From: Joanne Koong <joannelkoong@gmail.com>
To: Bernd Schubert <bernd.schubert@fastmail.fm>
Cc: miklos@szeredi.hu, linux-fsdevel@vger.kernel.org,
josef@toxicpanda.com, laoar.shao@gmail.com,
kernel-team@meta.com
Subject: Re: [PATCH v2 1/2] fuse: add optional kernel-enforced timeout for requests
Date: Mon, 5 Aug 2024 15:10:07 -0700 [thread overview]
Message-ID: <CAJnrk1YVC58PiU6_gJno7i439uHUkcLDzKY4mXmupybeDO7LWQ@mail.gmail.com> (raw)
In-Reply-To: <fc1ed986-fcd6-4a52-aed3-f3f61f2513a7@fastmail.fm>
On Mon, Aug 5, 2024 at 6:26 AM Bernd Schubert
<bernd.schubert@fastmail.fm> wrote:
>
>
>
> On 8/5/24 06:52, Joanne Koong wrote:
> > On Mon, Jul 29, 2024 at 5:28 PM Joanne Koong <joannelkoong@gmail.com> wrote:
> >>
> >> There are situations where fuse servers can become unresponsive or take
> >> too long to reply to a request. Currently there is no upper bound on
> >> how long a request may take, which may be frustrating to users who get
> >> stuck waiting for a request to complete.
> >>
> >> This commit adds a timeout option (in seconds) for requests. If the
> >> timeout elapses before the server replies to the request, the request
> >> will fail with -ETIME.
> >>
> >> There are 3 possibilities for a request that times out:
> >> a) The request times out before the request has been sent to userspace
> >> b) The request times out after the request has been sent to userspace
> >> and before it receives a reply from the server
> >> c) The request times out after the request has been sent to userspace
> >> and the server replies while the kernel is timing out the request
> >>
> >> While a request timeout is being handled, there may be other handlers
> >> running at the same time if:
> >> a) the kernel is forwarding the request to the server
> >> b) the kernel is processing the server's reply to the request
> >> c) the request is being re-sent
> >> d) the connection is aborting
> >> e) the device is getting released
> >>
> >> Proper synchronization must be added to ensure that the request is
> >> handled correctly in all of these cases. To this effect, there is a new
> >> FR_FINISHING bit added to the request flags, which is set atomically by
> >> either the timeout handler (see fuse_request_timeout()) which is invoked
> >> after the request timeout elapses or set by the request reply handler
> >> (see dev_do_write()), whichever gets there first. If the reply handler
> >> and the timeout handler are executing simultaneously and the reply handler
> >> sets FR_FINISHING before the timeout handler, then the request will be
> >> handled as if the timeout did not elapse. If the timeout handler sets
> >> FR_FINISHING before the reply handler, then the request will fail with
> >> -ETIME and the request will be cleaned up.
> >>
> >> Currently, this is the refcount lifecycle of a request:
> >>
> >> Synchronous request is created:
> >> fuse_simple_request -> allocates request, sets refcount to 1
> >> __fuse_request_send -> acquires refcount
> >> queues request and waits for reply...
> >> fuse_simple_request -> drops refcount
> >>
> >> Background request is created:
> >> fuse_simple_background -> allocates request, sets refcount to 1
> >>
> >> Request is replied to:
> >> fuse_dev_do_write
> >> fuse_request_end -> drops refcount on request
> >>
> >> Proper acquires on the request reference must be added to ensure that the
> >> timeout handler does not drop the last refcount on the request while
> >> other handlers may be operating on the request. Please note that the
> >> timeout handler may get invoked at any phase of the request's
> >> lifetime (eg before the request has been forwarded to userspace, etc).
> >>
> >> It is always guaranteed that there is a refcount on the request when the
> >> timeout handler is executing. The timeout handler will be either
> >> deactivated by the reply/abort/release handlers, or if the timeout
> >> handler is concurrently executing on another CPU, the reply/abort/release
> >> handlers will wait for the timeout handler to finish executing first before
> >> it drops the final refcount on the request.
