Re: kernel BUG at fs/inode.c:LINE! (2)

[Al Viro]

On Fri, Aug 28, 2020 at 06:54:13PM +0100, Al Viro wrote:
> On Fri, Aug 28, 2020 at 04:38:25PM +0100, Al Viro wrote:
> > On Fri, Aug 28, 2020 at 06:18:17AM -0700, syzbot wrote:
> > > Hello,
> > > 
> > > syzbot found the following issue on:
> > > 
> > > HEAD commit:    d012a719 Linux 5.9-rc2
> > > git tree:       upstream
> > > console output: https://syzkaller.appspot.com/x/log.txt?x=15aa650e900000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=891ca5711a9f1650
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=c92c93d1f1aaaacdb9db
> > > compiler:       clang version 10.0.0 (https://github.com/llvm/llvm-project/ c2443155a0fb245c8f17f2c1c72b6ea391e86e81)
> > > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=12ecb939900000
> > > C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=140a19a9900000
> > 
> > Trying to reproduce...
> 
> OK, I think I see what's going on.  ep_loop_check_proc() runs into an already
> doomed file that has already committed to getting killed (->f_count is already
> at 0), but still hadn't gotten through its epitems removal (e.g. has its
> eventpoll_release_file() sitting there trying to get epmutex).
> 
> Blindly bumping refcount here is worse than useless.  Try this, to verify that
> this is what's going on; it's _not_ a proper fix, but it should at least tell
> if we have something else going on.

... and what I think is the right way to fix the original race is (on top of
mainline) this:

[PATCH] Use list_empty_careful() in eventpoll_release()

... to avoid races with list_del_init() in clear_tfile_check_list().
Get rid of pinning files on check list in eventpoll.c - it's not needed
there.

Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
---
diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index e0decff22ae2..39eae45bff18 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -1995,7 +1995,6 @@ static int ep_loop_check_proc(void *priv, void *cookie, int call_nests)
 			 * during ep_insert().
 			 */
 			if (list_empty(&epi->ffd.file->f_tfile_llink)) {
-				get_file(epi->ffd.file);
 				list_add(&epi->ffd.file->f_tfile_llink,
 					 &tfile_check_list);
 			}
@@ -2042,7 +2041,6 @@ static void clear_tfile_check_list(void)
 		file = list_first_entry(&tfile_check_list, struct file,
 					f_tfile_llink);
 		list_del_init(&file->f_tfile_llink);
-		fput(file);
 	}
 	INIT_LIST_HEAD(&tfile_check_list);
 }
@@ -2206,7 +2204,6 @@ int do_epoll_ctl(int epfd, int op, int fd, struct epoll_event *epds,
 				if (ep_loop_check(ep, tf.file) != 0)
 					goto error_tgt_fput;
 			} else {
-				get_file(tf.file);
 				list_add(&tf.file->f_tfile_llink,
 							&tfile_check_list);
 			}
diff --git a/include/linux/eventpoll.h b/include/linux/eventpoll.h
index 8f000fada5a4..e2bdefd90cf8 100644
--- a/include/linux/eventpoll.h
+++ b/include/linux/eventpoll.h
@@ -46,11 +46,9 @@ static inline void eventpoll_release(struct file *file)
 	 * Fast check to avoid the get/release of the semaphore. Since
 	 * we're doing this outside the semaphore lock, it might return
 	 * false negatives, but we don't care. It'll help in 99.99% of cases
-	 * to avoid the semaphore lock. False positives simply cannot happen
-	 * because the file in on the way to be removed and nobody ( but
-	 * eventpoll ) has still a reference to this file.
+	 * to avoid the semaphore lock.
 	 */
-	if (likely(list_empty(&file->f_ep_links)))
+	if (likely(list_empty_careful(&file->f_ep_links)))
 		return;
 
 	/*