When the fuse-server terminates while the fuse-client or kernel
still has queued URING_CMDs, these commands retain references
to the struct file used by the fuse connection. This prevents
fuse_dev_release() from being invoked, resulting in a hung mount
point.
This patch addresses the issue by making queued URING_CMDs
cancelable, allowing fuse_dev_release() to proceed as expected
and preventing the mount point from hanging.
Signed-off-by: Bernd Schubert <bschubert@xxxxxxx>
---
fs/fuse/dev_uring.c | 87 ++++++++++++++++++++++++++++++++++++++++++---------
fs/fuse/dev_uring_i.h | 12 +++++++
2 files changed, 85 insertions(+), 14 deletions(-)
diff --git a/fs/fuse/dev_uring.c b/fs/fuse/dev_uring.c
index 8bdfb6fcfa51976cd121bee7f2e8dec1ff9aa916..be7eaf7cc569ff77f8ebdff323634b84ea0a3f63 100644
--- a/fs/fuse/dev_uring.c
+++ b/fs/fuse/dev_uring.c
@@ -168,6 +168,7 @@ void fuse_uring_destruct(struct fuse_conn *fc)
for (qid = 0; qid < ring->nr_queues; qid++) {
struct fuse_ring_queue *queue = ring->queues[qid];
+ struct fuse_ring_ent *ent, *next;
if (!queue)
continue;
@@ -177,6 +178,12 @@ void fuse_uring_destruct(struct fuse_conn *fc)
WARN_ON(!list_empty(&queue->ent_commit_queue));
WARN_ON(!list_empty(&queue->ent_in_userspace));
+ list_for_each_entry_safe(ent, next, &queue->ent_released,
+ list) {
+ list_del_init(&ent->list);
+ kfree(ent);
+ }
+
kfree(queue->fpq.processing);
kfree(queue);
ring->queues[qid] = NULL;
@@ -262,6 +269,7 @@ static struct fuse_ring_queue *fuse_uring_create_queue(struct fuse_ring *ring,
INIT_LIST_HEAD(&queue->ent_in_userspace);
INIT_LIST_HEAD(&queue->fuse_req_queue);
INIT_LIST_HEAD(&queue->fuse_req_bg_queue);
+ INIT_LIST_HEAD(&queue->ent_released);
queue->fpq.processing = pq;
fuse_pqueue_init(&queue->fpq);
@@ -294,24 +302,27 @@ static void fuse_uring_stop_fuse_req_end(struct fuse_ring_ent *ent)
/*
* Release a request/entry on connection tear down
*/
-static void fuse_uring_entry_teardown(struct fuse_ring_ent *ent,
- bool need_cmd_done)
+static void fuse_uring_entry_teardown(struct fuse_ring_ent *ent)
{
- /*
- * fuse_request_end() might take other locks like fi->lock and
- * can lead to lock ordering issues
- */
- lockdep_assert_not_held(&ent->queue->lock);
+ struct fuse_ring_queue *queue = ent->queue;
- if (need_cmd_done)
+ if (ent->need_cmd_done)
io_uring_cmd_done(ent->cmd, -ENOTCONN, 0,
IO_URING_F_UNLOCKED);
if (ent->fuse_req)
fuse_uring_stop_fuse_req_end(ent);
- list_del_init(&ent->list);
- kfree(ent);
+ /*
+ * The entry must not be freed immediately, due to access of direct
+ * pointer access of entries through IO_URING_F_CANCEL - there is a risk
+ * of race between daemon termination (which triggers IO_URING_F_CANCEL
+ * and accesses entries without checking the list state first
+ */
+ spin_lock(&queue->lock);
+ list_move(&ent->list, &queue->ent_released);
+ ent->state = FRRS_RELEASED;
+ spin_unlock(&queue->lock);
}
static void fuse_uring_stop_list_entries(struct list_head *head,
@@ -331,15 +342,15 @@ static void fuse_uring_stop_list_entries(struct list_head *head,
continue;
}
+ ent->need_cmd_done = ent->state != FRRS_USERSPACE;
+ ent->state = FRRS_TEARDOWN;
list_move(&ent->list, &to_teardown);
}
spin_unlock(&queue->lock);
/* no queue lock to avoid lock order issues */
list_for_each_entry_safe(ent, next, &to_teardown, list) {
- bool need_cmd_done = ent->state != FRRS_USERSPACE;
-
- fuse_uring_entry_teardown(ent, need_cmd_done);
+ fuse_uring_entry_teardown(ent);
queue_refs = atomic_dec_return(&ring->queue_refs);
WARN_ON_ONCE(queue_refs < 0);
@@ -447,6 +458,49 @@ void fuse_uring_stop_queues(struct fuse_ring *ring)
}
}
+/*
+ * Handle IO_URING_F_CANCEL, typically should come on daemon termination.
