incorrect direct io error handling (v3)
Subsystems (2)
RFC PATCH 1/1 incorrect direct io error handling (v3)
incorrect direct io error handling (v3)
Changes from v2:
- Remove BUG_ON(!mutex_is_locked(..)) for non blkdev.
- vmtruncate() called from generic_file_aio_write().
- depends on patch titled:
[PATH][RFC] mm: Move common segments checks to separate function
LOG:
If generic_file_direct_write() has fail (ENOSPC condition) inside __generic_file_aio_write_nolock() it may have instantiated
a few blocks outside i_size. And fsck will complain about wrong i_size (ext2, ext3 and reiserfs interpret i_size and biggest block difference as error), after fsck will fix error i_size will be increased to the biggest block, but this blocks contain gurbage from previous write attempt, this is not information leak, but its silence file data corruption. This issue affect fs regardless the values of blocksize or pagesize.
We need truncate any block beyond i_size after write have failed , do in simular generic_file_buffered_write() error path. Initialy i've proposed do it in __generic_file_aio_write_nolock() with explicit guarantee i_mutex always held, but not everybody was agree with it. So we may safely call vmtruncate() inside generic_file_aio_write(), here i_mutex already locked.
TEST_CASE:open("/mnt/test/BIG_FILE", O_WRONLY|O_CREAT|O_DIRECT, 0666) = 3 write(3, "aaaaaaaaaaaaaaa"..., 104857600) = -1 ENOSPC (No space left on device)
#stat /mnt/test/BIG_FILE
File: `/mnt/test/BIG_FILE'
Size: 0 Blocks: 110896 IO Block: 1024 regular empty file
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<^^^^^^^^file size is less than biggest block idx
Device: fe07h/65031d Inode: 14 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2007-01-24 20:03:38.000000000 +0300
Modify: 2007-01-24 20:03:38.000000000 +0300
Change: 2007-01-24 20:03:39.000000000 +0300
#fsck.ext3 -f /dev/VG/test
e2fsck 1.39 (29-May-2006)
Pass 1: Checking inodes, blocks, and sizes Inode 14, i_size is 0, should be 56556544. Fix<y>? yes Pass 2: Checking directory structure
-------
mm/filemap.c +16 -4
@@ -2058,8 +2058,9 @@ generic_file_direct_write(struct kiocb */* * Sync the fs metadata but not the minor inode changes and * of course not the data as we did direct DMA for the IO. -* i_mutex is held, which protects generic_osync_inode() from -* livelocking. AIO O_DIRECT ops attempt to sync metadata here. +* i_mutex may not being held, if so some specific locking +* ordering must protect generic_osync_inode() from livelocking. +* AIO O_DIRECT ops attempt to sync metadata here. */ if ((written >=0 || written == -EIOCBQUEUED ) && ((file->f_flags &O_SYNC ) ||IS_SYNC (inode))) {
@@ -2365,6 +2366,17 @@ ssize_t generic_file_aio_write(struct ki &iocb->ki_pos);mutex_unlock (&inode->i_mutex); +if (unlikely (ret <0 && (file->f_flags &O_DIRECT ))) { +ssize_t cnt =generic_segment_checks (nr_segs, iov,VERIFY_READ ); +loff_t isize =i_size_read (inode); +/* +* generic_file_direct_write() may have instantiated a few +* blocks outside i_size. Trim these off again. +*/ +if (cnt >0 && (pos + cnt > isize)) +vmtruncate (inode, isize); + } +if (ret >0 && ((file->f_flags &O_SYNC ) ||IS_SYNC (inode))) {ssize_t err;
@@ -2377,8 +2389,8 @@ ssize_t generic_file_aio_write(struct kiEXPORT_SYMBOL (generic_file_aio_write);/* -* Called under i_mutex for writes to S_ISREG files. Returns -EIO if something -* went wrong during pagecache shootdown. +* May be called without i_mutex for writes to S_ISREG files. +* Returns -EIO if something went wrong during pagecache shootdown. */ static ssize_t generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
Quoted 6 lines · On Wed, 24 Jan 2007 22:05:06 +0300 Dmitriy Monakhov <[email protected]> wrote:
incorrect direct io error handling (v3)
Changes from v2:
- Remove BUG_ON(!mutex_is_locked(..)) for non blkdev.
-vmtruncate()called fromgeneric_file_aio_write().
- depends on patch titled:
[PATH][RFC] mm: Move common segments checks to separate function
drat, I skipped that patch due to rejects, and because Nick is working on things in the same area.
Quoted 30 lines
LOG:
Ifgeneric_file_direct_write()has fail (ENOSPC condition) inside__generic_file_aio_write_nolock()it may have instantiated
a few blocks outsidei_size. And fsck will complain about wrongi_size(ext2, ext3 and reiserfs interpreti_sizeand biggest block difference as error), after fsck will fix errori_sizewill be increased to the biggest block, but this blocks contain gurbage from previous write attempt, this is not information leak, but its silence file data corruption. This issue affect fs regardless the values of blocksize or pagesize.
