From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755542Ab2F0BvS (ORCPT ); Tue, 26 Jun 2012 21:51:18 -0400 Received: from mail-qa0-f46.google.com ([209.85.216.46]:46543 "EHLO mail-qa0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753821Ab2F0BvR (ORCPT ); Tue, 26 Jun 2012 21:51:17 -0400 From: Filipe Brandenburger To: "J. Bruce Fields" , Al Viro , Matthew Wilcox Cc: linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, Filipe Brandenburger Subject: [PATCH v2 1/1] locks: prevent side-effects of locks_release_private before file_lock is initialized Date: Tue, 26 Jun 2012 21:50:48 -0400 Message-Id: <1340761848-1352-1-git-send-email-filbranden@gmail.com> X-Mailer: git-send-email 1.7.7.6 In-Reply-To: <1339815965-1171-1-git-send-email-filbranden@gmail.com> References: <1339815965-1171-1-git-send-email-filbranden@gmail.com> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org When calling fcntl(F_SETLEASE) for a second time on the same file descriptor, do_fcntl_add_lease will allocate and initialize a new file_lock to pass to __vfs_setlease. If that function decides to reuse the existing file_lock, it will free the newly allocated one to prevent leaking memory. However, the newly allocate file_lock was initialized with a valid file descriptor pointer and fl_lmops that contains a lm_release_private function, so calling locks_free_lock will trigger a call to lease_release_private_callback which will clear the fcntl(F_SETOWN) and fcntl(F_SETSIG) settings for the file descriptor even though the lease is not really being cleared at that point (as only the temporary file_lock is being freed.) This patch will fix this bug by calling kmem_cache_free directly instead of locks_free_lock if the file_lock will not be used. This will end up avoiding the call to lease_release_private_callback. This bug was tracked in kernel.org Bugzilla database: https://bugzilla.kernel.org/show_bug.cgi?id=43336 Signed-off-by: Filipe Brandenburger --- fs/locks.c | 22 +++++++++++++++++----- 1 files changed, 17 insertions(+), 5 deletions(-) diff --git a/fs/locks.c b/fs/locks.c index 814c51d..2a751d8 100644 --- a/fs/locks.c +++ b/fs/locks.c @@ -473,7 +473,10 @@ static struct file_lock *lease_alloc(struct file *filp, int type) error = lease_init(filp, type, fl); if (error) { - locks_free_lock(fl); + /* Free the lock by hand instead of calling locks_free_lock + * to prevent the call back to lease_release_private_callback + */ + kmem_cache_free(filelock_cache, fl); return ERR_PTR(error); } return fl; @@ -1538,19 +1541,28 @@ static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg) new = fasync_alloc(); if (!new) { - locks_free_lock(fl); + /* Free the lock by hand instead of calling locks_free_lock + * to prevent the call back to lease_release_private_callback + */ + kmem_cache_free(filelock_cache, fl); return -ENOMEM; } ret = fl; lock_flocks(); error = __vfs_setlease(filp, arg, &ret); + if (error || ret != fl) + /* + * Free the lock by hand instead of calling locks_free_lock + * to prevent the call back to lease_release_private_callback + * which will unset F_SETOWN and F_SETSIG for the file + * descriptor but that is not wanted as the lease was not + * really in use. + */ + kmem_cache_free(filelock_cache, fl); if (error) { unlock_flocks(); - locks_free_lock(fl); goto out_free_fasync; } - if (ret != fl) - locks_free_lock(fl); /* * fasync_insert_entry() returns the old entry if any. -- 1.7.7.6