From: Bernd Schubert <bschubert@ddn.com> mainline inclusion from mainline-v6.10-rc2 commit b6236c8407cba5d7a108facb1bcfab24994d3814 category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICJPON Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i... -------------------------------- 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@ddn.com> Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring Reviewed-by: Luis Henriques <luis@igalia.com> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com> Signed-off-by: Yifan Qiao <qiaoyifan4@huawei.com> Signed-off-by: Long Li <leo.lilong@huawei.com> --- fs/fuse/dev_uring.c | 69 +++++++++++++++++++++++++++++++++++++++++-- fs/fuse/dev_uring_i.h | 9 ++++++ 2 files changed, 75 insertions(+), 3 deletions(-) diff --git a/fs/fuse/dev_uring.c b/fs/fuse/dev_uring.c index 7bd33cef7203..d12d3b7be29a 100644 --- a/fs/fuse/dev_uring.c +++ b/fs/fuse/dev_uring.c @@ -151,6 +151,7 @@ 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; @@ -160,6 +161,12 @@ 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; @@ -243,6 +250,7 @@ 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); @@ -290,6 +298,15 @@ static void fuse_uring_entry_teardown(struct fuse_ring_ent *ent) /* remove entry from queue->fpq->processing */ list_del_init(&req->list); } + + /* + * 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 + */ + list_move(&ent->list, &queue->ent_released); + ent->state = FRRS_RELEASED; spin_unlock(&queue->lock); if (cmd) @@ -297,9 +314,6 @@ static void fuse_uring_entry_teardown(struct fuse_ring_ent *ent) if (req) fuse_uring_stop_fuse_req_end(req); - - list_del_init(&ent->list); - kfree(ent); } static void fuse_uring_stop_list_entries(struct list_head *head, @@ -319,6 +333,7 @@ static void fuse_uring_stop_list_entries(struct list_head *head, continue; } + ent->state = FRRS_TEARDOWN; list_move(&ent->list, &to_teardown); } spin_unlock(&queue->lock); @@ -433,6 +448,46 @@ 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 void fuse_uring_cancel(struct io_uring_cmd *cmd, + unsigned int issue_flags) +{ + struct fuse_ring_ent *ent = uring_cmd_to_ring_ent(cmd); + struct fuse_ring_queue *queue; + bool need_cmd_done = false; + + /* + * 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_AVAILABLE) { + ent->state = FRRS_USERSPACE; + list_move(&ent->list, &queue->ent_in_userspace); + need_cmd_done = true; + ent->cmd = NULL; + } + spin_unlock(&queue->lock); + + if (need_cmd_done) { + /* no queue lock to avoid lock order issues */ + io_uring_cmd_done(cmd, -ENOTCONN, 0, issue_flags); + } +} + +static void fuse_uring_prepare_cancel(struct io_uring_cmd *cmd, int issue_flags, + struct fuse_ring_ent *ring_ent) +{ + 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 */ @@ -840,6 +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(cmd, issue_flags, ent); fuse_uring_commit(ent, req, issue_flags); /* @@ -889,6 +945,8 @@ static void fuse_uring_do_register(struct fuse_ring_ent *ent, struct fuse_conn *fc = ring->fc; struct fuse_iqueue *fiq = &fc->iq; + fuse_uring_prepare_cancel(cmd, issue_flags, ent); + spin_lock(&queue->lock); ent->cmd = cmd; fuse_uring_ent_avail(ent, queue); @@ -1039,6 +1097,11 @@ int __maybe_unused fuse_uring_cmd(struct io_uring_cmd *cmd, return -EOPNOTSUPP; } + if ((unlikely(issue_flags & IO_URING_F_CANCEL))) { + fuse_uring_cancel(cmd, issue_flags); + return 0; + } + /* This extra SQE size holds struct fuse_uring_cmd_req */ if (!(issue_flags & IO_URING_F_SQE128)) return -EINVAL; diff --git a/fs/fuse/dev_uring_i.h b/fs/fuse/dev_uring_i.h index 0182be61778b..2102b3d0c1ae 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 */ @@ -79,6 +85,9 @@ 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; -- 2.39.2