set masterfd_stdout before regsitering ctrl_cb #71
Conversation
This file contains 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
exec with conmon ran into issues because the manager attempted to resize the window before stdout was configured. Thus, we got the ctrl_cb from the caller, but had no where to send the ioctl. Fix this by registering the callback after we set stdout. Signed-off-by: Peter Hunt <pehunt@redhat.com>
Closed
|
partial fix for containers/podman#3903 |
|
LGTM |
| * if we didn't set it here, we'd risk attempting to run ioctl on | ||
| * a negative fd, and fail to resize the window */ | ||
| g_unix_fd_add(terminal_ctrl_fd, G_IO_IN, ctrl_cb, NULL); | ||
|
|
There was a problem hiding this comment.
🥇 on an excellent comment
|
LGTM assuming happy tests |
apyrgio
pushed a commit
to freedomofpress/maint-dangerzone-conmon
that referenced
this pull request
Jan 24, 2024
Squashed commit of the following: commit 60b42f2 Author: Peter Hunt <pehunt@redhat.com> Date: Mon Jan 6 10:52:09 2020 -0500 bump to version 2.0.9 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 1560392 Author: Mrunal Patel <mrunalp@gmail.com> Date: Sat Jan 4 13:04:58 2020 -0800 Add TimedOutMessage to config to share with go code Signed-off-by: Mrunal Patel <mrunalp@gmail.com> commit b17d81b Author: Riccardo Schirone <sirmy15@gmail.com> Date: Fri Dec 20 17:38:42 2019 +0100 Fixes format string to limit the size of the string to 10 characters %10s just ensures that when printing, at least 10 characters are written. However, in this case we probably want *at most* 10 characters to be written. Signed-off-by: Riccardo Schirone <sirmy15@gmail.com> commit 3a11e26 Author: Riccardo Schirone <sirmy15@gmail.com> Date: Fri Dec 20 18:20:28 2019 +0100 g_file_{get,set}_contents return a gboolean to indicate failure, use it Signed-off-by: Riccardo Schirone <sirmy15@gmail.com> commit c2e2e67 Merge: fe2ebfd 0c66b66 Author: Daniel J Walsh <dwalsh@redhat.com> Date: Fri Dec 13 14:35:09 2019 -0500 Merge pull request containers#105 from haircommander/bump-2.0.8 Bump 2.0.8 commit 0c66b66 Author: Peter Hunt <pehunt@redhat.com> Date: Fri Dec 13 14:22:13 2019 -0500 bump to 2.0.9-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 5ca7df2 Author: Peter Hunt <pehunt@redhat.com> Date: Fri Dec 13 14:22:00 2019 -0500 bump to 2.0.8 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit fe2ebfd Merge: c8f7443 5bbd452 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Fri Dec 13 20:20:44 2019 +0100 Merge pull request containers#104 from haircommander/revert-oom Revert "conmon: check oom counter on cgroup v1" commit 5bbd452 Author: Peter Hunt <pehunt@redhat.com> Date: Fri Dec 13 12:09:27 2019 -0500 Revert "conmon: check oom counter on cgroup v1" This reverts commit 4100fb2. We found issues where the reading of the memory control file was racing with the cgroup manager's write. while we want to prevent false positives, not having false negatives is more important. Revert this commit until we find a more complete solution. Signed-off-by: Peter Hunt <pehunt@redhat.com> commit c8f7443 Merge: 036ff29 7d8faa7 Author: Mrunal Patel <mrunal@me.com> Date: Thu Dec 12 15:13:27 2019 -0800 Merge pull request containers#102 from haircommander/cgroup_v2_persist persist oom files on cgroup v2 commit 7d8faa7 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Dec 12 11:04:52 2019 -0500 persist oom files on cgroup v2 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 036ff29 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Dec 12 13:56:07 2019 -0500 bump to v2.0.8-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit d532cae Author: Peter Hunt <pehunt@redhat.com> Date: Thu Dec 12 13:55:53 2019 -0500 bump to v2.0.7 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit faccd40 Merge: 0760156 7d4c6be Author: Mrunal Patel <mrunal@me.com> Date: Thu Dec 12 10:49:47 2019 -0800 Merge pull request containers#100 from giuseppe/splice conmon: use splice(2) to copy from stdin commit 0760156 Merge: 4100fb2 dc23660 Author: Mrunal Patel <mrunal@me.com> Date: Thu Dec 12 10:48:58 2019 -0800 Merge pull request containers#101 from haircommander/bump-2.0.7 Bump 2.0.7 commit dc23660 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Dec 12 12:59:36 2019 -0500 bump to version 2.0.8-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit c35049c Author: