aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-07-13 19:39:25 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-07-13 19:39:25 -0400
commit1cf29683f4414296dc772a87caa207cab16c310c (patch)
tree90d05ba531c5ad48d8d171d5b78038a7df7728e7
parent4a390e07fc53ce9dd615d7b788e9ecc73f87ad94 (diff)
parent96577c43827697ca1af5982fa256a34786d0c720 (diff)
Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
* 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: jbd2: fix race between write_metadata_buffer and get_write_access ext4: Fix ext4_mb_initialize_context() to initialize all fields ext4: fix null handler of ioctls in no journal mode ext4: Fix buffer head reference leak in no-journal mode ext4: Move __ext4_journalled_writepage() to avoid forward declaration ext4: Fix mmap/truncate race when blocksize < pagesize && !nodellaoc ext4: Fix mmap/truncate race when blocksize < pagesize && delayed allocation ext4: Don't look at buffer_heads outside i_size. ext4: Fix goal inum check in the inode allocator ext4: fix no journal corruption with locale-gen ext4: Calculate required journal credits for inserting an extent properly ext4: Fix truncation of symlinks after failed write jbd2: Fix a race between checkpointing code and journal_get_write_access() ext4: Use rcu_barrier() on module unload. ext4: naturally align struct ext4_allocation_request ext4: mark several more functions in mballoc.c as noinline ext4: Fix potential reclaim deadlock when truncating partial block jbd2: Remove GFP_ATOMIC kmalloc from inside spinlock critical region ext4: Fix type warning on 64-bit platforms in tracing events header
-rw-r--r--fs/ext4/ext4.h14
-rw-r--r--fs/ext4/ext4_jbd2.c4
-rw-r--r--fs/ext4/ext4_jbd2.h6
-rw-r--r--fs/ext4/extents.c1
-rw-r--r--fs/ext4/ialloc.c2
-rw-r--r--fs/ext4/inode.c384
-rw-r--r--fs/ext4/ioctl.c20
-rw-r--r--fs/ext4/mballoc.c50
-rw-r--r--fs/jbd2/journal.c31
-rw-r--r--fs/jbd2/transaction.c68
-rw-r--r--include/trace/events/ext4.h48
11 files changed, 235 insertions, 393 deletions
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 0ddf7e55abe1..9714db393efe 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -93,20 +93,20 @@ typedef unsigned int ext4_group_t;
93struct ext4_allocation_request { 93struct ext4_allocation_request {
94 /* target inode for block we're allocating */ 94 /* target inode for block we're allocating */
95 struct inode *inode; 95 struct inode *inode;
96 /* how many blocks we want to allocate */
97 unsigned int len;
96 /* logical block in target inode */ 98 /* logical block in target inode */
97 ext4_lblk_t logical; 99 ext4_lblk_t logical;
98 /* phys. target (a hint) */
99 ext4_fsblk_t goal;
100 /* the closest logical allocated block to the left */ 100 /* the closest logical allocated block to the left */
101 ext4_lblk_t lleft; 101 ext4_lblk_t lleft;
102 /* phys. block for ^^^ */
103 ext4_fsblk_t pleft;
104 /* the closest logical allocated block to the right */ 102 /* the closest logical allocated block to the right */
105 ext4_lblk_t lright; 103 ext4_lblk_t lright;
106 /* phys. block for ^^^ */ 104 /* phys. target (a hint) */
105 ext4_fsblk_t goal;
106 /* phys. block for the closest logical allocated block to the left */
107 ext4_fsblk_t pleft;
108 /* phys. block for the closest logical allocated block to the right */
107 ext4_fsblk_t pright; 109 ext4_fsblk_t pright;
108 /* how many blocks we want to allocate */
109 unsigned int len;
110 /* flags. see above EXT4_MB_HINT_* */ 110 /* flags. see above EXT4_MB_HINT_* */
111 unsigned int flags; 111 unsigned int flags;
112}; 112};
diff --git a/fs/ext4/ext4_jbd2.c b/fs/ext4/ext4_jbd2.c
index ad13a84644e1..eb27fd0f2ee8 100644
--- a/fs/ext4/ext4_jbd2.c
+++ b/fs/ext4/ext4_jbd2.c
@@ -43,6 +43,8 @@ int __ext4_journal_forget(const char *where, handle_t *handle,
43 ext4_journal_abort_handle(where, __func__, bh, 43 ext4_journal_abort_handle(where, __func__, bh,
44 handle, err); 44 handle, err);