We need truncate any block beyondi_sizeafter write have failed , do in simulargeneric_file_buffered_write()error path. Initialy i've proposed do it in__generic_file_aio_write_nolock()with explicit guaranteei_mutexalways held, but not everybody was agree with it. So we may safely callvmtruncate()insidegeneric_file_aio_write(), herei_mutexalready locked.
TEST_CASE:open("/mnt/test/BIG_FILE", O_WRONLY|O_CREAT|O_DIRECT, 0666)= 3write(3, "aaaaaaaaaaaaaaa"..., 104857600)= -1 ENOSPC (No space left on device)#stat
/mnt/test/BIG_FILE
File: `/mnt/test/BIG_FILE'
Size: 0 Blocks: 110896 IO Block: 1024 regular empty file
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<^^^^^^^^file size is less than biggest block idxDevice: fe07h/65031d Inode: 14 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2007-01-24 20:03:38.000000000 +0300
Modify: 2007-01-24 20:03:38.000000000 +0300
Change: 2007-01-24 20:03:39.000000000 +0300
Quoted 9 lines
#fsck.ext3 -f
/dev/VG/test
e2fsck 1.39 (29-May-2006)
Pass 1: Checking inodes, blocks, and sizes Inode 14,i_sizeis 0, should be 56556544. Fix<y>? yes Pass 2: Checking directory structureSigned-off-by Dmitriy Monakhov <[email protected]>-------
mm/filemap.c +0 -0
Quoted 20 lines
index d01abb6..96840e5 100644
mm/filemap.c +4
-2@@ -2058,8 +2058,9 @@ generic_file_direct_write(struct kiocb */* * Sync the fs metadata but not the minor inode changes and * of course not the data as we did direct DMA for the IO. -* i_mutex is held, which protects generic_osync_inode() from -* livelocking. AIO O_DIRECT ops attempt to sync metadata here. +* i_mutex may not being held, if so some specific locking +* ordering must protect generic_osync_inode() from livelocking. +* AIO O_DIRECT ops attempt to sync metadata here. */ if ((written >=0 || written == -EIOCBQUEUED ) && ((file->f_flags &O_SYNC ) ||IS_SYNC (inode))) {@@ -2365,6 +2366,17 @@ ssize_t generic_file_aio_write(struct ki &iocb->ki_pos);mutex_unlock (&inode->i_mutex); +if (unlikely (ret <0 && (file->f_flags &O_DIRECT ))) {
mm/filemap.c ssize_t generic_file_aio_write(struct ki · line 2379
Quoted 10 lines
mm/filemap.c +10
-0+ssize_t cnt =generic_segment_checks (nr_segs, iov,VERIFY_READ ); +loff_t isize =i_size_read (inode); +/* +* generic_file_direct_write() may have instantiated a few +* blocks outside i_size. Trim these off again. +*/ +if (cnt >0 && (pos + cnt > isize)) +vmtruncate (inode, isize); + } +
vmtruncate() really wants i_mutex to be held. Can't we do that here?
Quoted 13 lines · Andrew Morton
Quoted 6 lines · On Wed, 24 Jan 2007 22:05:06 +0300 Dmitriy Monakhov <[email protected]> wrote:
incorrect direct io error handling (v3)
Changes from v2:
- Remove BUG_ON(!mutex_is_locked(..)) for non blkdev.
-vmtruncate()called fromgeneric_file_aio_write().
- depends on patch titled:
[PATH][RFC] mm: Move common segments checks to separate functiondrat, I skipped that patch due to rejects, and because Nick is working on things in the same area.
[skip]
Quoted 6 lines
Quoted 6 lines
(quoted diff) +0
-0if ((written >= 0 || written == -EIOCBQUEUED) && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {@@ -2365,6 +2366,17 @@ ssize_t generic_file_aio_write(struct ki &iocb->ki_pos); mutex_unlock(&inode->i_mutex);
Quoted 13 lines
Review ssize_t generic_file_aio_write(struct ki · line 2379
Quoted 11 lines
(quoted diff) +11
-0+ if (unlikely(ret < 0 && (file->f_flags & O_DIRECT))) { + ssize_t cnt = generic_segment_checks(nr_segs, iov, VERIFY_READ); + loff_t isize = i_size_read(inode); + /* + * generic_file_direct_write() may have instantiated a few + * blocks outside i_size. Trim these off again. + */ + if (cnt > 0 && (pos + cnt > isize)) + vmtruncate(inode, isize); + } +
vmtruncate()really wantsi_mutexto be held. Can't we do that here?
Yepp 110% true, baaahh ..... it looks like my brain was't clear at the time i wrote this. We have to do vmtruncate() brfore droping i_mutex , right after __generic_file_aio_write_nolock() call , i'm sorry to waste your time.