Peter Hunt <pehunt@redhat.com> Date: Thu Dec 12 12:59:18 2019 -0500 bump to version 2.0.7 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 7d4c6be Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Thu Dec 12 15:24:43 2019 +0100 conmon: use splice(2) to copy from stdin the first advantage is that it is much faster as data is not copied from the kernel to the userspace and the back to the kernel, and the second advantage is that we don't block on writ'ing on the fd all the data we've buffered. Closes: containers#99 Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit 4100fb2 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Thu Dec 12 12:14:05 2019 +0100 conmon: check oom counter on cgroup v1 when running on cgroup v1, also check whether the OOM counter is > 0 before writing the "oom" file. Previously, conmon was detecting an OOM every time an event fd was triggered. Unfortunately that is not enough, as an event on the eventfd is also triggered when the cgroup is deleted. That could cause a race condition where conmon sees the event from the cgroup deletion before the SIGCHLD+SIGUSR1. That could happen for example when systemd CollectMode is set for failed units. Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1782601 Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit 95ed45a Merge: 8ba9575 82ef9c7 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Thu Dec 12 12:21:59 2019 +0100 Merge pull request containers#97 from mrunalp/persist_oom_file Persist the oom file when --persist-dir is specified commit 82ef9c7 Author: Mrunal Patel <mrunalp@gmail.com> Date: Wed Dec 11 14:49:23 2019 -0800 Persist the oom file when --persist-dir is specified Signed-off-by: Mrunal Patel <mrunalp@gmail.com> commit 8ba9575 Author: Peter Hunt <pehunt@redhat.com> Date: Wed Dec 11 09:04:43 2019 -0500 bump to v2.0.7-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 29c3367 Author: Peter Hunt <pehunt@redhat.com> Date: Wed Dec 11 09:04:25 2019 -0500 bump to v2.0.6 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit ba14d9c Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Wed Dec 11 14:05:44 2019 +0100 conmon: kill the process group on timeout when the timeout is expired, instead of killing the process, make sure to kill the entire process group. It is not safe as using pid cgroup or a pid namespace, which we cannot use as the script must run inside of the container pid namespace, but it addresses most common cases like bash forking a new process. Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1774184 Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit bc9e976 Author: Peter Hunt <pehunt@redhat.com> Date: Tue Dec 10 12:48:56 2019 -0500 bump to v2.0.6-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit f91c170 Author: Peter Hunt <pehunt@redhat.com> Date: Tue Dec 10 12:48:39 2019 -0500 bump to v2.0.5 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit c792503 Author: Mrunal Patel <mrunalp@gmail.com> Date: Mon Dec 9 11:23:02 2019 -0800 Write exit file to container bundle directory Write the exit file under the container bundle directory so it is available even when we restart the node. Also, keeping it scoped there makes the cleanup easier. Signed-off-by: Mrunal Patel <mrunalp@gmail.com> commit fd5ac47 Merge: df8c6aa c6835ed Author: Daniel J Walsh <dwalsh@redhat.com> Date: Mon Dec 9 12:27:55 2019 -0500 Merge pull request containers#91 from haircommander/version-2.0.4 bump to version 2.0.4 commit c6835ed Author: Peter Hunt <pehunt@redhat.com> Date: Mon Dec 9 11:04:44 2019 -0500 bump to version 2.0.5-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit d930e38 Author: Peter Hunt <pehunt@redhat.com> Date: Mon Dec 9 11:04:10 2019 -0500 bump to version 2.0.4 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit df8c6aa Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Mon Dec 2 16:41:03 2019 +0100 conmon: fix tight loop on OOM inotify fails with EINVAL when trying to read an event with a too small buffer. Increase the size to "sizeof(struct inotify_event) + NAME_MAX + 1", that is enough to read at least one event. Closes: containers#87 Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit 42bce45 Merge: f6d23b5 5ae9942 Author: Daniel J Walsh <dwalsh@redhat.com> Date: Fri Nov 29 06:34:52 2019 -0500 Merge pull request containers#89 from DaanDeMeyer/reduce-error-impact