Set default connection timeout if no smtp check host specified. - Fix min timer value, zero to 0.000001Sec. - Add fixing min time for vs_co_timeout_handler() and rs_co_timeout_handler(). - Fix parameter of read_timer(), it treat Mintime and Maxtime as microseconds. - vrrp: vrrp_dispatcher_read() performance extension We took time with Quentin to simulate and rework this code. We introduced 2 imbricated while loop: (1) First one is catching recvfrom EINTR (this code trig only on kernel older than 2.6.22 where signalfd was firstly introduced). Newer kernel will immediately break the loop (hey guys: if you are running older than 2.6.22 it is worth considering upgrading). (2) Second loop will continue reading from socket until same VRID advert has been received during the same cycle. After simulating, it appears that during contention with a lot of VRRP instances (around 1500), this design is needed to relax socket recvq from growing. This can be viewed as a Poll-Mode activation during contention and fallback to regular I/O MUX during normal operations. This loop breaks immediately and re-submit opration to I/O MUX when there is no more to be read. - Fix conversion from long for double in read_timer(). - Remove variable timer of unsigned long cast in read_timer(). When Double type variable timer is cast to long type, it's scale falls. - changes from2.0.12 - Documentation related. Remove keepalived.conf.SYNOPSIS content to make a pointer to manpage. Update README manifest to reflect actual Keepalived goal and features. - Improve error message if process events connector not enabled in kernel. - Add option to disable track-process functionality Issue #1099 reported that their kernel did not support the proc events connector, and it would therefore be helpful to have an option to build keepalived without the track-process functionality. This commit adds the --disable-track-process configure option. - Fix vrrp instances going to fault state when have virtual routes If an interface going down caused a vrrp instance to go to fault state, and the vrrp instance also had virtual routes, the state of the vrrp instance would be set to backup when the deletion of the virtual route was detected. This commit ensures that the vrrp instance stays in fault state until the interface is brought up again. - Remove Red Hat Linux 9 and RH Enterprise Linux 3 from spec file. Red Hat Linux 9 and Red Hat Enterprise Linux 3 are both based on Linux 2.4, which is no longer supported by keepalived. The options in the spec file for Reh Hat Linux 9 have twice caused people to specify wrong options to configure when trying to build keepalived, so the options are removed to i) avoid confusion and ii) they are not longer relevant. - Add global option vrrp_min_garp. By default keepalived sends 5 gratuitous ARP/NA messages after transitioning to master, and 5 more 5 seconds later. This isn't necessary with modern switches, and so if the vrrp_min_garp option is set, only one gratuitious ARP/NA message is sent after transition to master, and no repeat messages are sent 4 seconds later. - Standardise definition of _INCLUDE_UNUSED_CODE_ - Remove out of date comment re VRRP over IPv6. - Correct typo in keepalived.conf.5. - Directly use structure sizes for packet header lengths. - vrrp_state_fault_rx() is not used. Wrap the function in conditional compilation so it is not compiled - Convert so list loops to use LIST_FOREACH. - Don't recalculate vrrp packet header address. vrrp_get_header() calculates the address of the vrrp header in a received packet, but it was being recalculated in vrrp_in_chk(). This commit passes the already calculated address to vrrp_in_chk(). - Ensure a received packet has an AH header if and only if AH auth. Ensure that a received packet has an AH header if we expect AH authentication, and doesn't have an AH header if we don't expect AH authentication. - Ensure all protocol headers received before return pointer to vrrp header vrrp_get_header() returns a pointer to the vrrp header, but it now returns NULL if insufficient data has been received to include all the (IP, possibly AH, and VRRP) headers (this does not include the VIPs in the VRRP packet). This means that when a pointer to the VRRP header is returned, all fields in all protocol headers can safely be accessed. - Add check of received IPv6 hop count in multicast adverts The VRRP RFC requires that IPv6 hop count MUST be checked to be 255, just as the TTL for IPv6 must be 255. Previously that wasn't being checked, since IPv6 raw sockets don't provide access to the IPv6 header. Using recvmsg() rather than recvfrom(), and setting socket option IPV6_RECVHOPLIMIT allows keepalived to receive the hop count as ancillary data, and that can now be checked. - Improve reading from vrrp receive sockets. Previously no check was made of the return value from recvfrom()/ recvmsg(). This meant than an error could occur (e.g. EINTR), or no data might be returned, and keepalived would still attempt to process the receive buffer as though data had been received. - Enhance and streamline checking of validity of received VRRP packet This includes checking that a packet is multicast, unless unicast is expected in which case it is checked for unicast, ensuring that if AH authentication is used, the next header protocol is VRRP. The sequence of some checks is revised to ensure that the fields being checked are valid to be accessed prior to accessing them, e.g. check that the packet is VRRP version 2 before checking the authentication. - Stop clearing receive buffer before receiving VRRP packets. This is no longer necessary now that the appropriate checks are made of the return status of recvmsg(), and also that the checks of received packet length and packet headers now do all necessary checks. - Add compile time checks for IPV6_RECVHOPLIMIT/IPV6_RECVPKTINFO support. - Update keepalived.spec.in build-requires. The kernel package required for building keepalived is kernel-headers not kernel-devel. Also, it is superfluous to have package kernel in the build-requires! - Add missing file (build.setup) to tarball. - Fix calculating print format to rlim_t in configure.ac. - Fix compiler warnings on 32 bit systems re HASH_UPDATE. Removing all the casts stopped the warnings. - Use PRI_rlim_t when printing rlim_t types. - Use %zd/%zu for ssize_t/size_t to avoid warnings on 32 bit systems. - Fix some space/tab formatting. - Stop declaring some timer definitions unsigned to stop compiler warnings. TIMER_HZ, TIMER_CENTI_HZ, NSEC_PER_SEC were causing some compiler warnings on some systems due to being defined with a 'U' unsigned suffix. Removing the unsigned specifier stopped the compiler warnings. - Fix compiler warning due to incorrect format specifier. An int64_t should use % PRIi64 and not %ld - Stop an uninitialized variable compiler warning. - Fix MEM_CHECK debugging on processors without unaligned memory access. - Don't attempt to use unopened socket for getting ipset version. - Tidy up an error message. - vrrp: make vrrp_dispatcher_read() async while catching error. During investigations we decided to update previous patch to resubmit into I/O MUX on read error. It will make read procedure I/O MUX freindly by removing potential sync operation potentially leading to a global I/O MUX desync. We aggreed, the situation is really and very exceptionnal but could happen. - vrrp: vrrp_arp_thread split. Split the function for maintainability purpose. ------------------------------------------------------------------- Sat Jan 19 02:22:09 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - fix build on 42.3/sle12 by disabling http regexp check support - add nftables to the BR - cleanup BR support for sle11, moved almost all BR to pkgconfig style - disable dbus instance creation support as it is marked as dangerous ------------------------------------------------------------------- Fri Jan 18 15:39:47 UTC 2019 