Rebase onto upstream drbd-utils v9.32.0 and sync netlink headers to a6d1604c (36/25)#4
Open
astef wants to merge 85 commits into
Open
Rebase onto upstream drbd-utils v9.32.0 and sync netlink headers to a6d1604c (36/25)#4astef wants to merge 85 commits into
astef wants to merge 85 commits into
Conversation
Enables loading of event files into DRBDmon that were created using the --timestamps option with drbdsetup events2 by preprocessing the event lines before they are passed on to DRBDmon.
The device config has been neglected so far. Initially, it only contained the max-bio-size and the intentionally diskless flag. Later I added the block-size member. I forgot to output these fields in the output of 'drbdsetup show'. This patch adds outputting the device configuration options in the disk options.
This allows dnf/yum updates to succeed without --allowerasing.
Fedora just released version 42, which means some tools were missing from the test job. We also pin to fedora 42.
This is most likely a fallout from changes introduced in f962315 We have "versioned man pages" and create symlinks to them (e.g., drbdmeta.8.gz -> drbdmeta-9.0.8.gz). Creating and deleting these is managed in post and preun scripts. drbdmon is special as there is no versioned man page, there is only drbdmon.8.gz (the final file, no symlinks). As is, we tried to rm the file twice, once in preun and then as part of "files". This resulted in a completely harmless message that a non-existent file was tried to get removed.
The long option name is --initialize-bitmap, not --initialize-bitmap-mode. Fix the man page.
Signed-off-by: Aleksandr Stefurishin <aleksandr.stefurishin@flant.com>
Error out with a useful message. Previously, it would just say 'input in flex scanner failed'.
Allow "dump" to take a specific volume (for example specifying "minor-N"). Also add a new sub command, sh-list-config-file, that will list <path-to-config-file>:<line-number>:resource-name
When handling supposedly constant strings, we should use (const char *) more often.
The message "Can not open '%s'.\n.", with that trailing dot, has been accidentally introduced in the very first attempts of writing drbdadm, drbdadm_main.c was not even four weeks old at that time. It has been copy-pasted around since 2002 unnoticed. Finally drop those trailing dots.
Its counterpart --config-to-exclude is already aggregating. This should allow to "plausibility check" multiple changes to multiple config files with one invocation.
Source code pattern is: r = ctx_by_name(&ctx, "id", ...); if (!ctx.res) log error if (r) exit 1 If you specify a number of resources on the command line, and the first one is defined, we forgot to log the error, but still do the exit 1. We need to reset ctx with each call to ctx_by_...(). This also changes the logic: we no longer exit on the first "non-existing" resource, but log the error, accumulate the "exit 1", then still continue with all other resource names that are listed on the command line.
This is a revert/rework of 6407c42 (drbdadm: reset ctx in ctx_by_* functions, 2025-05-15) we need to reset (v84 drbdadm already did), but must not reset the ctx->arg respectively ctx->cmd, that does not change while iterating over resources, and is not expected to be reset to NULL. Also add two command line test cases that would have caught this on the previous commit.
These are available on all the distros we care about, the oldest currently being SLES 12sp5 and RHEL 7.9.
No need to template this in autotools.
No need to set this with autotools. Contrary to previous suggestions, packaging guidelines do seem to indicate that things like "make" and "gcc" should be included.
We no longer support any systems using rpm < 4.6. The oldest distros we support both use RPM 4.11.X
Some files that are packaged as the drbd-pacemaker package are installed in the wrong make section. This causes rpmbuild errors if run with pacemaker support disabled.
No need to set a bash completion suffix, as well as weird optional requirements for bash completions. Also, move the completions to the system location instead of the admin config directory.
Our udev rules have been supported for more than 20 years, no need to check for ancient udev versions to disable the rules. Should it be needed, we already support optional removal of the udev builds.
It appears that phil forgot to add sh-list-adjustable to the is_adjust condition, which led to a segfault and connection-mesh errors. I am also adding a test case to highlight the issue. That unveiled that if there were a running resource that needs to be taken down, the sh-list-adjustable would output that resource name twice. Also fixing that corner case.
