NTFS: Fix a nasty deadlock that appeared in recent kernels.

      The situation: VFS inode X on a mounted ntfs volume is dirty.  For
      same inode X, the ntfs_inode is dirty and thus corresponding on-disk
      inode, i.e. mft record, which is in a dirty PAGE_CACHE_PAGE belonging
      to the table of inodes, i.e. $MFT, inode 0.
      What happens:
      Process 1: sys_sync()/umount()/whatever...  calls
      __sync_single_inode() for $MFT -> do_writepages() -> write_page for
      the dirty page containing the on-disk inode X, the page is now locked
      -> ntfs_write_mst_block() which clears PageUptodate() on the page to
      prevent anyone else getting hold of it whilst it does the write out.
      This is necessary as the on-disk inode needs "fixups" applied before
      the write to disk which are removed again after the write and
      PageUptodate is then set again.  It then analyses the page looking
      for dirty on-disk inodes and when it finds one it calls
      ntfs_may_write_mft_record() to see if it is safe to write this
      on-disk inode.  This then calls ilookup5() to check if the
      corresponding VFS inode is in icache().  This in turn calls ifind()
      which waits on the inode lock via wait_on_inode whilst holding the
      global inode_lock.
      Process 2: pdflush results in a call to __sync_single_inode for the
      same VFS inode X on the ntfs volume.  This locks the inode (I_LOCK)
      then calls write-inode -> ntfs_write_inode -> map_mft_record() ->
      read_cache_page() for the page (in page cache of table of inodes
      $MFT, inode 0) containing the on-disk inode.  This page has
      PageUptodate() clear because of Process 1 (see above) so
      read_cache_page() blocks when it tries to take the page lock for the
      page so it can call ntfs_read_page().
      Thus Process 1 is holding the page lock on the page containing the
      on-disk inode X and it is waiting on the inode X to be unlocked in
      ifind() so it can write the page out and then unlock the page.
      And Process 2 is holding the inode lock on inode X and is waiting for
      the page to be unlocked so it can call ntfs_readpage() or discover
      that Process 1 set PageUptodate() again and use the page.
      Thus we have a deadlock due to ifind() waiting on the inode lock.
      The solution: The fix is to use the newly introduced
      ilookup5_nowait() which does not wait on the inode's lock and hence
      avoids the deadlock.  This is safe as we do not care about the VFS
      inode and only use the fact that it is in the VFS inode cache and the
      fact that the vfs and ntfs inodes are one struct in memory to find
      the ntfs inode in memory if present.  Also, the ntfs inode has its
      own locking so it does not matter if the vfs inode is locked.