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.11 - Fix segfault while shutting down when SNMP activity occurs. Issue #1061 identified that keepalived could segfault when it shut down. It appears that this was caused by data being received on the file descriptors that the snmp agent requests keepalived to monitor with epoll(). Since the read threads weren't being processed during a shutdown, the first time an snmp fd was ready, keepalived discarded the read thread. The second time that fd became ready there was no thread to handle the fd, and, since the assert() statement was not compiled in, non existant data was queued to the thread ready queue. This commit changes the assert() calls to continue, so that non existant data is no longer queued to the thread ready queue. - While shutting down, continue to handle snmp agent fds. Since we don't shutdown the snmp connection until the very end of the shutdown process (we need to be able to send snmp traps), we should continue to handle the snmp fds on behalf of the snmp agent while shutting down. - Ensure snmp agent is in correct state when initialising/closing Make sure the snmp agent is not already initialised before initialising it, and make sure it has been initialised before closing it. - Disable asserts in bfd code by default and add --enable-asserts Asserts were enabled by default in the bfd code, which shouldn't be the case. Add --enable-asserts configure option so that the asserts tests can be enabled while debugging. - Remove debugging log message accidently left in. - Update receive buffers when interface is created. The receive buffer size used by keepalived is based on the largest MTU of any interface that keepalived uses. If dynamic interfaces are being used and an interface is created after keepalived has started, the MTU of the new interface may be larger than the previous largest, so the receive buffer may need to be increased in size. Further, if vrrp_rx_bufs_policy is MTU, then the kernel receive buffers on the receive socket may need to be increased. - Handle MTU sizes being changed. Issue #1068 identified that the MTU size wasn't being updated in keepalived if it changed. This commit now updates the MTU size and adjusts receive buffer sizes accordingly. - Fix syntax error in configure.ac. - Fix double free when global data smtp_helo_name copied from local_name Issue #1071 identified a double free fault. It occurred when smtp_helo_name was not set, in which case it was set to point to the same malloc'd memory as local_name. At termination keepalived freed both local_name and smtp_helo_name. If keepalived needs to use local_name for smtp_helo_name it now malloc's additional memory to copy the string into. - Rename TIMER_MAX to TIMER_MAXIMUM. ulibC defines TIMER_MAX, so to avoid naming conflict rename it. This issue was reported by Paul Gildea <gildeap@tcd.ie> who also provided the patch. - Fix segfault when smtp alerts configured. - First working version of nftables. - Restructed code around how iptables/nftables are called This commit also allows building keepalived without iptables support, thereby allowing only nftables support. Adding any other mechanism to handle no_accept mode, i.e. blocking receiving and sending to/from VIPs should be added to vrrp_firewall.c, in a similar way to how nftables/iptables are used. - Update doc files re nftables. - Make nftables handle dont_track_primary appropriately. - Fix config reload with nftables. - Set base chain priorities from configuration. - Use iptables by default if neither iptables or nftables configured. But if the build of keepalived does not include iptables, then use nftables default. - Stop dumping keywords - left turned on after debugging. - Make umask configuration apply to created file. - Add libmnl and libnftnl to travis file. - Fix compilation failure when NFTNL_EXPR_LOOKUP_FLAGS not defined. - Fix compilation failure when build with nftables but without iptables. - Fix order of include files in configure COLLISION test. Since Linux 4.4.11 (commit 1575c09) including linux/if.h after net/if.h works, whereas until glibc fix their headers including net/if.h after linux/if.h causes compiler redefinition errors. Unfortunately the test for the collision was done the wrong way round, as identified in issue #1079. The patch included in the issue report corrects the order of inclusion of the header files. What we should do is ensure that glibc header files are included before Linux header files, so that at least if kernel headers from 4.4.11 onwards are used, the conflict will not occur. - Set CLOEXEC on netlink sockets. - Correct error message for invalid route metric. - Add track_process for vrrp to monitor if another process is running. Configurations frequently include a track_script to check that a process is running, often haproxy or nginx. Using any of pgrep, pkill, killall, pidof, etc, has an overhead of reading all /proc/[1-9]*/status and/or /proc/[1-9]*/cmdline files. In particular reading the cmdline files has a significant overhead on a system that is swapping, since the cmdline files provide access to part of the address space of each process, which may need to be fetched from the swap space. This commit reads the /proc/[1-9]*/stat and/or the /proc/[1-9]*/cmdline files only when keepalived starts, and after that uses the process events connector to track process creation and termination. keepalived will ignore zombie processes, whereas pgrep etc include them. A minimum number of instances of a process can be specified, and also a delay so that if a process is restarted, it won't cause monitoring vrrp instances to immediately transition to fault state but to wait the configured time and it the monitored process starts again it won't transition to fault state. There are potential difficulties with the process event connector if a large number of process events occur very rapidly, since there can be a receive buffer overrun on the netlink socket. This code will detect that happening, increase the receive buffer size, and reread the processes from /proc. - Add missing #include to track_process.c. - Fix number of elements of fd_set read for snmp select info. - Remove thread_event_t when EPOLL_CTL_DEL fails. If snmpd closes a file descriptor, when keepalived attempts to unregister the fd from epoll an error is returned. However, we still need to remove the thread_event_t from the io_events rbtree. - Fix connection to snmpd after it has to reconnect. Issue #1080 identified that keepalived wasn't handling a connection failure and reconnect to snmpd properly. The problem was created when the change from select() to epoll() was made. This commit makes keepalived unregister and reregister the snmp file descriptors after snmpd reconnects. - Fix retry count for SMTP_CHECK checker. The checker was doing one too few retries. - Make healthchecker failure reporting consistent Some healthcheckers were reporting all failures, and others only when the retries expired. This commit by default makes the checkers only report failure when the retries expire, unless the global keyword checker_log_all_failures or log_all_failures on the specific checker is configured. - After reload, reinitialise current track processes state. - Remove unused variable in track_process.c. - Add configure checks re --with-kernel-dir. - Convert remaining select() to epoll_wait(). keepalived was using select() for handling the termination of child processes, but the main scheduling loop now uses epoll_wait(), so convert the select() to epoll_wait() from consistency. - Stop keepalived leaving zombie child processes. keepalived wasn't reaping the termination of its child processes, so this commit adds waitpid() calls once it knows the processes have terminated. - Fix make distclean and make distcheck. - Also skip route not configured with down interface. Otherwise, if keepalived has virtual_routes configured, we create a virtual interface and bring it up and