We prepare an error description string already. If drbdsetup exits with error, at least print that error description if available.
Pass options to genl_connect(). For now, that is send- and receive buffer sizes. "Too much" activity causes many events to be multicast via netlink. That may overrun the receive buffer. We want to be able to increase the receive buffer size in large or busy setups.
If we not immediately succeed setting the SO_RCVBUF, retry with SO_RCVBUFFORCE.
Allow to override rcvbuf size from environment or command line: DRBD_GENL_RCVBUF_SZ events2 --rcvbuf Uses our "m_strol", so units like 10m are fine.
drbdsetup(8) man page update about how to adjust the netlink receive buffer size.
This reverts commit 7cd43be. I will re-implement this in another way.
Prepare for improving `drbdadm adjust` by giving it the complete connection state for each connection.
There is no point in having the CFG_STAGE values spread throughout the code in the callsites of schedule_deferred_cmd(). Centralize them in struct adm_cmd.
…cies So far, the CFG_STAGE (enum drbd_cfg_stage) constants allowed a static ordering of the scheduled commands. From now on, scheduled_deferred_cmd() returns a pointer to a deferred_cmd and takes such a pointer as an optional dependency as an argument.
When drbd.conf changes how a diskless node reaches nodes holding the data, applying the changes might be nontrivial. Consider a diskless primary, having one peer with an UpToDate disk. One might remove the connection between the diskless node and that node with the backing disk and add another connection to another diskful node at the same time. That is a valid change when the two diskful nodes have a connection. So far, `drbdadm adjust` has failed to apply such a change, as it attempts to remove the connection first and then add the other one. Fixing that by leveraging an explicit dependency from the del-peer to a wait-connect for the new peer. The failure of a disconnect command triggered the previous implementation, so it relied on the DRBD state engine itself to reject the first disconnect. The new implementation is overly cautious, so it attempts to establish the new connection before deleting the old one, even if it is not strictly necessary.
Consider a diskless primary, having two storage servers configured as peers. When the diskless primary switches to another network, drbdadm adjust should reconfigure the connections first that are not currently connected, give the reconfigured connection time to establish and finally apply changes to the connections that were established initially.
…just This is a basic test that verifies the commands emitted by drbdadm adjust when it starts from an empty state. This was useful for verifying that the work on the drbdadm adjust code does not break the existing behaviour.
In this case we do not need to wait for the new connection to be established before setting the options on the existing connection.
The only user of this script was drbd-reactor in its early days. This was rewritten in the drbd-reactor project a long time ago in rust. We have enough users for the rust implementation which used that for years. The "better" (i.e., non-shell) version can be found here: https://github.com/LINBIT/drbd-reactor/blob/master/src/bin/ocf-rs-wrapper.rs https://github.com/LINBIT/drbd-reactor/blob/master/example/ocf.rs%40.service
double_quote_string() does not handle NULL values up to now.
Program received signal SIGSEGV, Segmentation fault.