+ *
+ * Releasing the last entry should trigger fuse_dev_release() if
+ * the daemon was terminated
+ */
+static int fuse_uring_cancel(struct io_uring_cmd *cmd, unsigned int issue_flags)
+{
+ struct fuse_ring_ent *ent = fuse_uring_cmd_to_ring_ent(cmd);
+ struct fuse_ring_queue *queue;
+ bool need_cmd_done = false;
+ int ret = 0;
+
+ /*
+ * direct access on ent - it must not be destructed as long as
+ * IO_URING_F_CANCEL might come up
+ */
+ queue = ent->queue;
+ spin_lock(&queue->lock);
+ if (ent->state == FRRS_WAIT) {
+ ent->state = FRRS_USERSPACE;
+ list_move(&ent->list, &queue->ent_in_userspace);
+ need_cmd_done = true;
+ }
+ spin_unlock(&queue->lock);
+
+ if (need_cmd_done) {
+ io_uring_cmd_done(cmd, -ENOTCONN, 0, issue_flags);
+ } else {
+ /* io-uring handles resending */
+ ret = -EAGAIN;
+ }
+
+ return ret;
+}
+
+static void fuse_uring_prepare_cancel(struct io_uring_cmd *cmd, int issue_flags,
+ struct fuse_ring_ent *ring_ent)
+{
+ fuse_uring_cmd_set_ring_ent(cmd, ring_ent);
+ io_uring_cmd_mark_cancelable(cmd, issue_flags);
+}
+
/*
* Checks for errors and stores it into the request
*/
@@ -841,6 +895,7 @@ static int fuse_uring_commit_fetch(struct io_uring_cmd *cmd, int issue_flags,
spin_unlock(&queue->lock);
/* without the queue lock, as other locks are taken */
+ fuse_uring_prepare_cancel(ring_ent->cmd, issue_flags, ring_ent);
fuse_uring_commit(ring_ent, issue_flags);
/*
@@ -890,6 +945,8 @@ static void _fuse_uring_register(struct fuse_ring_ent *ring_ent,
struct fuse_conn *fc = ring->fc;
struct fuse_iqueue *fiq = &fc->iq;
+ fuse_uring_prepare_cancel(ring_ent->cmd, issue_flags, ring_ent);
+
spin_lock(&queue->lock);
fuse_uring_ent_avail(ring_ent, queue);
spin_unlock(&queue->lock);
@@ -1039,6 +1096,9 @@ int __maybe_unused fuse_uring_cmd(struct io_uring_cmd *cmd,
return -EOPNOTSUPP;
}
+ if ((unlikely(issue_flags & IO_URING_F_CANCEL)))
+ return fuse_uring_cancel(cmd, issue_flags);
+
/* This extra SQE size holds struct fuse_uring_cmd_req */
if (!(issue_flags & IO_URING_F_SQE128))
return -EINVAL;
@@ -1170,7 +1230,6 @@ void fuse_uring_queue_fuse_req(struct fuse_iqueue *fiq, struct fuse_req *req)
if (ring_ent) {
struct io_uring_cmd *cmd = ring_ent->cmd;
-
err = -EIO;
if (WARN_ON_ONCE(ring_ent->state != FRRS_FUSE_REQ))
goto err;
diff --git a/fs/fuse/dev_uring_i.h b/fs/fuse/dev_uring_i.h
index 8c5e3ac630f245192c380d132b665d95b8f446a4..4e670022ada2827657feec8e5165e56dbfb86037 100644
--- a/fs/fuse/dev_uring_i.h
+++ b/fs/fuse/dev_uring_i.h
@@ -28,6 +28,12 @@ enum fuse_ring_req_state {
/* The ring entry is in or on the way to user space */
FRRS_USERSPACE,
+
+ /* The ring entry is in teardown */
+ FRRS_TEARDOWN,
+
+ /* The ring entry is released, but not freed yet */
+ FRRS_RELEASED,
};
/** A fuse ring entry, part of the ring queue */
@@ -49,6 +55,9 @@ struct fuse_ring_ent {
*/
unsigned int state;
+ /* The entry needs io_uring_cmd_done for teardown */
+ unsigned int need_cmd_done:1;
+
struct fuse_req *fuse_req;
/* commit id to identify the server reply */
@@ -84,6 +93,9 @@ struct fuse_ring_queue {
/* entries in userspace */
struct list_head ent_in_userspace;
+ /* entries that are released */
+ struct list_head ent_released;
+
/* fuse requests waiting for an entry slot */
struct list_head fuse_req_queue;
--
2.43.0