down, current code will bring VRRP state to FAULT and never return. - Stop vrrp process entering infinite loop when track script times out Issue #1093 identified that the vrrp process was entering an infinite loop after a track script timed out. This was due to a child process thread having an RB tree for PIDs as well as for the timeout, and if a child process timed out, the thread wasn't being removed from the PID RB tree. This commit now ensures it is removed. - Fix the abbreviation of Shortest Expected Delay. - Don't free unallocated memory if not tracking processes. - vrrp: Rewrote JSON code Remove dependency to json-c extralib by using a simple streaming JSON writter. Refactored code to make it simple to maintain. - vrrp: Fix JSON handling for v{route;rule}. - autoconf: fix nftables selection We need to inhibit nftable compilation if compiling system has kernel header file nf_tables.h but not libnftnl nor libmnl. ------------------------------------------------------------------- Wed Nov 28 12:27:13 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.10 - Fix compiling on Alpine Linux. - Stop printf compiler warning on Alpine Linux due to rlim_t. - manpage cosmetic. - Fix removing snmpd read threads when snmpd becomes unavailable. - Update to support libipset version 7. - Use ipset_printf for ipset messages so can go to log. - When opening files for write, ensure files can only be read by root. Issue #1048 referred to CVE-2018-19046 regarding files used for debugging purposes could potentially be read by non root users. This commit ensures that such log files cannot be opened by non root users. - Disable fopen_safe() append mode by default If a non privileged user creates /tmp/keepalived.log and has it open for read (e.g. tail -f), then even though keepalived will change the owner to root and remove all read/write permissions from non owners, the application which already has the file open will be able to read the added log entries. Accordingly, opening a file in append mode is disabled by default, and only enabled if --enable-smtp-alert-debug or --enable-log-file (which are debugging options and unset by default) are enabled. This should further alleviate security concerns related to CVE-2018-19046. - vrrp: add support to constant time memcmp. Just an update to use best practise security design pattern. While comparing password or hmac you need to ensure comparison function is time constant in order to figth against any timing attacks. We turn off potential compiler optimizations for this particular function to avoid any short circuit. - Make sure a non privileged user cannot read keepalived file output Ensure that when a file such as /tmp/keepalived.data is written, no non privileged can have a previous version of that file already open, thereby allowing them to read the data. This should fully resolve CVE-2018-19046. - drop b7a98f9265ffb5927c4d54c9a30726c76e65bb52.patch: included in update ------------------------------------------------------------------- Sat Nov 10 21:01:14 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - added b7a98f9265ffb5927c4d54c9a30726c76e65bb52.patch to fix building with libipset >= 7 ------------------------------------------------------------------- Fri Nov 9 16:07:40 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.9 - Fix updating a timer thread's timeout. Issue #1042 identified that the BFD process could segfault. This was tracked down to a timer thread which had already expired having its timeout updated by timer_thread_update_timeout(). The sands timer should only be updated if the thread is on a waiting queue, and not if it has already timed out or it is unused. - Don't requeue read thread if it is not waiting. This update matches commit 09a2a37 - Fix updating a timer thread's timeout should. - Allow BFD instance to recover after send error. If sendto failed in bfd_send_packet(), the bfd instance was put into admin down state, but there was no means for the bfd instance to transition out of admin down state. This commit makes keepalived log the first instance of a sequence of failures to send a bfd packet, but does not bring the bfd instance down in case the error is a transient error. If the error is longer lasting, the remote system will timeout, transition to down state, and send a message saying it is down. Once the bfd instance can start sending again the bfd instance can now transition again to up state. - Make DGB definition use log_message() rather than syslog(). - Fix building with --enable-debug configure option. - Start list of required kernel features in INSTALL file. Issue #1024 asked what kernel features are needed to support keepalived. The simple answer was that it isn't recorded anywhere, so this is a start of making a list of the features required. - Make list_remove() call list free function and add list_transfer(). If an element is being removed from a list, the free function should be called. list_transfer() allows a list element to be moved from one list to another without freeing and reallocating the list element control information. - Add mem_check diagnostics re calling functions of list functions. When using mem_check, mallocs and frees were recorded against the list functions, and the originating functions weren't identified. This patch adds recording of the functions calling the list functions so that the originating function is identified. - Simplify the processing of comments in configuration files. This commit moves the handling (and removal) of comments to a single function (called from read_line()) which simplifies the processing of config files. - Add ~SEQ(start, step, end) config functionality Where a configuration has repeated blocks of configuration where the only thing that changes is a numeric value (e.g. for VRIDs from 1 to 255) this allows the block to be defined once, and a single line using ~SEQ can then generate all the blocks. - Use REALLOC when building a multiline definition. The code used to use MALLOC, strcpy() and FREE, but REALLOC can do all this for us. - Improve mem-check diagnostics. When using an allocation list of over 50,000 entries, it was quite slow searching thtough all the entries to find the matching memory allocation, and to find free entries. This commit changes to using malloc() to create entries, and a red-black tree to hold the entries. It also has a separate list of free entries. This commit also adds 4 more types of memory allocation error, and improves the consistency of the entries in the log files. - Don't attempt to delete VMAC when underlying interface is deleted. If the underlying interface of one of our vmacs is deleted, and we know the vmac has been deleted, don't attempt to delete it again. - Include master state in determining if vmacs are up or down Netlink doesn't send messages for a state change of a macvlan when the master device changes state, so we have to track that for ourselves. - Turn off parser debugging. - Make test/mk_if create iptables chains. - Handle interfaces not existing when keepalived terminates. If the underlying interface of a vmac we created has been deleted, the vmac will not exist so don't attempt to delete it again. Also, don't attempt to reset the configuration of the underlying interface. - Handle the underlying interface of a macvlan interface going up/down. The kernel doesn't send netlink messages for macvlans going up or down when the underlying interface transitions (it doesn't even update their status to say they are up/down), but the interfaces don't work. We need to track the state of the underlying interfaces and propagate that to the macvlan interfaces. - Fix duplicate value in track_t enum. - Fix check for matching track types. - Treat macvtap interfaces in the same way as macvlan interfaces. - Improve handling of interfaces not existing when keepalived starts. - Fix handling interface deletion and creation of vmacs on macvlan i/fs. - When interface created, open