45 } 45 }
46 else
47 brelse(bh);
46 return err; 48 return err;
47} 49}
48 50
@@ -57,6 +59,8 @@ int __ext4_journal_revoke(const char *where, handle_t *handle,
57 ext4_journal_abort_handle(where, __func__, bh, 59 ext4_journal_abort_handle(where, __func__, bh,
58 handle, err); 60 handle, err);
59 } 61 }
62 else
63 brelse(bh);
60 return err; 64 return err;
61} 65}
62 66
diff --git a/fs/ext4/ext4_jbd2.h b/fs/ext4/ext4_jbd2.h
index be2f426f6805..139fb8cb87e4 100644
--- a/fs/ext4/ext4_jbd2.h
+++ b/fs/ext4/ext4_jbd2.h
@@ -131,9 +131,11 @@ int __ext4_journal_get_undo_access(const char *where, handle_t *handle,
131int __ext4_journal_get_write_access(const char *where, handle_t *handle, 131int __ext4_journal_get_write_access(const char *where, handle_t *handle,
132 struct buffer_head *bh); 132 struct buffer_head *bh);
133 133
134/* When called with an invalid handle, this will still do a put on the BH */
134int __ext4_journal_forget(const char *where, handle_t *handle, 135int __ext4_journal_forget(const char *where, handle_t *handle,
135 struct buffer_head *bh); 136 struct buffer_head *bh);
136 137
138/* When called with an invalid handle, this will still do a put on the BH */
137int __ext4_journal_revoke(const char *where, handle_t *handle, 139int __ext4_journal_revoke(const char *where, handle_t *handle,
138 ext4_fsblk_t blocknr, struct buffer_head *bh); 140 ext4_fsblk_t blocknr, struct buffer_head *bh);
139 141
@@ -281,10 +283,10 @@ static inline int ext4_should_order_data(struct inode *inode)
281 283
282static inline int ext4_should_writeback_data(struct inode *inode) 284static inline int ext4_should_writeback_data(struct inode *inode)
283{ 285{
284 if (EXT4_JOURNAL(inode) == NULL)
285 return 0;
286 if (!S_ISREG(inode->i_mode)) 286 if (!S_ISREG(inode->i_mode))
287 return 0; 287 return 0;
288 if (EXT4_JOURNAL(inode) == NULL)
289 return 1;
288 if (EXT4_I(inode)->i_flags & EXT4_JOURNAL_DATA_FL) 290 if (EXT4_I(inode)->i_flags & EXT4_JOURNAL_DATA_FL)
289 return 0; 291 return 0;
290 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA) 292 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 50322a09bd01..73ebfb44ad75 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -1977,6 +1977,7 @@ int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
1977 */ 1977 */
1978 /* 1 bitmap, 1 block group descriptor */ 1978 /* 1 bitmap, 1 block group descriptor */
1979 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb); 1979 ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
1980 return ret;
1980 } 1981 }
1981 } 1982 }
1982 1983
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index 2f645732e3b7..29e6dc7299b8 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -833,7 +833,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
833 if (!goal) 833 if (!goal)
834 goal = sbi->s_inode_goal; 834 goal = sbi->s_inode_goal;
835 835
836 if (goal && goal < le32_to_cpu(sbi->s_es->s_inodes_count)) { 836 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
837 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); 837 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
838 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); 838 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
839 ret2 = 0; 839 ret2 = 0;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 60a26f3a6f8b..f9c642b22efa 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -78,16 +78,14 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
78 * but there may still be a record of it in the journal, and that record 78 * but there may still be a record of it in the journal, and that record
79 * still needs to be revoked. 79 * still needs to be revoked.
80 * 80 *
81 * If the handle isn't valid we're not journaling so there's nothing to do. 81 * If the handle isn't valid we're not journaling, but we still need to
82 * call into ext4_journal_revoke() to put the buffer head.