0x000000000040ea21 in double_quote_string (str=0x0) at config_flags.c:668
(gdb) bt
0 0x000000000040ea21 in double_quote_string (str=0x0) at config_flags.c:668
1 0x0000000000404d48 in __print_options (attr=<optimized out>, ctx=0x42bd50 <show_net_options_ctx>,
sect_name=sect_name@entry=0x41bab2 "net", opened=true, opened@entry=false, close=close@entry=true) at drbdsetup.c:1357
2 0x0000000000407edb in __print_options (attr=<optimized out>,
ctx=<error reading variable: Cannot access memory at address 0x0>, sect_name=<optimized out>, opened=<optimized out>,
close=<optimized out>) at drbdsetup.c:1381
3 print_options (attr=<optimized out>, ctx=<error reading variable: Cannot access memory at address 0x0>,
sect_name=0x41bab2 "net") at drbdsetup.c:1383
4 print_options (attr=<optimized out>, ctx=<error reading variable: Cannot access memory at address 0x0>,
sect_name=0x41bab2 "net") at drbdsetup.c:1381
5 show_connection (connection=0x436d40, peer_devices=<optimized out>) at drbdsetup.c:2250
6 show_resource_list (resources_list=0x436340, old_objname=0x7fffffffe632 "r0") at drbdsetup.c:2440
7 show_cmd (cm=<optimized out>, argc=<optimized out>, argv=<optimized out>) at drbdsetup.c:2574
8 0x000000000040c997 in drbdsetup_main (argc=<optimized out>, argv=0x7fffffffe390) at drbdsetup.c:4861
9 0x00007ffff7c2a1ca in __libc_start_call_main (main=main@entry=0x400f40 <main>, argc=argc@entry=4,
argv=argv@entry=0x7fffffffe388) at ../sysdeps/nptl/libc_start_call_main.h:58
10 0x00007ffff7c2a28b in __libc_start_main_impl (main=0x400f40 <main>, argc=4, argv=0x7fffffffe388, init=<optimized out>,
fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffe378) at ../csu/libc-start.c:360
11 0x0000000000400f75 in _start ()
Fixes: 8b48c59 ("drbdadm adjust: detect and drop options unsupported by kernel")
For consistency with all other statements. Fixes: 8b48c59 ("drbdadm adjust: detect and drop options unsupported by kernel")
Its the only distribution that does not support the syntax.
Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com>
our preset file acutally depends on pkg-config and systemd.pc. We can add pkg-config as a general dependency. systemd.pc moved from "systemd" to "sytemd-dev", so we can not add it here, but downstream distros should add the one that fits.
Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com>
Expose the new kernel resource option quorum-dynamic-voters as a boolean in drbdsetup and drbdadm. Defaults to yes, preserving the existing quorum behavior. Update drbd-headers submodule to include the new option definition. Signed-off-by: David Magton <david.magton@flant.com>
Expose the new kernel disk option non-voting as a boolean in drbdsetup
and drbdadm. Defaults to no, preserving existing behavior.
Usage:
drbdsetup disk-options <minor> --non-voting=yes
or in drbd.conf:
on shadow {
disk { non-voting yes; }
}
Signed-off-by: David Magton <david.magton@flant.com>
The flant-9.2 branch was renamed to flant in 3p-drbd-headers, causing submodule checkout failures during build. Signed-off-by: David Magton <david.magton@flant.com>
Move the drbd-headers submodule from ee91191b to a6d1604c, which renumbers our two custom netlink fields: - non_voting (disk_conf): 28 -> 36 - quorum_dynamic_voters (res_opts): 17 -> 25 This keeps drbdsetup/drbdadm wire-compatible with the kernel module, which adopted the same renumbering in 9.2.18-flant.4. Both repos now pin the identical headers commit a6d1604c.
flant-9.32.0 is flant rebased onto upstream drbd-utils v9.32.0, with the drbd-headers submodule re-pinned to a6d1604c (netlink IDs 36/25) to match 3p-drbd 9.2.18-flant.4+. This merge records the pre-rebase branch as an ancestor so PR #4 can fast-forward; the tree is taken entirely from flant-9.32.0 (strategy=ours). The base only added the gitlink ee91191b (a direct ancestor of a6d1604c) and the two drbdsetup options, both already present in the rebased head (verified). The GitHub "conflict" was only the submodule gitlink, which GitHub cannot fast-forward without the headers objects; this merge resolves it. Only obsolete metadata is superseded.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
12 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Moves our userspace-tools fork from v9.31.0 onto upstream
v9.32.0, keeps our twodrbdsetupoptions, and re-pinsdrbd-headersto match the kernel's renumbered netlink ABI.v9.32.0— the newest utils still ABI-compatible with the DRBD-9.2 flant headers (v9.33+/9.34 moved to an incompatible headers line).drbdsetup --non-votingand--quorum-dynamic-voters.drbd-headers→a6d1604c(non_voting=36,quorum_dynamic_voters=25) to match3p-drbd9.2.18-flant.4+.Ships with kernel
flant-9.2.18(9.2.18-flant.9) and headersflant— identicala6d1604con all three sides.