sockets on it if used by VRRP directly If an interface is created that has vrrp instances configured on it that don't use VMACs, or use vmac_xmit_base, then the raw sockets must be opened. - Force seeing a transition to up state when an interface is created. - Fix netlink remnant data error. - Add command line and configuration option to set umask. Issue #1048 identified that files created by keepalived are created with mode 0666. This commit changes the default to 0644, and also allows the umask to be specified in the configuration or as a command line option. - Fix compile warning introduced in commit c6247a9. Commit c6247a9 - "Add command line and configuration option to set umask" introduced a compile warning, although the code would have worked OK. - When opening files for write, ensure they aren't symbolic links. Issue #1048 identified that if, for example, a non privileged user created a symbolic link from /etc/keepalvied.data to /etc/passwd, writing to /etc/keepalived.data (which could be invoked via DBus) would cause /etc/passwd to be overwritten. This commit stops keepalived writing to pathnames where the ultimate component is a symbolic link, by setting O_NOFOLLOW whenever opening a file for writing. This might break some setups, where, for example, /etc/keepalived.data was a symbolic link to /home/fred/keepalived.data. If this was the case, instead create a symbolic link from /home/fred/keepalived.data to /tmp/keepalived.data, so that the file is still accessible via /home/fred/keepalived.data. There doesn't appear to be a way around this backward incompatibility, since even checking if the pathname is a symbolic link prior to opening for writing would create a race condition. - Make netlink error messages more meaningful. - Fix compiling without support for macvlans. - fix uninitialized structure. The linkinfo and linkattr structures were not initialized, so we should not expect that unexistant attributes are set to NULL. Add the missing memset(). - fix socket allocation with dynamic interfaces. When there are several vrrp instance binding different interfaces that don't exist at startup, their ifindex is set to 0 in the sock. The function already_exist_sock() that lookup for an existing socket will always return the first sock because the ifindex is the same. Later, when an interface appears, the fd will be created for one instance, and all instances will wrongly use this fd to send the advertisments. Fix this by using the interface structure pointer instead of the ifindex as the key for sock lookup. The problem was identified by Olivier Matz who also provided a patch fixing the problem. This patch is a slight rework of Olivier's patch, better using the existing data structures that keepalived already holds. - When creating a macvlan interface, use AF_UNSPEC rather than AF_INET. - Stop using libnl for configuring interfaces. Since there is code to configure the interfaces using netlink without using libnl, there is no point in having code to do it using libnl. - Fix building on Centos 6.5. - Stop including some files not needed after libnl removal for i/fs. - Fix some compilation issues when building without vrrp support. - Stop using linbl for mcast group membership and setting rx buf sizes. Since there is code to handle multicast group membership and setting kernel netlink receive buffer sizes without using libnl, there is no point in having code to do it using libnl. This now means that the vrrp functionality no longer uses libnl. - Add some sanity checking of configure options. Certain invalid combinations of configure options could cause compile errors, e.g. --disable-vrrp --enable-vrrp-fd-debug. This commit ensures that invalid combinations aren't allowed, in order to stop the compile errors. - Fix invalid configuration combination caught by previous commit. - Use netlink to set/clear rp_filter on interfaces. - Fix configure for building without vrrp. - Actually update the .travis.yml file to fix the problem. - Fix conditional compilation re epoll-thread-dump debugging. - Update INSTALL file now no longer use libnl-route-3. - Stop cast to incompatible function type warnings from gcc 8.1. - Update snapcraft.yaml not to include libnl-route-3. - keepalived exit with non-zero exit code if config file not readable. - Allow specifying default config file at configure time. - Use keepalived define for exit code when malloc failure. - Fix configuring fixed interface type. - Add configuring keepalived default configuration file. - Fix return value in get_time_rtt() error path. - Update generation of git-commit.h. - snapcraft.yaml: Enable all sensible build options. Preserve build time version in the snap version. Expose genhash. - snapcraft.yaml: Build keepalived with Linux 3.13 headers. - snap: Add an install hook to make sure a keepalived configuration exists. - snap: Move the hooks to the correct location. - snap: Make sure /etc/keepalived exists. - Fix building with IP_MULTICAST_ALL in linux/in.h but not netinet/in.h Issue #1054 identified that configure was checking the definition of IP_MULTICAST_ALL in linux/in.h but including netinet/in.h, which also has the definition, but only from glibc 2.17. This commit creates a local definition (in lib/config.h) of IP_MULTICAST_ALL if it is defined in linux/in.h but not in netinet/in.h. The reason for this is that compiles using linux/in.h fail due to conflicting definitions. - Fix creating iptables tables in mk_if. - Update .travis.yml to use xenial. - Update .travis.yml to add --enable-regex option. - Tidy up .travis.yml file. - snap: Build multiple keepalived binaries. - Updated snapcraft builds to support multiple kernel versions. - drop patches: - 5241e4d7b177d0b6f073cfc9ed5444bf51ec89d6.patch - c6247a9ef2c7b33244ab1d3aa5d629ec49f0a067.patch - 04f2d32871bb3b11d7dc024039952f2fe2750306.patch - refreshed patch: linux-4.15.patch ------------------------------------------------------------------- Thu Nov 8 12:44:47 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.8 - Improve identifing interface as macvlan when reading interface details - Enslave a VMAC to the VRF master of the underlying interface. - Use addattr32 rather than addattr_l for if_index. - Only include VRF support if kernel headers support it. - Fix --enable-timer-debug configure option. - Fix some configure.ac enable option tests. - Include stdbool.h in process.c. - Fix diagnostic message re ignoring weight of tracked interface. - Fix track_bfds with weights. - Correct conditional compilation definition name. - Fix memory leak in HTTP_GET/SSL_GET. - Fix two memory leaks in DNS_CHECK. - Don't consider retries for BFD_CHECK. The BFD_CHECKer doesn't support retries, and the check was causing the checker not to transition to down state. - Fix memory leak with BFD_CHECK. - Restart global notify FIFO handler after reload. - modify @WITH_REGEX@ to @WITH_REGEX_TRUE@ - Fix compiling without BFD support. - Stop bfd process sending double the number of packets. If a bfd process received an initial bfd packet, it scheduled a second bfd_sender_thread thereby causing two packets to be sent in every interval. - Use timerfd for select timeouts rather than select timeout parameter This is a precursor to moving to using epoll. - Use epoll rather than select. epoll is both more efficient than select and also doesn't have a file descriptor limit of 1024, which limited the number of vrrp instances that could be managed. This commit also introduces read-black trees and the list_head list type. - Add --enable-timer-check option for logging calls for getting time Calls to update the current time from the kernel are made too frequently, and this patch logs when the calls are made, and how long since the previous call, so unnecessary calls can be removed. - Add debug option for monitoring epoll queues. This is enabled by --enable-epoll-debug and replaces --enable-timer-debug. - Use system monotonic clock to generate a