Initialize the endpoint for attach before calling fork. commit 5ae9942 Author: Daan De Meyer <daan.j.demeyer@gmail.com> Date: Wed Nov 27 21:24:16 2019 +0100 Initialize the endpoint for attach before calling fork. This way, when an error occurs in the attach endpoint logic, conmon can exit immediately without having to clean up the child process. commit f6d23b5 Merge: 098fcce 951052b Author: Daniel J Walsh <dwalsh@redhat.com> Date: Mon Nov 11 15:48:46 2019 -0500 Merge pull request containers#83 from haircommander/bump-2.0.4 Bump to 2.0.3 commit 951052b Author: Peter Hunt <pehunt@redhat.com> Date: Mon Nov 11 10:16:39 2019 -0500 version 2.0.4-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit eb5fa88 Author: Peter Hunt <pehunt@redhat.com> Date: Mon Nov 11 10:16:23 2019 -0500 version 2.0.3 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 098fcce Merge: 002da25 496e8e4 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Mon Nov 11 16:09:42 2019 +0100 Merge pull request containers#82 from haircommander/try_other_pipe Handle window resize events in a different fifo commit 496e8e4 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Nov 7 22:01:48 2019 -0500 config: include the two ctrl events Rather than having a comment saying "this should stay in sync", let's define a variable and import it there, so we're guarenteed to stay in sync Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 1ed4f02 Author: Peter Hunt <pehunt@redhat.com> Date: Tue Nov 5 15:25:00 2019 -0500 Handle window resize events in a different fifo The terminal_ctrl_fd was being overloaded with events. CRI-O wants to reopen the log files at any time, and podman wants to not drop window resize events on exec. But we can have both. Introduce winsz_fd_*, a special fifo pair for window resize events. The terminal_ctrl passes window resize events to the write side, and the read side can process them (after masterfd_stdout is initialized, so we have somewhere to write to and we don't drop events). Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 002da25 Merge: 19a1d5c 8d66f44 Author: Daniel J Walsh <dwalsh@redhat.com> Date: Tue Nov 5 09:13:28 2019 -0500 Merge pull request containers#81 from giuseppe/opt-tag src: add new option --log-tag commit 8d66f44 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Tue Nov 5 09:59:07 2019 +0100 src: add new option --log-tag it specifies an additional tag to use for logging messages. When using journald, it adds a tag "CONTAINER_TAG". Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit 19a1d5c Author: Sascha Grunert <sgrunert@suse.com> Date: Tue Oct 15 13:08:12 2019 +0200 Add 'trace' log level support The trace log level is currently not supported by conmon, but seems to be a valid level for container runtimes like CRI-O and podman. This results that the container currently won't start if the log level is set to 'trace'. Beside this, we now case insensitively compare the log level string to make it more fail safe. Signed-off-by: Sascha Grunert <sgrunert@suse.com> commit bc758d8 Merge: e37aa48 cb4d0e3 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Fri Oct 11 08:23:02 2019 +0200 Merge pull request containers#77 from haircommander/v2.0.2-bump bump to 2.0.2 commit cb4d0e3 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Oct 10 08:42:17 2019 -0400 version 2.0.3-dev Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 65fe022 Author: Peter Hunt <pehunt@redhat.com> Date: Thu Oct 10 08:41:45 2019 -0400 version 2.0.2 Signed-off-by: Peter Hunt <pehunt@redhat.com> commit e37aa48 Merge: 0e888a9 067c0a5 Author: Daniel J Walsh <dwalsh@redhat.com> Date: Thu Oct 10 06:42:21 2019 -0400 Merge pull request containers#76 from giuseppe/fix-podman-hang conmon: close the sync pipe before exit command commit 067c0a5 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Wed Oct 9 17:41:06 2019 +0200 conmon: close the sync pipe before exit command before calling the exit command be sure the sync pipe is closed so that the waiting process can stop waiting. It solves an issue with Podman where the cleanup process could hang since the container is already locked by the podman process waiting on the sync pipe. The reproducer here: podman --runtime /bin/false run --rm alpine true Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit 0e888a9 Author: Sascha Grunert <sgrunert@suse.com> Date: Mon