epoll: avoid double-inserts in case of EFAULT

In commit f337b9c58332bdecde965b436e47ea4c94d30da0 ("epoll: drop
unnecessary test") Thomas found that there is an unnecessary (always
true) test in ep_send_events().  The callback never inserts into
->rdllink while the send loop is performed, and also does the
~EP_PRIVATE_BITS test.  Given we're holding the mutex during this time,
the conditions tested inside the loop are always true.

HOWEVER.

The test "!ep_is_linked(&epi->rdllink)" wasn't there because we insert
into ->rdllink, but because the send-events loop might terminate before
the whole list is scanned (-EFAULT).

In such cases, when the loop terminates early, and when a (leftover)
file received an event while we're performing the lockless loop, we need
such test to avoid to double insert the epoll items.  The list_splice()
done a few steps below, will correctly re-insert the ones that were left
on "txlist".

This should fix the kenrel.org bugzilla entry 11831.

Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index 99368bda0261321290a6afa52ad74a797f79cd54..aec5c13f634177bab34f108e9f4891e300b873a3 100644 (file)
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -930,8 +930,15 @@ errxit:
         * inside the main ready-list here.
         */
        for (nepi = ep->ovflist; (epi = nepi) != NULL;
-            nepi = epi->next, epi->next = EP_UNACTIVE_PTR)
-               list_add_tail(&epi->rdllink, &ep->rdllist);
+            nepi = epi->next, epi->next = EP_UNACTIVE_PTR) {
+               /*
+                * If the above loop quit with errors, the epoll item might still
+                * be linked to "txlist", and the list_splice() done below will
+                * take care of those cases.
+                */
+               if (!ep_is_linked(&epi->rdllink))
+                       list_add_tail(&epi->rdllink, &ep->rdllist);
+       }
        /*
         * We need to set back ep->ovflist to EP_UNACTIVE_PTR, so that after
         * releasing the lock, events will be queued in the normal way inside