monotonic clock. Rather than have our own code for creating a monotonic clock, use the kernel's monotonic clock. - Make some functions in timer.c inline. The functions had one line of code so inlining them is more efficient. - Fix requeueing read and write threads after read/write timeouts. - Fix initial allocating and final freeing of thread_master epoll_events. - When cleaning up threads, also clean up their thread_events. - Add thread_close_fd() function to release thread_event_t on close When a file descriptor that has been monitored by epoll is closed the thread_event_t structure used for managing epoll for that fd has to be release. Therefore calls to close() and replace by calls to thread_close_fd(). - Make parent process write log entry when it is reloading. - Move checking for thread timeouts to timerfd_handler There is no point in checking for thread timeouts if the timerfd isn't readable; in other words only check for thread timeouts if the timer has expired. - Make bfd reschuling timer threads more efficient. - Streamline DNS_CHECK code. - Fix buffer overrun with track file path names. - Add timestamp when writing mem_check entries to file. - Ensure thread_event_t released for ready threads at termination. - Increase open file limit if large number of VRRP instances. Each VRRP instance can use up to 2 file descriptors, and so if there are more than 500 ish VRRP instances the number of open files can exceed the default per process limit (1024 on my system). The commit allows 2 file descriptors per vrrp instance plus a few more, and if the RLIMIT_NOFILE value returned by getrlimit isn't high enough, keepalived will increase the limit. - Ensure that child processes run with standard priorities/limits. When child processes such as notify scripts, track_scripts and MISC_CHECK scripts are run, they should not inherit any elevated priorities, system limits etc from the parent keepalived process. - Change multiple spaces to tabs in scheduler.h. - Add family to sockpool listing. - Fix a multiline definition expansion issue. - Free allocated cache when closing/freeing netlink socket. When running on a system with 500+ interfaces configured and adding 1000 VMAC interfaces, the heap was growing by 340Mb due the netlink cahce not being freed after creating each VMAC interface. With this patch the heap only grow by 3.7Mb (if creating 1000 VMAC interfaces the heap grep by 905Mb now reduced to 6.1Mb). - Stop using netlink cache when adding and configuring VMAC interfaces. When running on a system with 500+ interfaces configured and adding 1000 VMAC interfaces, it was taking 2.3 seconds to add the interfaces. Without populating a netlink cache each time a VMAC interface is created it now takes 0.38 seconds to add the interfaces (if creating 1000 VMAC interfaces it was taking 6.1 seconds, now reduced to 0.89 seconds, and the heap growth is reduced from 6.1Mb to 3.9Mb). - Add function rtnk_link_get_kernel for dynamic linking. - Fix compiling without JSON support. - Add support for recording perf profiling data for vrrp process. - Add comment re usage of MAX_ALLOC_LIST. - Some streamlining of scheduler.c. - Merge --enable-epoll-debug and --enable-dump-threads functionality. - Let thread_add_unuse() set thread type, and use thread_add_unuse() more. - Use break rather than return in process_threads(). - Fix segfault when reloading with HTTP_GET and no regex configured. - Merge the next-generation scheduler. - Make all debug options need enabling at runtime. Previously if configure enabled a debug option its output was always recorded, which meant that if one didn't want the output, configure/ compile was needed. This commit adds command line options that need to be set in order to turn the debugging on. - Remove unwanted debug message. - Fix parsing --debug options. - Fix rb tree insertion with timers. - Add missing functions for thread debugging. - Add vrrp instance VMAC flags when dumping configuration. - Ensure parent thread terminates if child has permanant config error. - Ensure don't delete VMAC interface if keepalived didn't create it. and sundry fixes. - If receive lower priority advert, send GARP messages for sync group. A recent update to issue #542 identified that following recovery from a split brain situation, GARP messages weren't being sent. It transpired that, if a member of a sync group in master state received a lower priority advert and vrrp_higher_prio_send_advert is set, a further (lower priority) advert is sent, and the instance and all the members of the sync group transition to backup (the other members of the sync group don't send a further advert since they haven't received a higher priority advert). This meant that the other members of the sync group on the keepalived instance that remained master didn't receive a lower priority advert, and so didn't send further GARP messages. This commit changes keepalived's behaviour, so that if a vrrp instance is sending GARP messages due to receiving a lower priority advert and it is a member of a sync group, keepalived will also send GARP messages for any other member of the sync group that have garp_lower_prio_rep set. - Allow 0.0.0.0 and default/default6 for rule/route to/from addresses. - Check return value of SSL_CTX_new(). - Check return values of SSL_new() and BIO_new_socket(). - Only allow subnet masks with routes or virtual IP addresses. For example, if specifying a via address or preferred source address for a route, it isn't valid to specify a subnet mask. - Add inet/inet6 to specify ip route/rule family if ambiguous. - Remove superfluous parameter from parse_route(). - Add "any" and "all" as synonyms for "default". - Fix memory leak if route destination address is wrong address family. - Add ttl-propagate route option. - Fix checking return status of kill(). - Fix building with --enable-debug configure option. - Stop delay in reload when using network namespaces. If running in a network namespace, getaddrinfo() could take over 30 seconds before timing out while trying to contact a name server. To alleviate this, the hostname is remembered from when keepalived started. - Fix spelling of propagate in propagate_signal(). - Fix effective_priority after reload if tracked interface down. - Cosmetic grammatical changes. - Add debug option for dumping vrrp fd lists. - Fix calculation for vrrp fd timers. Starting or reloading keepalived when an interface that was tracked interface was failed was stopping other vrrp instances that were on the same interface but not using VMACs coming up. - Move code for initialising tracking priorities to vrrp_track.c. - Don't overwrite track file on reload. - Don't attempt to write track file if path not specified. - Fix compiling when not using --enable-vrrp-fd-debug. - Fix compiling with configure --enable-vrrp-fd-debug. - Add sync group track_bfds and track file status to config dump. - Move initialisation of track_files. - Don't alter effective_priority if track_file take vrrp instance down. - Don't log vrrp instance in fault state at reload if already fault. - Fix calculating fd timer if all vrrp sands are set to TIMER_DISABLED. - Don't make all sync groups transition to backup on reload If a sync group was in master state, and can still be after a reload then allow it to stay in master state. - Don't have track_bfd list in vrrp_sgroup_t in BFD not enabled. - Fix memory leak re vrrp_sgroup_t track lists. - Tidy up some freeing of MALLOC'd memory. Use FREE_PTR if it is not known if the pointer is valid, and don't clear the pointer afterr FREE/FREE_PTR since FREE does it anyway. - Add memory.c list size definition and move definition from memory.h. - Increase size of checksum value for MEM_CHECK. - Don't store checksum of memory allocation block. It can be calculated from the size, so do so. - Make the checksum for memory allocation blocks unsigned. - Use an enum for memory allocation block types. - Update comment re debug bit for memory