From: Eric Biggers <ebiggers@xxxxxxxxxx>
When IOCB_CMD_POLL is used on a userfaultfd, aio_poll() disables IRQs
and takes kioctx::ctx_lock, then userfaultfd_ctx::fd_wqh.lock. This may
have to wait for userfaultfd_ctx::fd_wqh.lock to be released by
userfaultfd_ctx_read(), which can be waiting for
userfaultfd_ctx::fault_pending_wqh.lock or
userfaultfd_ctx::event_wqh.lock. But elsewhere the fault_pending_wqh
and event_wqh locks are taken with IRQs enabled. Since the IRQ handler
may take kioctx::ctx_lock, lockdep reports that a deadlock is possible.
Fix it by always disabling IRQs when taking the fault_pending_wqh and
event_wqh locks.
Commit ae62c16e105a ("userfaultfd: disable irqs when taking the
waitqueue lock") didn't fix this because it only accounted for the
fd_wqh lock, not the other locks nested inside it.
Reported-by: syzbot+fab6de82892b6b9c6191@xxxxxxxxxxxxxxxxxxxxxxxxx
Reported-by: syzbot+53c0b767f7ca0dc0c451@xxxxxxxxxxxxxxxxxxxxxxxxx
Reported-by: syzbot+a3accb352f9c22041cfa@xxxxxxxxxxxxxxxxxxxxxxxxx
Fixes: bfe4037e722e ("aio: implement IOCB_CMD_POLL")
Cc: <stable@xxxxxxxxxxxxxxx> # v4.19+
Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx>
---
fs/userfaultfd.c | 42 ++++++++++++++++++++++++++----------------
1 file changed, 26 insertions(+), 16 deletions(-)
diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index ae0b8b5f69e6..ccbdbd62f0d8 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -40,6 +40,16 @@ enum userfaultfd_state {
/*
* Start with fault_pending_wqh and fault_wqh so they're more likely
* to be in the same cacheline.
+ *
+ * Locking order:
+ * fd_wqh.lock
+ * fault_pending_wqh.lock
+ * fault_wqh.lock
+ * event_wqh.lock
+ *
+ * To avoid deadlocks, IRQs must be disabled when taking any of the above locks,
+ * since fd_wqh.lock is taken by aio_poll() while it's holding a lock that's
+ * also taken in IRQ context.
*/
struct userfaultfd_ctx {
/* waitqueue head for the pending (i.e. not read) userfaults */
@@ -458,7 +468,7 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
blocking_state = return_to_userland ? TASK_INTERRUPTIBLE :
TASK_KILLABLE;
- spin_lock(&ctx->fault_pending_wqh.lock);
+ spin_lock_irq(&ctx->fault_pending_wqh.lock);
/*
* After the __add_wait_queue the uwq is visible to userland
* through poll/read().
@@ -470,7 +480,7 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
* __add_wait_queue.
*/
set_current_state(blocking_state);
- spin_unlock(&ctx->fault_pending_wqh.lock);
+ spin_unlock_irq(&ctx->fault_pending_wqh.lock);
if (!is_vm_hugetlb_page(vmf->vma))
must_wait = userfaultfd_must_wait(ctx, vmf->address, vmf->flags,
@@ -552,13 +562,13 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
* kernel stack can be released after the list_del_init.
*/
if (!list_empty_careful(&uwq.wq.entry)) {
- spin_lock(&ctx->fault_pending_wqh.lock);
+ spin_lock_irq(&ctx->fault_pending_wqh.lock);
/*
* No need of list_del_init(), the uwq on the stack
* will be freed shortly anyway.
*/
list_del(&uwq.wq.entry);
- spin_unlock(&ctx->fault_pending_wqh.lock);
+ spin_unlock_irq(&ctx->fault_pending_wqh.lock);
}
/*
@@ -583,7 +593,7 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
init_waitqueue_entry(&ewq->wq, current);
release_new_ctx = NULL;
- spin_lock(&ctx->event_wqh.lock);
+ spin_lock_irq(&ctx->event_wqh.lock);
/*
* After the __add_wait_queue the uwq is visible to userland
* through poll/read().
