]> git.itanic.dy.fi Git - linux-stable/commitdiff
io_uring: fix link traversal locking
authorPavel Begunkov <asml.silence@gmail.com>
Fri, 26 Nov 2021 14:38:15 +0000 (14:38 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 1 Dec 2021 08:04:42 +0000 (09:04 +0100)
commit 6af3f48bf6156a7f02e91aca64e2927c4bebda03 upstream.

WARNING: inconsistent lock state
5.16.0-rc2-syzkaller #0 Not tainted
inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
ffff888078e11418 (&ctx->timeout_lock
){?.+.}-{2:2}
, at: io_timeout_fn+0x6f/0x360 fs/io_uring.c:5943
{HARDIRQ-ON-W} state was registered at:
  [...]
  spin_unlock_irq include/linux/spinlock.h:399 [inline]
  __io_poll_remove_one fs/io_uring.c:5669 [inline]
  __io_poll_remove_one fs/io_uring.c:5654 [inline]
  io_poll_remove_one+0x236/0x870 fs/io_uring.c:5680
  io_poll_remove_all+0x1af/0x235 fs/io_uring.c:5709
  io_ring_ctx_wait_and_kill+0x1cc/0x322 fs/io_uring.c:9534
  io_uring_release+0x42/0x46 fs/io_uring.c:9554
  __fput+0x286/0x9f0 fs/file_table.c:280
  task_work_run+0xdd/0x1a0 kernel/task_work.c:164
  exit_task_work include/linux/task_work.h:32 [inline]
  do_exit+0xc14/0x2b40 kernel/exit.c:832

674ee8e1b4a41 ("io_uring: correct link-list traversal locking") fixed a
data race but introduced a possible deadlock and inconsistentcy in irq
states. E.g.

io_poll_remove_all()
    spin_lock_irq(timeout_lock)
    io_poll_remove_one()
        spin_lock/unlock_irq(poll_lock);
    spin_unlock_irq(timeout_lock)

Another type of problem is freeing a request while holding
->timeout_lock, which may leads to a deadlock in
io_commit_cqring() -> io_flush_timeouts() and other places.

Having 3 nested locks is also too ugly. Add io_match_task_safe(), which
would briefly take and release timeout_lock for race prevention inside,
so the actuall request cancellation / free / etc. code doesn't have it
taken.

Reported-by: syzbot+ff49a3059d49b0ca0eec@syzkaller.appspotmail.com
Reported-by: syzbot+847f02ec20a6609a328b@syzkaller.appspotmail.com
Reported-by: syzbot+3368aadcd30425ceb53b@syzkaller.appspotmail.com
Reported-by: syzbot+51ce8887cdef77c9ac83@syzkaller.appspotmail.com
Reported-by: syzbot+3cb756a49d2f394a9ee3@syzkaller.appspotmail.com
Fixes: 674ee8e1b4a41 ("io_uring: correct link-list traversal locking")
Cc: stable@kernel.org # 5.15+
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/397f7ebf3f4171f1abe41f708ac1ecb5766f0b68.1637937097.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/io_uring.c

index edd30baa568db14146a0f9d737e184229af68f62..074718333b1053d5ea05cd803d7849f7f7f7474d 100644 (file)
@@ -1204,6 +1204,7 @@ static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl)
 
 static bool io_match_task(struct io_kiocb *head, struct task_struct *task,
                          bool cancel_all)
+       __must_hold(&req->ctx->timeout_lock)
 {
        struct io_kiocb *req;
 
@@ -1219,6 +1220,44 @@ static bool io_match_task(struct io_kiocb *head, struct task_struct *task,
        return false;
 }
 
+static bool io_match_linked(struct io_kiocb *head)
+{
+       struct io_kiocb *req;
+
+       io_for_each_link(req, head) {
+               if (req->flags & REQ_F_INFLIGHT)
+                       return true;
+       }
+       return false;
+}
+
+/*
+ * As io_match_task() but protected against racing with linked timeouts.
+ * User must not hold timeout_lock.
+ */
+static bool io_match_task_safe(struct io_kiocb *head, struct task_struct *task,
+                              bool cancel_all)
+{
+       bool matched;
+
+       if (task && head->task != task)
+               return false;
+       if (cancel_all)
+               return true;
+
+       if (head->flags & REQ_F_LINK_TIMEOUT) {
+               struct io_ring_ctx *ctx = head->ctx;
+
+               /* protect against races with linked timeouts */
+               spin_lock_irq(&ctx->timeout_lock);
+               matched = io_match_linked(head);
+               spin_unlock_irq(&ctx->timeout_lock);
+       } else {
+               matched = io_match_linked(head);
+       }
+       return matched;
+}
+
 static inline void req_set_fail(struct io_kiocb *req)
 {
        req->flags |= REQ_F_FAIL;
@@ -5697,17 +5736,15 @@ static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
        int posted = 0, i;
 
        spin_lock(&ctx->completion_lock);
-       spin_lock_irq(&ctx->timeout_lock);
        for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
                struct hlist_head *list;
 
                list = &ctx->cancel_hash[i];
                hlist_for_each_entry_safe(req, tmp, list, hash_node) {
-                       if (io_match_task(req, tsk, cancel_all))
+                       if (io_match_task_safe(req, tsk, cancel_all))
                                posted += io_poll_remove_one(req);
                }
        }
-       spin_unlock_irq(&ctx->timeout_lock);
        spin_unlock(&ctx->completion_lock);
 
        if (posted)
@@ -9520,19 +9557,8 @@ static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
 {
        struct io_kiocb *req = container_of(work, struct io_kiocb, work);
        struct io_task_cancel *cancel = data;
-       bool ret;
 
-       if (!cancel->all && (req->flags & REQ_F_LINK_TIMEOUT)) {
-               struct io_ring_ctx *ctx = req->ctx;
-
-               /* protect against races with linked timeouts */
-               spin_lock_irq(&ctx->timeout_lock);
-               ret = io_match_task(req, cancel->task, cancel->all);
-               spin_unlock_irq(&ctx->timeout_lock);
-       } else {
-               ret = io_match_task(req, cancel->task, cancel->all);
-       }
-       return ret;
+       return io_match_task_safe(req, cancel->task, cancel->all);
 }
 
 static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
@@ -9542,14 +9568,12 @@ static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
        LIST_HEAD(list);
 
        spin_lock(&ctx->completion_lock);
-       spin_lock_irq(&ctx->timeout_lock);
        list_for_each_entry_reverse(de, &ctx->defer_list, list) {
-               if (io_match_task(de->req, task, cancel_all)) {
+               if (io_match_task_safe(de->req, task, cancel_all)) {
                        list_cut_position(&list, &ctx->defer_list, &de->list);
                        break;
                }
        }
-       spin_unlock_irq(&ctx->timeout_lock);
        spin_unlock(&ctx->completion_lock);
        if (list_empty(&list))
                return false;