detect error. - In memory alloc debug code report free or realloc for not alloc'd. - Allow for PIDs up to 2^22 (7 decimal digits). - Add function for dumping memory allocation while running. - Fix max memory allocation size calculations. - Fix reporting original and new file/line/func for realloc. - Check matching block for realloc is allocated. The same memory block may have been previously allocated and freed, so we need to make sure that the block we find is currently marked as allocated. - Use a new MEMCHECK struct for realloc overrun detected It was marking the allocated block as an overrun block, whereas it needs to be an allocated block, so use a new block to mark the overrun. - Tidy up working of a couple of memory allocation messages. - Use for loops rather than while blocks in memory allocation code. - Report number of mallocs and reallocs with MEMCHECK. - Attempt to log first free after double free in MEMCHECK. - Streamline use of buf/buffer in memory.c. - Always use first free entry in alloc_list for MEMCHECK. - Define MEMCHECK alloc_list size via configure. - Align keepalived_free() and keepalived_realloc(). - Make char * const where possible for MEMCHECK. - Merge MEMCHECK keepalived_free() and keepalived_realloc(). Most of the code was common between the two (or should have been), so it makes sense for them to use common code. - Ensure only relevant thread types run during shutdown. - Fix building without --enable-mem-check. - Use rbtree search for finding child thread on child termination. It was doing a linear search of the rbtree in timeout order. This commit adds another rbtree for child processes (vrrp track scripts and check_misc scripts), sorted by PID, to make the search by PID more efficient. - Make rbtree compare function thread_timer_cmp() more efficient. - Remove child_remover functionality - it was superfluous. - Fix checking that there are no duplicate vrrp instances configured The tuple {interface, family, vrid} must be unique. The check for this was being made completely incorrectly. - Delay creating vrrp notify FIFO. - Remove struct sockaddr_storage saddr from sock_t. - Use an rbtree for finding vrrp instance for received advert. Previously the code search a list of pointers to vrrp instances and looked for a matching fd and vrid. In order to optimise this, it was implemented using an mlist whose index was a hash of the fd and vrid. This commit changes the approach and uses an rbtree for each sock_t. Since the sock_t that the advert was received on is known, the rbtree search is only searching for a match on the vrid. Not only is this more efficient, but it is simpler, uses standard code, and reduces the code by over 60 lines. - Use an rbtree for finding vrrp instance for socket timeout. Previously the code search a list of pointers to vrrp instances and looked for matching file descriptor and sands < time_now. In order to optimise this, it was implemented using an mlist whose index was a hash of the fd. This commit changes the approach and uses a second rbtree for each sock_t. Since the sock_t that the timeout occurred on is known, the rbtree search is only searching for a match of the sands. Not only is this more efficient, but it is simpler, uses standard code, and reduces the code by over 220 lines. - Remove superfluous checks of rbtree node != NULL in rb_move(). - Remove superfluous check of node != NULL in rb_next(). - Update rbtree code to Linux 4.18.10. - Fix debug logging of sands timers before time_now. - Update rb_for_each_entry etc and rb_move to use rb_entry_safe. With the added definition of rb_entry_safe in the rbtree code updated to Linux 4.18.10, the refinition of rb_entry was reverted to the kernel definition. That meant that rb_for_each_entry, rb_for_eacn_entry_safe and rb_move neded to be updated to use rb_entry_safe rather than rb_entry. - Add support functions for rbtree rb_root_cached. This is in preparation for the use of rb_root_cached in the next patch. - Use cached rbtrees where the key is a timeval_t sands When the key of an rbtree is a timeval_t sands keepalived will frequently need to access the first node of the tree in order to calculate the next timeout. This applies to the read, write, child and timer threads queues, and also the vrrp queues on a sock_t. The use of cached rbtrees for these is ideal since it gives direct access to the first node of the queue. - Add thread_add_read_sands to avoid introducing timer errors. When using thread_add_read and the timeout was held as timeval_t, it was converted to and offset from time_now, and then converted back to a timeval_t, but time_now was updated, resulting in a slightly different value being used as the timeout. Using thread_add_read_sands() avoids the double conversion and results in the timeout being more accurate. - Replace NETLINK_TIMER with TIMER_NEVER. It makes the code easier to read, and since NETLINK_TIMER was defined to be TIMER_NEVER it doesn't change the functionality. - Handle preempt delays not expiring at same time on sync group If different vrrp instances in a sync group had preempt delays that expired at different times keepalived looped with very small to epoll_wait() until all preempt delays had expired, causing high CPU utilisation. Keepalived now reschedules vrrp instances with a delay of 3 * advert_int + skew time while waiting for all vrrp instances in the sync group to expire their preempt delays. - Fix segfault when receive netlink message for default route added. - Move vrf_master_index into conditional compilation block. - Store interface macvlan type. - Make vrp_master_ifp point to self for VRF master interfaces. - Log if cannot create a VMAC due to existing interface with same name. - Handle delete/create of macvlan i/fs which aren't keepalived's. - Tidying up keepalived_netlink.c. - Handle VRFs changing on macvlan i/fs which have VMACs configured on them. - Fix recreating our VMACs if they are deleted. - Fix detecting address add/deletion from underlying i/f of our vmacs. - Don't use configured_ifp or base_ifp if not _HAVE_VRRP_VMAC_. - Distinguish between VMAC on real i/f and no VMAC on macvlan i/f If keepalived is configured to have a non VMAC interface on a macvlan interface, we want to use the macvlan interface rather than the underlying interface, whereas if we have a VMAC interface on a macvlan interface, we create the VMAC on the underlying interface of the macvlan. - Update duplicate VRID check where vrrp instance configured on macvlan. If a VRRP instance is configured on a macvlan interface, the duplicate VRID check needs to be done on the underlying interface. - Check for VRID conflicts when changeable interfaces are added For example, a vrrp instance could be configured on a macvlan, and that macvlan could be deleted and recreated with another base interface. The VRIDs in this case need to be checked for duplicates against the base interface, and so the VRID check needs to be done dynamically. In order to allow VRID conflicts to produce config errors at startup, by default keepalived assumes that there won't be interface movements as described above, and will only handle it if the global_defs option 'dynamic_interfaces' is used along with the option 'allow_if_changes'. - Remove some comments inserted for tracking changes to code. - Fix building with --enable-debug configure option. - Check that '{'s and '}'s are balanced in the configuration file. - Allow more flexibility re placing of { and }. - Improve reporting additional '}'s in configuration. - Minor improvements re thread handling and cancellation. - Remove unused THREAD_IF_UP and THREAD_IF_DOWN. - Replace getpagesize() with sysconf(_SC_PAGESIZE). - Increase netlink receive buffer for dumps to 16KiB. - Dynamically set the netlink receive buffer size. - Sort out setting netlink receive buffer size. - added patches for changes found during the review of the dbus code: (boo#1015141) CVE-2018-19044 for