kill-the-BKL/reiserfs: lock only once in reiserfs_truncate_file
authorFrederic Weisbecker <fweisbec@gmail.com>
Tue, 14 Apr 2009 03:34:24 +0000 (05:34 +0200)
committerFrederic Weisbecker <fweisbec@gmail.com>
Mon, 14 Sep 2009 05:18:03 +0000 (07:18 +0200)
Impact: fix a deadlock

reiserfs_truncate_file() can be called from multiple context where
the write lock can be already hold or not.

This function also acquire (possibly recursively) the write
lock. Subsequent releases before sleeping will not actually release
the lock because we may be in more than one lock depth degree.

A typical case is:

reiserfs_file_release {
acquire_the_lock()
reiserfs_truncate_file()
reacquire_the_lock()
journal_begin() {
do_journal_begin_r() {
reiserfs_wait_on_write_block() {
/*
 * Not released because still one
 * depth owned
 */
release_lock()
wait_for_event()

At this stage the event never happen because the one which provides
it needs the write lock.

We use reiserfs_write_lock_once() here to ensure that we don't acquire the
write lock recursively.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alessio Igor Bogani <abogani@texware.it>
Cc: Jeff Mahoney <jeffm@suse.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Cc: Chris Mason <chris.mason@oracle.com>
LKML-Reference: <1239680065-25013-3-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
fs/reiserfs/inode.c

index 1893c8198439efd4c80314c0f2fa307d91186fc1..cc70b56bf6f27d573d5a72f884e6c6256af6fd05 100644 (file)
@@ -2079,8 +2079,9 @@ int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
        int error;
        struct buffer_head *bh = NULL;
        int err2;
+       int lock_depth;
 
-       reiserfs_write_lock(inode->i_sb);
+       lock_depth = reiserfs_write_lock_once(inode->i_sb);
 
        if (inode->i_size > 0) {
                error = grab_tail_page(inode, &page, &bh);
@@ -2149,14 +2150,17 @@ int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
                page_cache_release(page);
        }
 
-       reiserfs_write_unlock(inode->i_sb);
+       reiserfs_write_unlock_once(inode->i_sb, lock_depth);
+
        return 0;
       out:
        if (page) {
                unlock_page(page);
                page_cache_release(page);
        }
-       reiserfs_write_unlock(inode->i_sb);
+
+       reiserfs_write_unlock_once(inode->i_sb, lock_depth);
+
        return error;
 }