On Fri, 6 Jun 2025 at 07:57, Sergey Senozhatsky
<senozhatsky@xxxxxxxxxxxx> wrote:
>
> Hi,
>
> We are seeing a number of cases when blocked fuse requests prevent
> the system from suspending, which is a little important on laptops.
> Usually something like this:
>
> [ 36.281038] Freezing user space processes
> [ 56.284961] Freezing user space processes failed after 20.003 seconds (1 tasks refusing to freeze, wq_busy=0):
> [ 56.285069] task:secagentd state:D stack:0 pid:1792 ppid:1711 flags:0x00004006
> [ 56.285084] Call Trace:
> [ 56.285091] <TASK>
> [ 56.285111] schedule+0x612/0x2230
> [ 56.285136] fuse_get_req+0x108/0x2d0
> [ 56.285179] fuse_simple_request+0x40/0x630
> [ 56.285203] fuse_getxattr+0x15d/0x1c0
> [...]
>
> Which looks like wait_event_killable_exclusive() in fuse_get_req().
> And we were wondering if we could do something about it. For example,
> perhaps, something like:
>
> ---
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index f182a4ca1bb32..587cea3a0407d 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -241,7 +241,7 @@ static struct fuse_req *fuse_get_req(struct fuse_conn *fc, bool for_background)
>
> if (fuse_block_alloc(fc, for_background)) {
> err = -EINTR;
> - if (wait_event_killable_exclusive(fc->blocked_waitq,
> + if (wait_event_freezable_killable_exclusive(fc->blocked_waitq,
> !fuse_block_alloc(fc, for_background)))
> goto out;
> }
This looks fine. We can turn each wait into a freezable one inside
fuse. But that still would leave core locks (inode lock, rename lock,
page lock, etc) unfreezable. Turning those into freezable isn't
realistic...
But a partial solution might still be better than no solution.
Thanks,
Miklos