https://github.com/acassen/keepalived/commit/04f2d32871bb3b11d7dc024039952f2fe2750306.patch CVE-2018-19045 for https://github.com/acassen/keepalived/commit/c6247a9ef2c7b33244ab1d3aa5d629ec49f0a067.patch https://github.com/acassen/keepalived/commit/5241e4d7b177d0b6f073cfc9ed5444bf51ec89d6.patch - enable dbus support on TW by default (boo#1015141) - enable json stats dump support ------------------------------------------------------------------- Thu Sep 13 07:28:25 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - use %license ------------------------------------------------------------------- Thu Sep 13 01:38:32 UTC 2018 - Marcus Rueckert <mrueckert@suse.de> - update to 2.0.7 see /usr/share/doc/packages/keepalived/ChangeLog - refreshed keepalive-init.patch: - reduced patch to minimal changes - made sure it actually reads our sysconfig file - refreshed linux-4.15.patch - enable http regexp support: new BR pcre2-devel - update rpmlintrc to actually match the error message: glob vs regexp ------------------------------------------------------------------- Mon Apr 30 14:41:46 UTC 2018 - crrodriguez@opensuse.org - Only Require insserv on distributions without systemd. - Fix systemd related requires/buildRequires - Do not run scriptlets that use insserv when using systemd ------------------------------------------------------------------- Thu Feb 22 10:07:17 UTC 2018 - dmueller@suse.com - add linux-4.15.patch ------------------------------------------------------------------- Wed Feb 21 14:52:29 UTC 2018 - dmueller@suse.com - update to 1.4.1: * Improve and fix use of getopt_long(). We musn't use a long option val of 1, since getopt_long() can return that value. getopt_long() also returns longindex == 0 when there is no matching long option, and there needs to be careful checking if there is an error to work out whether a long or short option was used, which is needed for meaningful error messages. * Write assert() messages to syslog. assert()s are nasty things, but at least let's get the benefit of them, and write the messages to syslog, rather than losing them down stderr. * Enable sorry server at startup if quorum down due to alpha mode If alpha mode is configured on sufficient checkers so that a virtual server doesn't have a quorum, we need to add the sorry server at startup, otherwise it won't be added until a quorum has been achieved and subsequently lost again. In the case where some of the checkers remain in the down state at startup, this would have meant that the sorry server never got added. * For virtual servers, ensure quorum <= number of real servers If the quorum were gigher than the number of real servers, the quorum for the real server to come up could never be achieved, so if the quorum is greater than the number of real servers, reduce it to the number of real servers. * Fix some SNMP keepalived checker integer types and default values. Some virtual server and real server values were being sent to SNMP with a signed type whereas the value is unsigned, so set the type field correctly. Some virtual server and real server values that apply to checkers are set to nonsense default values in order to determine if a value has been specified. Handle these values when reporting them to SNMP replying with 0 rather than a nonsense value. * Fix some MALLOC/FREE issues with notify FIFOs. * Add instance_name/config_id to alert emails' subjects if configured. If multiple instances of keepalived are running, either different instance_names and/or config_ids, it is useful to know which keepalived instance the email relates to. * Ensure that email body string isn't unterminated. Using strncpy() needs to ensure that there is a nul termination byte, so this commits adds always writing a nul byte to the end of the buffer. * Remove duplicate fault notification. * Fix problem with scripts found via PATH with a '/' in parameters. Recent discussions on issue #101 led to discovering that if an executable without a fully qualified name was specified as a script and there was a '/' character in the parameters, then the path resolution would not work. * Send SNMP traps when go from backup to fault due to sync group. Commit 020a9ab added executing notify_fault for vrrp instances transitioning from backup to fault state due to another instance in the sync group going to fault state. This commit adds sending SNMP traps in the same circumstance. * Revert "Add instance_name/config_id to alert emails' subjects if configured". This should be handled by setting router_id * Add config option to send smtp-alerts to file rather than send emails This is useful for debugging purposes. * Add additional entry to Travis-CI build matrix. * Fix segfault if no sorry server configured for a virtual server. ------------------------------------------------------------------- Mon Jan 22 13:03:55 UTC 2018 - mrueckert@suse.de - enable json stats and config dump support new BR: pkgconfig(json-c) - disable dynamic loading of libipset and link it instead - enable stacktrace support - turn on snmp-rfcv2 and snmp-rfcv3 support - do not reference the keepalived.socket in the rpm scriptlets ------------------------------------------------------------------- Fri Jan 12 08:53:51 UTC 2018 - lars@linux-schulserver.de - update to 1.4.0 * Add Linux build and runtime versions to -v output. * Log kernel version and build kernel version to log at startup. * Don't sleep for 1 send when exiting vrrp process if no vrrp instances. * With large configurations the syslog can get flooded and drop output. This commit adds options to not log to syslog, and also to log all output to files. * Add option to only flush log files before forking. * Don't poll netlink for all interfaces each time add a VMAC. We can poll for the individual interface details which significantly reduces what we have to process. * Print interface details in keepalived.data output. * Add high performace child finder code. The code to find the relevant thread to execute afer a child process (either a vrrp track script or a misc_check healthchecker) was doing a linear search for the matching pid, which if there are a large number of child processes running could become time consuming. The code now will enable high performance child finding, based on using mlists hashed by the pid, if there are 32 or more vrrp track scripts or misc check healthcheckers. The size of the mlist is based on the number of scripts, with a limit of 256. * Improve high performance child termination timeout code. * Preserve filename in script path name resolution. Some executables change their behaviour depending on the name by which they are invoked (e.g. /usr/sbin/pidof when it is a link to /usr/sbin/killall5). Using realpath() changes the file name part if it is a symbolic link. This commit resolves all symbolic links to directories, but leaves the file name part unaltered. It then checks the security of both the path to the link and the path to the real file. * Handle scripts names that are symbolic links properly. * Fix some RFC SNMP issues. * Fix removing left-over addresses if keepalived aborts. * Update openssl use to stop using deprecated functions openssl from version 1.1 deprecated certain functions that keepalived was using. This commit ceases using those functions if the version of openssl is >= 1.1. * Allow sync groups with only 1 member, but issue a warning. * Add replaceable parameters in configuration files. * Add multiline configuration definitions. * Fix keepalived.conf(5) man page. * Suppress error message when removing leftover addresses at startup. => find more changes at /usr/share/doc/packages/keepalived/ - rebase keepalive-init.patch - use upstream systemd service file instead providing an own one => removed keepalived.service - remove executable bit from samples in docdir - check that LVS support is enabled - optionally enable dump configuration and stats as JSON (via bcond) => BuildRequire libjson-c-devel - restrict /etc/keepalived permissions