RepoMirrors
/
btrfs-progs
mirror of https://github.com/kdave/btrfs-progs
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

btrfs-progs: fix exclusive op enqueue timeout 

There's a report that 'btrfs balance start --enqueue' does not properly
wait when there are multiple instances started. The command does a busy
wait instead of timeouts.

Strace output:

     0.000006 pselect6(5, NULL, NULL, [4], {tv_sec=60, tv_nsec=0}, NULL) = 1 (except [4], left {tv_sec=59, tv_nsec=999999716})
     0.000008 pselect6(5, NULL, NULL, [4], {tv_sec=29, tv_nsec=999999000}, NULL) = 1 (except [4], left {tv_sec=29, tv_nsec=999998786})

After the first select there's almost the entire time left, the second
one starts right after it.

Polling/selecting sysfs files is possible under some conditions:

- the file descriptor must be reopened before each poll/select
- the whole buffer must be read too

With that in place it now works as expected. The remaining timeout logic
is slightly adjusted to wait at most 10 seconds so the pending jobs do
not wait too long if there's still a lot of time left from the first
select.

Issue: #746
Signed-off-by: David Sterba <dsterba@suse.com>
This commit is contained in:
David Sterba 2024-02-20 09:53:45 +01:00
parent 06d0ed483d
commit d03594b031
1 changed files with 20 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
common/utils.c
View File

@ -1345,26 +1345,45 @@ int check_running_fs_exclop(int fd, enum exclusive_operation start, bool enqueue
fflush(stdout);
}
/*
* The sysfs file descriptor needs to be reopened and all data read
* before each select().
*/
while (exclop > 0) {
fd_set fds;
struct timeval tv = { .tv_sec = 60, .tv_usec = 0 };
char tmp[1024];
close(sysfs_fd);
sysfs_fd = sysfs_open_fsid_file(fd, "exclusive_operation");
if (sysfs_fd < 0)
return sysfs_fd;
FD_ZERO(&fds);
FD_SET(sysfs_fd, &fds);
ret = read(sysfs_fd, tmp, sizeof(tmp));
ret = select(sysfs_fd + 1, NULL, NULL, &fds, &tv);
if (ret < 0) {
ret = -errno;
break;
}
if (ret > 0) {
close(sysfs_fd);
sysfs_fd = sysfs_open_fsid_file(fd, "exclusive_operation");
if (sysfs_fd < 0)
return sysfs_fd;
FD_ZERO(&fds);
FD_SET(sysfs_fd, &fds);
ret = read(sysfs_fd, tmp, sizeof(tmp));
/*
* Notified before the timeout, check again before
* returning. In case there are more operations
* waiting, we want to reduce the chances to race so
* reuse the remaining time to randomize the order.
*/
tv.tv_sec /= 2;
tv.tv_sec = (tv.tv_sec % 10) + 1;
ret = select(sysfs_fd + 1, NULL, NULL, &fds, &tv);
exclop = get_fs_exclop(fd);
if (exclop <= 0)