Oct 7 11:45:45 2019 +0200 Add static binary build to Cirrus CI Signed-off-by: Sascha Grunert <sgrunert@suse.com> commit 422ce21 Merge: 7286971 d7c90bd Author: Daniel J Walsh <dwalsh@redhat.com> Date: Wed Sep 18 07:34:52 2019 -0400 Merge pull request containers#74 from haircommander/conn_sock_config Add CONN_SOCK_BUF_SIZE to config commit d7c90bd Author: Peter Hunt <pehunt@redhat.com> Date: Tue Sep 17 15:04:35 2019 -0400 Add CONN_SOCK_BUF_SIZE to config instead of having a note that says to sync the value Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 7286971 Author: Valentin Rothberg <rothberg@redhat.com> Date: Fri Sep 13 16:20:48 2019 +0200 bump to v2.0.2-dev Signed-off-by: Valentin Rothberg <rothberg@redhat.com> commit 4dc8bcf Author: Valentin Rothberg <rothberg@redhat.com> Date: Fri Sep 13 16:20:28 2019 +0200 v2.0.1 * set masterfd_stdout before regsitering ctrl_cb * Add static & containerized targets to Makefile. * Makefile: don't fail if clean is called without a build * Fix compilation with clang * conmon: propagate error code from the container Signed-off-by: Valentin Rothberg <rothberg@redhat.com> commit 35437e4 Merge: 4d36cb6 9e1e1ff Author: Valentin Rothberg <rothberg@redhat.com> Date: Fri Sep 13 10:11:09 2019 +0200 Merge pull request containers#71 from haircommander/stdout-wait-ctrl set masterfd_stdout before regsitering ctrl_cb commit 9e1e1ff Author: Peter Hunt <pehunt@redhat.com> Date: Thu Sep 12 15:59:20 2019 -0400 set masterfd_stdout before regsitering ctrl_cb exec with conmon ran into issues because the manager attempted to resize the window before stdout was configured. Thus, we got the ctrl_cb from the caller, but had no where to send the ioctl. Fix this by registering the callback after we set stdout. Signed-off-by: Peter Hunt <pehunt@redhat.com> commit 4d36cb6 Merge: fac48d0 58ce137 Author: Daniel J Walsh <dwalsh@redhat.com> Date: Wed Sep 4 16:06:00 2019 -0400 Merge pull request containers#69 from mgoltzsche/static-build Add static & dockerized targets to Makefile. commit 58ce137 Author: Max Goltzsche <max.goltzsche@gmail.com> Date: Tue Sep 3 21:57:38 2019 +0200 Add static & containerized targets to Makefile. Signed-off-by: Max Goltzsche <max.goltzsche@gmail.com> commit fac48d0 Merge: 8a56404 6ef63df Author: Daniel J Walsh <dwalsh@redhat.com> Date: Sat Aug 31 06:25:52 2019 -0400 Merge pull request containers#68 from agners/master Makefile: don't fail if clean is called without a build commit 6ef63df Author: Stefan Agner <stefan@agner.ch> Date: Fri Aug 30 17:49:22 2019 +0200 Makefile: don't fail if clean is called without a build Do not fail when calling the clean target without building first. Signed-off-by: Stefan Agner <stefan@agner.ch> commit 8a56404 Merge: af8e5a6 6d4f502 Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Tue Aug 20 16:33:23 2019 +0200 Merge pull request containers#66 from openSUSE/clang-fix Fix compilation with clang commit 6d4f502 Author: Sascha Grunert <sgrunert@suse.com> Date: Tue Aug 20 15:27:38 2019 +0200 Fix compilation with clang This fixes the missing field initializers error when compiling with clang: ``` src/conmon.c:119:7: error: missing field 'short_name' initializer {NULL}}; ^ ``` Signed-off-by: Sascha Grunert <sgrunert@suse.com> commit af8e5a6 Merge: a1cdfff 74da3fb Author: Daniel J Walsh <dwalsh@redhat.com> Date: Mon Aug 19 10:25:01 2019 -0400 Merge pull request containers#65 from giuseppe/conmon-return-exit-code-container conmon: propagate error code from the container commit 74da3fb Author: Giuseppe Scrivano <gscrivan@redhat.com> Date: Mon Aug 19 13:34:13 2019 +0200 conmon: propagate error code from the container at exit propagate the return code from the watched container. This is useful when running in systemd so it can properly detect a failure and restart the container if necessary. Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> commit a1cdfff Author: Peter Hunt <pehunt@redhat.com> Date: Fri Aug 9 09:50:56 2019 -0400 v2.0.1-dev Signed-off-by: Peter Hunt <pehunt@redhat.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
exec with conmon ran into issues because the manager attempted to resize the window before stdout was configured. Thus, we got the ctrl_cb from the caller, but had no where to send the ioctl.
Fix this by registering the callback after we set stdout.
Signed-off-by: Peter Hunt pehunt@redhat.com