> >>
> >> Signed-off-by: Joanne Koong <joannelkoong@gmail.com>
> >> ---
> >> fs/fuse/dev.c | 187 +++++++++++++++++++++++++++++++++++++++++++++--
> >> fs/fuse/fuse_i.h | 14 ++++
> >> fs/fuse/inode.c | 7 ++
> >> 3 files changed, 200 insertions(+), 8 deletions(-)
> >>
> >> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> >> index 9eb191b5c4de..9992bc5f4469 100644
> >> --- a/fs/fuse/dev.c
> >> +++ b/fs/fuse/dev.c
> >> @@ -31,6 +31,8 @@ MODULE_ALIAS("devname:fuse");
> >>
> >> static struct kmem_cache *fuse_req_cachep;
> >>
> >> +static void fuse_request_timeout(struct timer_list *timer);
> >> +
> >> static struct fuse_dev *fuse_get_dev(struct file *file)
> >> {
> >> /*
> >> @@ -48,6 +50,8 @@ static void fuse_request_init(struct fuse_mount *fm, struct fuse_req *req)
> >> refcount_set(&req->count, 1);
> >> __set_bit(FR_PENDING, &req->flags);
> >> req->fm = fm;
> >> + if (fm->fc->req_timeout)
> >> + timer_setup(&req->timer, fuse_request_timeout, 0);
> >> }
> >>
> >> static struct fuse_req *fuse_request_alloc(struct fuse_mount *fm, gfp_t flags)
> >> @@ -277,12 +281,15 @@ static void flush_bg_queue(struct fuse_conn *fc)
> >> * the 'end' callback is called if given, else the reference to the
> >> * request is released
> >> */
> >> -void fuse_request_end(struct fuse_req *req)
> >> +static void do_fuse_request_end(struct fuse_req *req, bool from_timer_callback)
> >> {
> >> struct fuse_mount *fm = req->fm;
> >> struct fuse_conn *fc = fm->fc;
> >> struct fuse_iqueue *fiq = &fc->iq;
> >>
> >> + if (from_timer_callback)
> >> + req->out.h.error = -ETIME;
> >> +
> >> if (test_and_set_bit(FR_FINISHED, &req->flags))
> >> goto put_request;
> >>
> >> @@ -296,8 +303,6 @@ void fuse_request_end(struct fuse_req *req)
> >> list_del_init(&req->intr_entry);
> >> spin_unlock(&fiq->lock);
> >> }
> >> - WARN_ON(test_bit(FR_PENDING, &req->flags));
> >> - WARN_ON(test_bit(FR_SENT, &req->flags));
> >> if (test_bit(FR_BACKGROUND, &req->flags)) {
> >> spin_lock(&fc->bg_lock);
> >> clear_bit(FR_BACKGROUND, &req->flags);
> >> @@ -324,13 +329,105 @@ void fuse_request_end(struct fuse_req *req)
> >> wake_up(&req->waitq);
> >> }
> >>
> >> + if (!from_timer_callback && req->timer.function)
> >> + timer_delete_sync(&req->timer);
> >> +
> >> if (test_bit(FR_ASYNC, &req->flags))
> >> req->args->end(fm, req->args, req->out.h.error);
> >> put_request:
> >> fuse_put_request(req);
> >> }
> >> +
> >> +void fuse_request_end(struct fuse_req *req)
> >> +{
> >> + WARN_ON(test_bit(FR_PENDING, &req->flags));
> >> + WARN_ON(test_bit(FR_SENT, &req->flags));
> >> +
> >> + do_fuse_request_end(req, false);
> >> +}
> >> EXPORT_SYMBOL_GPL(fuse_request_end);
> >>
> >> +static void timeout_inflight_req(struct fuse_req *req)
> >> +{
> >> + struct fuse_conn *fc = req->fm->fc;
> >> + struct fuse_iqueue *fiq = &fc->iq;
> >> + struct fuse_pqueue *fpq;
> >> +
> >> + spin_lock(&fiq->lock);
> >> + fpq = req->fpq;
> >> + spin_unlock(&fiq->lock);
> >> +
> >> + /*
> >> + * If fpq has not been set yet, then the request is aborting (which
> >> + * clears FR_PENDING flag) before dev_do_read (which sets req->fpq)
> >> + * has been called. Let the abort handler handle this request.
> >> + */
> >> + if (!fpq)
> >> + return;
> >> +
> >> + spin_lock(&fpq->lock);
> >> + if (!fpq->connected || req->out.h.error == -ECONNABORTED) {
> >> + /*
> >> + * Connection is being aborted or the fuse_dev is being released.
