aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4
diff options
context:
space:
mode:
authorAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>2008-04-29 08:11:12 -0400
committerTheodore Ts'o <tytso@mit.edu>2008-04-29 08:11:12 -0400
commit267e4db9ac28a09973476e7ec2cb6807e609d35a (patch)
tree54eae44c7c0086800901866424ef1526e0ce863e /fs/ext4
parent3977c965ec35ce1a7eac988ad313f0fc9aee9660 (diff)
ext4: Fix race between migration and mmap write
Fail migrate if we allocated new blocks via mmap write. If we write to holes in the file via mmap, we end up allocating new blocks. This block allocation happens without taking inode->i_mutex. Since migrate is protected by i_mutex and migrate expects that no new blocks get allocated during migrate, fail migrate if new blocks get allocated. We can't take inode->i_mutex in the mmap write path because that would result in a locking order violation between i_mutex and mmap_sem. Also adding a separate rw_sempahore for protection is really high overhead for a rare operation such as migrate. Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Acked-by: Jan Kara <jack@suse.cz> Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Diffstat (limited to 'fs/ext4')
-rw-r--r--fs/ext4/inode.c13
-rw-r--r--fs/ext4/migrate.c39
2 files changed, 46 insertions, 6 deletions
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 8fab233cb05f..24a2604dde7b 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -985,6 +985,16 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
985 } else { 985 } else {
986 retval = ext4_get_blocks_handle(handle, inode, block, 986 retval = ext4_get_blocks_handle(handle, inode, block,
987 max_blocks, bh, create, extend_disksize); 987 max_blocks, bh, create, extend_disksize);
988
989 if (retval > 0 && buffer_new(bh)) {
990 /*
991 * We allocated new blocks which will result in
992 * i_data's format changing. Force the migrate
993 * to fail by clearing migrate flags
994 */
995 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
996 ~EXT4_EXT_MIGRATE;
997 }
988 } 998 }
989 up_write((&EXT4_I(inode)->i_data_sem)); 999 up_write((&EXT4_I(inode)->i_data_sem));
990 return retval; 1000 return retval;
@@ -2976,7 +2986,8 @@ static int ext4_do_update_inode(handle_t *handle,
2976 if (ext4_inode_blocks_set(handle, raw_inode, ei)) 2986 if (ext4_inode_blocks_set(handle, raw_inode, ei))
2977 goto out_brelse; 2987 goto out_brelse;
2978 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); 2988 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2979 raw_inode->i_flags = cpu_to_le32(ei->i_flags); 2989 /* clear the migrate flag in the raw_inode */
2990 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
2980 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != 2991 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2981 cpu_to_le32(EXT4_OS_HURD)) 2992 cpu_to_le32(EXT4_OS_HURD))
2982 raw_inode->i_file_acl_high = 2993 raw_inode->i_file_acl_high =
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index 5c1e27de7755..9b4fb07d192c 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -327,7 +327,7 @@ static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
327} 327}
328 328
329static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode, 329static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
330 struct inode *tmp_inode) 330 struct inode *tmp_inode)
331{ 331{
332 int retval; 332 int retval;
333 __le32 i_data[3]; 333 __le32 i_data[3];
@@ -339,7 +339,7 @@ static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
339 * i_data field of the original inode 339 * i_data field of the original inode
340 */ 340 */
341 retval = ext4_journal_extend(handle, 1); 341 retval = ext4_journal_extend(handle, 1);
342 if (retval != 0) { 342 if (retval) {
343 retval = ext4_journal_restart(handle, 1); 343 retval = ext4_journal_restart(handle, 1);
344 if (retval) 344 if (retval)
345 goto err_out; 345 goto err_out;
@@ -351,6 +351,18 @@ static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
351 351
352 down_write(&EXT4_I(inode)->i_data_sem); 352 down_write(&EXT4_I(inode)->i_data_sem);
353 /* 353 /*
354 * if EXT4_EXT_MIGRATE is cleared a block allocation
355 * happened after we started the migrate. We need to
356 * fail the migrate
357 */
358 if (!(EXT4_I(inode)->i_flags & EXT4_EXT_MIGRATE)) {
359 retval = -EAGAIN;
360 up_write(&EXT4_I(inode)->i_data_sem);
361 goto err_out;
362 } else
363 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
364 ~EXT4_EXT_MIGRATE;
365 /*
354 * We have the extent map build with the tmp inode. 366 * We have the extent map build with the tmp inode.
355 * Now copy the i_data across 367 * Now copy the i_data across
356 */ 368 */
@@ -508,6 +520,17 @@ int ext4_ext_migrate(struct inode *inode, struct file *filp,
508 * switch the inode format to prevent read. 520 * switch the inode format to prevent read.
509 */ 521 */
510 mutex_lock(&(inode->i_mutex)); 522 mutex_lock(&(inode->i_mutex));
523 /*
524 * Even though we take i_mutex we can still cause block allocation
525 * via mmap write to holes. If we have allocated new blocks we fail
526 * migrate. New block allocation will clear EXT4_EXT_MIGRATE flag.
527 * The flag is updated with i_data_sem held to prevent racing with
528 * block allocation.
529 */
530 down_read((&EXT4_I(inode)->i_data_sem));
531 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags | EXT4_EXT_MIGRATE;
532 up_read((&EXT4_I(inode)->i_data_sem));
533
511 handle = ext4_journal_start(inode, 1); 534 handle = ext4_journal_start(inode, 1);
512 535
513 ei = EXT4_I(inode); 536 ei = EXT4_I(inode);
@@ -559,9 +582,15 @@ err_out:
559 * tmp_inode 582 * tmp_inode
560 */ 583 */
561 free_ext_block(handle, tmp_inode); 584 free_ext_block(handle, tmp_inode);
562 else 585 else {
563 retval = ext4_ext_swap_inode_data(handle, inode, 586 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
564 tmp_inode); 587 if (retval)
588 /*
589 * if we fail to swap inode data free the extent
590 * details of the tmp inode
591 */
592 free_ext_block(handle, tmp_inode);
593 }
565 594
566 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */ 595 /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
567 if (ext4_journal_extend(handle, 1) != 0) 596 if (ext4_journal_extend(handle, 1) != 0)