Signed-off-by: Anton Altaparmakov <aia21@cantab.net>

2 files changed, 61 insertions, 10 deletions

diff --git a/fs/ntfs/ChangeLog b/fs/ntfs/ChangeLog
index 3d2cac4061d6..9709fac6531d 100644
--- a/fs/ntfs/ChangeLog
+++ b/fs/ntfs/ChangeLog
@@ -132,6 +132,48 @@ ToDo/Notes:
 	  the already mapped runlist fragment which causes
 	  ntfs_mapping_pairs_decompress() to fail and return error.  Update
 	  ntfs_attr_find_vcn_nolock() accordingly.
+	- Fix a nasty deadlock that appeared in recent kernels.
+	  The situation: VFS inode X on a mounted ntfs volume is dirty.  For
+	  same inode X, the ntfs_inode is dirty and thus corresponding on-disk
+	  inode, i.e. mft record, which is in a dirty PAGE_CACHE_PAGE belonging
+	  to the table of inodes, i.e. $MFT, inode 0.
+	  What happens:
+	  Process 1: sys_sync()/umount()/whatever...  calls
+	  __sync_single_inode() for $MFT -> do_writepages() -> write_page for
+	  the dirty page containing the on-disk inode X, the page is now locked
+	  -> ntfs_write_mst_block() which clears PageUptodate() on the page to
+	  prevent anyone else getting hold of it whilst it does the write out.
+	  This is necessary as the on-disk inode needs "fixups" applied before
+	  the write to disk which are removed again after the write and
+	  PageUptodate is then set again.  It then analyses the page looking
+	  for dirty on-disk inodes and when it finds one it calls
+	  ntfs_may_write_mft_record() to see if it is safe to write this
+	  on-disk inode.  This then calls ilookup5() to check if the
+	  corresponding VFS inode is in icache().  This in turn calls ifind()
+	  which waits on the inode lock via wait_on_inode whilst holding the
+	  global inode_lock.
+	  Process 2: pdflush results in a call to __sync_single_inode for the
+	  same VFS inode X on the ntfs volume.  This locks the inode (I_LOCK)
+	  then calls write-inode -> ntfs_write_inode -> map_mft_record() ->
+	  read_cache_page() for the page (in page cache of table of inodes
+	  $MFT, inode 0) containing the on-disk inode.  This page has
+	  PageUptodate() clear because of Process 1 (see above) so
+	  read_cache_page() blocks when it tries to take the page lock for the
+	  page so it can call ntfs_read_page().
+	  Thus Process 1 is holding the page lock on the page containing the
+	  on-disk inode X and it is waiting on the inode X to be unlocked in
+	  ifind() so it can write the page out and then unlock the page.
+	  And Process 2 is holding the inode lock on inode X and is waiting for
+	  the page to be unlocked so it can call ntfs_readpage() or discover
+	  that Process 1 set PageUptodate() again and use the page.
+	  Thus we have a deadlock due to ifind() waiting on the inode lock.
+	  The solution: The fix is to use the newly introduced
+	  ilookup5_nowait() which does not wait on the inode's lock and hence
+	  avoids the deadlock.  This is safe as we do not care about the VFS
+	  inode and only use the fact that it is in the VFS inode cache and the
+	  fact that the vfs and ntfs inodes are one struct in memory to find
+	  the ntfs inode in memory if present.  Also, the ntfs inode has its
+	  own locking so it does not matter if the vfs inode is locked.
 
 2.1.22 - Many bug and race fixes and error handling improvements.
 
diff --git a/fs/ntfs/mft.c b/fs/ntfs/mft.c
index 3d0ba8e60adc..ac9ff39aa834 100644
--- a/fs/ntfs/mft.c
+++ b/fs/ntfs/mft.c
@@ -948,20 +948,23 @@ BOOL ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
 	na.name_len = 0;
 	na.type = AT_UNUSED;
 	/*
-	 * For inode 0, i.e. $MFT itself, we cannot use ilookup5() from here or
-	 * we deadlock because the inode is already locked by the kernel
-	 * (fs/fs-writeback.c::__sync_single_inode()) and ilookup5() waits
-	 * until the inode is unlocked before returning it and it never gets
-	 * unlocked because ntfs_should_write_mft_record() never returns.  )-:
-	 * Fortunately, we have inode 0 pinned in icache for the duration of
-	 * the mount so we can access it directly.
+	 * Optimize inode 0, i.e. $MFT itself, since we have it in memory and
+	 * we get here for it rather often.
 	 */
 	if (!mft_no) {
 		/* Balance the below iput(). */
 		vi = igrab(mft_vi);
 		BUG_ON(vi != mft_vi);
-	} else
-		vi = ilookup5(sb, mft_no, (test_t)ntfs_test_inode, &na);
+	} else {
+		/*
+		 * Have to use ilookup5_nowait() since ilookup5() waits for the
+		 * inode lock which causes ntfs to deadlock when a concurrent
+		 * inode write via the inode dirty code paths and the page
+		 * dirty code path of the inode dirty code path when writing
+		 * $MFT occurs.
+		 */
+		vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na);
+	}
 	if (vi) {
 		ntfs_debug("Base inode 0x%lx is in icache.", mft_no);
 		/* The inode is in icache. */
@@ -1016,7 +1019,13 @@ BOOL ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no,
 	na.mft_no = MREF_LE(m->base_mft_record);
 	ntfs_debug("Mft record 0x%lx is an extent record.  Looking for base "
 			"inode 0x%lx in icache.", mft_no, na.mft_no);
-	vi = ilookup5(sb, na.mft_no, (test_t)ntfs_test_inode, &na);
+	if (!na.mft_no) {
+		/* Balance the below iput(). */
+		vi = igrab(mft_vi);
+		BUG_ON(vi != mft_vi);
+	} else
+		vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode,
+				&na);
 	if (!vi) {
 		/*
 		 * The base inode is not in icache, write this extent mft