to root ------------------------------------------------------------------- Mon Nov 27 11:26:58 UTC 2017 - jengelh@inai.de - Do not suppress errors from useradd. - Ensure neutrality of description. ------------------------------------------------------------------- Thu Nov 27 09:11:55 UTC 2017 - igarcia@suse.com - update to 1.3.9: Revert using github tarball and use original source again. Too many fixes and features to list, refer to /usr/share/doc/packages/keepalived/ChangeLog for a detailed list. ------------------------------------------------------------------- Thu Nov 23 13:38:30 UTC 2017 - rbrown@suse.com - Replace references to /var/adm/fillup-templates with new %_fillupdir macro (boo#1069468) ------------------------------------------------------------------- Thu Feb 16 12:27:53 UTC 2017 - mrueckert@suse.de - use tarball from https://github.com/acassen/keepalived/issues/524 the original tarball did not build. This has the necessary fix applied. for the 1.3.4 update see the TODO entry in the preamble. ------------------------------------------------------------------- Wed Feb 15 11:38:16 UTC 2017 - mrueckert@suse.de - update to 1.3.3 Some minor fix, extensions and updates. snapcraft support. Refer to /usr/share/doc/packages/keepalived/ChangeLog for more infos. ------------------------------------------------------------------- Mon Dec 12 14:05:25 UTC 2016 - mrueckert@suse.de - fix building with libnfnetlink. the additional include path needs to be in CPPFLAGS instead of CFLAGS now. - enabled a few more features: - enhanced snmp support (V2/V3 RFC) - make sure we build with ipset/libiptc and routes support - prepared dbus support: waiting for boo#1015141 ------------------------------------------------------------------- Mon Dec 12 12:59:54 UTC 2016 - mrueckert@suse.de - update 1.3.2 - Security focused on notify heplers. Some minor fix and extensions. - changes from 1.3.1 - Quick script fix for regression brought by last release. - changes from 1.3.0 - New MAJOR release with stabilization fixes. Support to DBus. Conf extensions. Parser error log. Security extensions to run scripts more secure. - changes from 1.2.24 - MAJOR release with stabilization fixes and new features like support to network namespace. Refer to /usr/share/doc/packages/keepalived/ChangeLog for more infos. ------------------------------------------------------------------- Wed Jul 20 09:07:35 UTC 2016 - michael@stroeder.com - update to 1.2.23 Some VRRP fixes. Some Healthcheckers fixes. Refer to ChangeLog for more infos. ------------------------------------------------------------------- Fri Jul 8 10:32:22 UTC 2016 - mrueckert@suse.de - update to 1.2.22 Some VRRP fixes. Refer to ChangeLog for more infos. - update to 1.2.21 Some fixes for last major release 1.2.20. Extensions on vrrp framework. Refer to ChangeLog for more infos. - update to 1.2.20 BUNCH of extensions, fixes, cleanup & production considerations. Distro packages maintainers are strongly encouraged to upgrade. - new BR libnfnetlink-devel - we no longer ship the VRRP-MIB ------------------------------------------------------------------- Thu Feb 11 10:44:31 UTC 2016 - lars@linux-schulserver.de - enhanced keepalive-init.patch : + replace tabs with spaces + read /etc/sysconfig/keepalived, if exists and use the settings there instead of the default KEEPALIVED_OPTIONS in case the user changed them ------------------------------------------------------------------- Thu Jan 28 12:13:36 UTC 2016 - mrueckert@suse.de - use package name buildrequires on sle11 to fix building ------------------------------------------------------------------- Thu Jan 28 11:46:11 UTC 2016 - mrueckert@suse.de - enable snmp for better monitoring - enable sha1 support ------------------------------------------------------------------- Wed Oct 7 11:45:41 UTC 2015 - dimstar@opensuse.org - Update to version 1.2.19: + vrrp: fix checksum computation in vrrp v2 for socket family AF_INET. + Some cosmetics at Makefile stuff. - Changes from version 1.2.18: + some cosmetics changes (in memory and parser). + remove dead/not used code. + revert notify script brought by last release. + revert VRRP preemption speed up extension. + vrrp: ix vrrp removes incorrect IPv4 address when VIPs are removed. + vrrp: Re-enable VRRPv2 checksum on inbound pkts. - Changes from version 1.2.17: + zalloc use xalloc for consistency. + memory: fix wrong size calculation in zfree. + Fix keepalived snmp configuration. + Change comments to match kernel style. + smtp: Fix wrong algorithm in RCPT-TO building. + Lots of vrrp fixes. - Changes from version 1.2.16: + Properly close netlink channel to avoid fd leak. + Use getaddrinfo instead of gethostbyname to workaround glibc gethostbyname function buffer overflow (boo#949238). + Lots of ipvs fixes. ------------------------------------------------------------------- Wed Oct 7 10:31:50 UTC 2015 - mrueckert@suse.de - no longer install the init script on systemd systems ------------------------------------------------------------------- Wed Mar 11 13:21:29 UTC 2015 - dimstar@opensuse.org - Update to version 1.2.15: + Bugfixes. - Changes from version 1.2.14: + VRRP bugfixes and extensions. IPVS bugfixes and code code cleanup. - Changes from version 1.2.13: + VRRP fixes and extensions. Extrend and unify checker framework. ------------------------------------------------------------------- Mon Feb 2 01:32:37 UTC 2015 - crrodriguez@opensuse.org - Build with -DOPENSSL_NO_SSL_INTERN, if package starts accessing the SSL library internals it must fail to build now, in upcoming openSSL versions structures are opaque. - BuildRequire libnl3 - Do not strip binaries, fix -debuginfo packages. ------------------------------------------------------------------- Sun Nov 09 05:21:00 UTC 2014 - Led <ledest@gmail.com> - fix bashisms in pre script ------------------------------------------------------------------- Thu Jul 31 14:28:08 UTC 2014 - dimstar@opensuse.org - Rename rpmlintrc to %{name}-rpmlintrc. Follow the packaging guidelines. ------------------------------------------------------------------- Tue Feb 11 08:12:55 UTC 2014 - boris@steki.net - updated to latest upstream version 1.2.12 + Fix reallocation issue introduced in last merge. + Fix some minor memory leaks. + Better libnl support and selection. + VRRP unicast TTL fix. + Support to newer libnl. + More IPv6 support. + Fix/extend VRRP gratuitous ARP handling. + Support xmit VRRP packets from base VMAC interface. + VRRP multicast group tweaking. + Fixed VRRP socket sync while leaving FAULT state. + Code cleanup and cosmetics. ------------------------------------------------------------------- Tue Jan 7 10:55:42 UTC 2014 - speilicke@suse.com - Add cyrus-sasl for old distros ------------------------------------------------------------------- Tue Nov 19 14:01:47 UTC 2013 - speilicke@suse.com - Update to version 1.2.9: + Extended VRRP code for faster sync and transition. + Fixed VRRP unicast code to support routed packet. + Fixed VRRP checksum computation. + Extended VRRP code tweaking IPv6 VIP install by disabling DAD algo and setting deprecated flag. + Fixed some issues in checker framework while processing hysteresis. + Extended checker framework to support use of status_code and digest at a time. - Changes from version 1.2.8: + Add support for VRRP unicast. + Add support for VRRP IPv6 routes. + Add support to LVS One-Packet Scheduling. + Add CLI core framework. + Misc bugfixes, typo and cosmetics. - Drop keepalived_man_fix.patch: merged upstream ------------------------------------------------------------------- Tue Nov 20 16:11:59 UTC 2012 - mrueckert@suse.de - initial package of 1.2.7
Locations
Projects
Search
Status Monitor
Help
OpenBuildService.org
Documentation
API Documentation
Code of Conduct
Contact
Support
@OBShq
Terms
openSUSE Build Service is sponsored by
The Open Build Service is an
openSUSE project
.
Sign Up
Log In
Places
Places
All Projects
Status Monitor