> >> + * The abort / release will clean up the request
> >> + */
> >> + spin_unlock(&fpq->lock);
> >> + return;
> >> + }
> >> +
> >> + if (!test_bit(FR_PRIVATE, &req->flags))
> >> + list_del_init(&req->list);
> >> +
> >> + spin_unlock(&fpq->lock);
> >> +
> >> + do_fuse_request_end(req, true);
> >> +}
> >> +
> >> +static void timeout_pending_req(struct fuse_req *req)
> >> +{
> >> + struct fuse_conn *fc = req->fm->fc;
> >> + struct fuse_iqueue *fiq = &fc->iq;
> >> + bool background = test_bit(FR_BACKGROUND, &req->flags);
> >> +
> >> + if (background)
> >> + spin_lock(&fc->bg_lock);
> >> + spin_lock(&fiq->lock);
> >> +
> >> + if (!test_bit(FR_PENDING, &req->flags)) {
> >> + spin_unlock(&fiq->lock);
> >> + if (background)
> >> + spin_unlock(&fc->bg_lock);
> >> + timeout_inflight_req(req);
> >> + return;
> >> + }
> >> +
> >> + if (!test_bit(FR_PRIVATE, &req->flags))
> >> + list_del_init(&req->list);
> >> +
> >> + spin_unlock(&fiq->lock);
> >> + if (background)
> >> + spin_unlock(&fc->bg_lock);
> >> +
> >> + do_fuse_request_end(req, true);
> >> +}
> >> +
> >> +static void fuse_request_timeout(struct timer_list *timer)
> >> +{
> >> + struct fuse_req *req = container_of(timer, struct fuse_req, timer);
> >> +
> >> + /*
> >> + * Request reply is being finished by the kernel right now.
> >> + * No need to time out the request.
> >> + */
> >> + if (test_and_set_bit(FR_FINISHING, &req->flags))
> >> + return;
> >> +
> >> + if (test_bit(FR_PENDING, &req->flags))
> >> + timeout_pending_req(req);
> >> + else
> >> + timeout_inflight_req(req);
> >> +}
> >> +
> >> static int queue_interrupt(struct fuse_req *req)
> >> {
> >> struct fuse_iqueue *fiq = &req->fm->fc->iq;
> >> @@ -409,7 +506,8 @@ static void request_wait_answer(struct fuse_req *req)
> >>
> >> static void __fuse_request_send(struct fuse_req *req)
> >> {
> >> - struct fuse_iqueue *fiq = &req->fm->fc->iq;
> >> + struct fuse_conn *fc = req->fm->fc;
> >> + struct fuse_iqueue *fiq = &fc->iq;
> >>
> >> BUG_ON(test_bit(FR_BACKGROUND, &req->flags));
> >> spin_lock(&fiq->lock);
> >> @@ -421,6 +519,10 @@ static void __fuse_request_send(struct fuse_req *req)
> >> /* acquire extra reference, since request is still needed
> >> after fuse_request_end() */
> >> __fuse_get_request(req);
> >> + if (req->timer.function) {
> >> + req->timer.expires = jiffies + fc->req_timeout;
> >> + add_timer(&req->timer);
> >> + }
> >> queue_request_and_unlock(fiq, req);
> >>
> >> request_wait_answer(req);
> >> @@ -539,6 +641,10 @@ static bool fuse_request_queue_background(struct fuse_req *req)
> >> if (fc->num_background == fc->max_background)
> >> fc->blocked = 1;
> >> list_add_tail(&req->list, &fc->bg_queue);
> >> + if (req->timer.function) {
> >> + req->timer.expires = jiffies + fc->req_timeout;
> >> + add_timer(&req->timer);
> >> + }
> >> flush_bg_queue(fc);
> >> queued = true;
> >> }
> >> @@ -1268,6 +1374,9 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
> >> req = list_entry(fiq->pending.next, struct fuse_req, list);
> >> clear_bit(FR_PENDING, &req->flags);
> >> list_del_init(&req->list);
> >> + /* Acquire a reference in case the timeout handler starts executing */
> >> + __fuse_get_request(req);
> >> + req->fpq = fpq;
> >> spin_unlock(&fiq->lock);
> >>
> >> args = req->args;
> >> @@ -1280,6 +1389,7 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
> >> if (args->opcode == FUSE_SETXATTR)
> >> req->out.h.error = -E2BIG;
> >> fuse_request_end(req);
> >> + fuse_put_request(req);
> >> goto restart;
> >
> > While rereading through fuse_dev_do_read, I just realized we also need
> > to handle the race condition for the error edge cases (here and in the
> > "goto out_end;"), since the timeout handler could have finished
> > executing by the time we hit the error edge case. We need to
> > test_and_set_bit(FR_FINISHING) so that either the timeout_handler or
> > dev_do_read cleans up the request, but not both. I'll fix this for v3.
>
> I know it would change semantics a bit, but wouldn't it be much easier /
> less racy if fuse_dev_do_read() would delete the timer when it takes a
> request from fiq->pending and add it back in (with new timeouts) before
> it returns the request?