@@ -613,15 +623,15 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
break;
}
- spin_unlock(&ctx->event_wqh.lock);
+ spin_unlock_irq(&ctx->event_wqh.lock);
wake_up_poll(&ctx->fd_wqh, EPOLLIN);
schedule();
- spin_lock(&ctx->event_wqh.lock);
+ spin_lock_irq(&ctx->event_wqh.lock);
}
__set_current_state(TASK_RUNNING);
- spin_unlock(&ctx->event_wqh.lock);
+ spin_unlock_irq(&ctx->event_wqh.lock);
if (release_new_ctx) {
struct vm_area_struct *vma;
@@ -918,10 +928,10 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
* the last page faults that may have been already waiting on
* the fault_*wqh.
*/
- spin_lock(&ctx->fault_pending_wqh.lock);
+ spin_lock_irq(&ctx->fault_pending_wqh.lock);
__wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL, &range);
__wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, &range);
- spin_unlock(&ctx->fault_pending_wqh.lock);
+ spin_unlock_irq(&ctx->fault_pending_wqh.lock);
/* Flush pending events that may still wait on event_wqh */
wake_up_all(&ctx->event_wqh);
@@ -1134,7 +1144,7 @@ static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait,
if (!ret && msg->event == UFFD_EVENT_FORK) {
ret = resolve_userfault_fork(ctx, fork_nctx, msg);
- spin_lock(&ctx->event_wqh.lock);
+ spin_lock_irq(&ctx->event_wqh.lock);
if (!list_empty(&fork_event)) {
/*
* The fork thread didn't abort, so we can
@@ -1180,7 +1190,7 @@ static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait,
if (ret)
userfaultfd_ctx_put(fork_nctx);
}
- spin_unlock(&ctx->event_wqh.lock);
+ spin_unlock_irq(&ctx->event_wqh.lock);
}
return ret;
@@ -1219,14 +1229,14 @@ static ssize_t userfaultfd_read(struct file *file, char __user *buf,
static void __wake_userfault(struct userfaultfd_ctx *ctx,
struct userfaultfd_wake_range *range)
{
- spin_lock(&ctx->fault_pending_wqh.lock);
+ spin_lock_irq(&ctx->fault_pending_wqh.lock);
/* wake all in the range and autoremove */
if (waitqueue_active(&ctx->fault_pending_wqh))
__wake_up_locked_key(&ctx->fault_pending_wqh, TASK_NORMAL,
range);
if (waitqueue_active(&ctx->fault_wqh))
__wake_up(&ctx->fault_wqh, TASK_NORMAL, 1, range);
- spin_unlock(&ctx->fault_pending_wqh.lock);
+ spin_unlock_irq(&ctx->fault_pending_wqh.lock);
}
static __always_inline void wake_userfault(struct userfaultfd_ctx *ctx,
@@ -1881,7 +1891,7 @@ static void userfaultfd_show_fdinfo(struct seq_file *m, struct file *f)
wait_queue_entry_t *wq;
unsigned long pending = 0, total = 0;
- spin_lock(&ctx->fault_pending_wqh.lock);
+ spin_lock_irq(&ctx->fault_pending_wqh.lock);
list_for_each_entry(wq, &ctx->fault_pending_wqh.head, entry) {
pending++;
total++;
@@ -1889,7 +1899,7 @@ static void userfaultfd_show_fdinfo(struct seq_file *m, struct file *f)
list_for_each_entry(wq, &ctx->fault_wqh.head, entry) {
total++;
}
- spin_unlock(&ctx->fault_pending_wqh.lock);
+ spin_unlock_irq(&ctx->fault_pending_wqh.lock);
/*
* If more protocols will be added, there will be all shown
--
2.22.0