>
Ooo I really like this idea! I'm worried though that this might allow
potential scenarios where the fuse_dev_do_read gets descheduled after
disarming the timer and a non-trivial amount of time elapses before it
gets scheduled back (eg on a system where the CPU is starved), in
which case the fuse req_timeout value will be (somewhat of) a lie. If
you and others think this is likely fine though, then I'll incorporate
this into v3 which will make this logic a lot simpler :)
Thanks,
Joanne
> Untested:
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 9992bc5f4469..444f667e2f43 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -1379,6 +1379,15 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
> req->fpq = fpq;
> spin_unlock(&fiq->lock);
>
> + if (req->timer.function) {
> + /* request gets handled, remove the previous timeout */
> + timer_delete_sync(&req->timer);
> + if (test_bit(FR_FINISHED, &req->flags)) {
> + fuse_put_request(req);
> + goto restart;
> + }
> + }
> +
> args = req->args;
> reqsize = req->in.h.len;
>
> @@ -1433,24 +1442,10 @@ static ssize_t fuse_dev_do_read(struct fuse_dev *fud, struct file *file,
> if (test_bit(FR_INTERRUPTED, &req->flags))
> queue_interrupt(req);
>
> - /*
> - * Check if the timeout handler is running / ran. If it did, we need to
> - * remove the request from any lists in case the timeout handler finished
> - * before dev_do_read moved the request to the processing list.
> - *
> - * Check FR_SENT to distinguish whether the timeout or the write handler
> - * is finishing the request. However, there can be the case where the
> - * timeout handler and resend handler are running concurrently, so we
> - * need to also check the FR_PENDING bit.
> - */
> - if (test_bit(FR_FINISHING, &req->flags) &&
> - (test_bit(FR_SENT, &req->flags) || test_bit(FR_PENDING, &req->flags))) {
> - spin_lock(&fpq->lock);
> - if (!test_bit(FR_PRIVATE, &req->flags))
> - list_del_init(&req->list);
> - spin_unlock(&fpq->lock);
> - fuse_put_request(req);
> - return -ETIME;
> + if (req->timer.function) {
> + /* re-arm the request */
> + req->timer.expires = jiffies + fc->req_timeout;
> + add_timer(&req->timer);
> }
>
> fuse_put_request(req);
>
> Thanks,
> Bernd
next prev parent reply other threads:[~2024-08-05 22:10 UTC|newest]
Thread overview: 34+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-07-30 0:23 [PATCH v2 0/2] fuse: add timeout option for requests Joanne Koong
2024-07-30 0:23 ` [PATCH v2 1/2] fuse: add optional kernel-enforced timeout " Joanne Koong
2024-08-04 22:46 ` Bernd Schubert
2024-08-05 4:45 ` Joanne Koong
2024-08-05 13:05 ` Bernd Schubert
2024-08-05 4:52 ` Joanne Koong
2024-08-05 13:26 ` Bernd Schubert
2024-08-05 22:10 ` Joanne Koong [this message]
2024-08-06 15:43 ` Bernd Schubert
2024-08-06 17:08 ` Joanne Koong
2024-08-05 7:32 ` Jingbo Xu
2024-08-05 22:53 ` Joanne Koong
2024-08-06 2:45 ` Jingbo Xu
2024-08-06 16:43 ` Joanne Koong
2024-08-06 15:50 ` Bernd Schubert
2024-07-30 0:23 ` [PATCH v2 2/2] fuse: add default_request_timeout and max_request_timeout sysctls Joanne Koong
2024-07-30 7:49 ` kernel test robot
2024-07-30 9:14 ` kernel test robot
2024-08-05 7:38 ` Jingbo Xu
2024-08-06 1:26 ` Joanne Koong
2024-07-30 5:59 ` [PATCH v2 0/2] fuse: add timeout option for requests Yafang Shao
2024-07-30 18:16 ` Joanne Koong
2024-07-31 2:13 ` Yafang Shao
2024-07-31 17:52 ` Joanne Koong
2024-07-31 18:46 ` Joanne Koong
2024-08-01 2:47 ` Yafang Shao
2024-08-02 19:05 ` Joanne Koong
2024-08-04 7:46 ` Yafang Shao
2024-08-05 5:05 ` Joanne Koong
2024-08-06 16:23 ` Joanne Koong
2024-08-06 17:11 ` Bernd Schubert
2024-08-06 18:26 ` Joanne Koong
2024-08-06 18:37 ` Joanne Koong
2024-08-06 20:08 ` Bernd Schubert
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAJnrk1YVC58PiU6_gJno7i439uHUkcLDzKY4mXmupybeDO7LWQ@mail.gmail.com \
--to=joannelkoong@gmail.com \
--cc=bernd.schubert@fastmail.fm \
--cc=josef@toxicpanda.com \
--cc=kernel-team@meta.com \
--cc=laoar.shao@gmail.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=miklos@szeredi.hu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).