82 */ 83 */
83int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, 84int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
84 struct buffer_head *bh, ext4_fsblk_t blocknr) 85 struct buffer_head *bh, ext4_fsblk_t blocknr)
85{ 86{
86 int err; 87 int err;
87 88
88 if (!ext4_handle_valid(handle))
89 return 0;
90
91 might_sleep(); 89 might_sleep();
92 90
93 BUFFER_TRACE(bh, "enter"); 91 BUFFER_TRACE(bh, "enter");
@@ -1513,14 +1511,14 @@ retry:
1513 * Add inode to orphan list in case we crash before 1511 * Add inode to orphan list in case we crash before
1514 * truncate finishes 1512 * truncate finishes
1515 */ 1513 */
1516 if (pos + len > inode->i_size) 1514 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1517 ext4_orphan_add(handle, inode); 1515 ext4_orphan_add(handle, inode);
1518 1516
1519 ext4_journal_stop(handle); 1517 ext4_journal_stop(handle);
1520 if (pos + len > inode->i_size) { 1518 if (pos + len > inode->i_size) {
1521 vmtruncate(inode, inode->i_size); 1519 ext4_truncate(inode);
1522 /* 1520 /*
1523 * If vmtruncate failed early the inode might 1521 * If truncate failed early the inode might
1524 * still be on the orphan list; we need to 1522 * still be on the orphan list; we need to
1525 * make sure the inode is removed from the 1523 * make sure the inode is removed from the
1526 * orphan list in that case. 1524 * orphan list in that case.
@@ -1614,7 +1612,7 @@ static int ext4_ordered_write_end(struct file *file,
1614 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, 1612 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1615 page, fsdata); 1613 page, fsdata);
1616 copied = ret2; 1614 copied = ret2;
1617 if (pos + len > inode->i_size) 1615 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1618 /* if we have allocated more blocks and copied 1616 /* if we have allocated more blocks and copied
1619 * less. We will have blocks allocated outside 1617 * less. We will have blocks allocated outside
1620 * inode->i_size. So truncate them 1618 * inode->i_size. So truncate them
@@ -1628,9 +1626,9 @@ static int ext4_ordered_write_end(struct file *file,
1628 ret = ret2; 1626 ret = ret2;
1629 1627
1630 if (pos + len > inode->i_size) { 1628 if (pos + len > inode->i_size) {
1631 vmtruncate(inode, inode->i_size); 1629 ext4_truncate(inode);
1632 /* 1630 /*
1633 * If vmtruncate failed early the inode might still be 1631 * If truncate failed early the inode might still be
1634 * on the orphan list; we need to make sure the inode 1632 * on the orphan list; we need to make sure the inode
1635 * is removed from the orphan list in that case. 1633 * is removed from the orphan list in that case.
1636 */ 1634 */
@@ -1655,7 +1653,7 @@ static int ext4_writeback_write_end(struct file *file,
1655 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, 1653 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1656 page, fsdata); 1654 page, fsdata);
1657 copied = ret2; 1655 copied = ret2;
1658 if (pos + len > inode->i_size) 1656 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1659 /* if we have allocated more blocks and copied 1657 /* if we have allocated more blocks and copied
1660 * less. We will have blocks allocated outside 1658 * less. We will have blocks allocated outside
1661 * inode->i_size. So truncate them 1659 * inode->i_size. So truncate them
@@ -1670,9 +1668,9 @@ static int ext4_writeback_write_end(struct file *file,
1670 ret = ret2; 1668 ret = ret2;
1671 1669
1672 if (pos + len > inode->i_size) { 1670 if (pos + len > inode->i_size) {
1673 vmtruncate(inode, inode->i_size); 1671 ext4_truncate(inode);
1674 /* 1672 /*
1675 * If vmtruncate failed early the inode might still be 1673 * If truncate failed early the inode might still be
1676 * on the orphan list; we need to make sure the inode 1674 * on the orphan list; we need to make sure the inode
1677 * is removed from the orphan list in that case. 1675 * is removed from the orphan list in that case.
1678 */ 1676 */
@@ -1722,7 +1720,7 @@ static int ext4_journalled_write_end(struct file *file,
1722 1720
1723 unlock_page(page); 1721 unlock_page(page);
1724 page_cache_release(page); 1722 page_cache_release(page);
1725 if (pos + len > inode->i_size) 1723 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1726 /* if we have allocated more blocks and copied 1724 /* if we have allocated more blocks and copied
1727 * less. We will have blocks allocated outside 1725 * less. We will have blocks allocated outside
1728 * inode->i_size. So truncate them 1726 * inode->i_size. So truncate them
@@ -1733,9 +1731,9 @@ static int ext4_journalled_write_end(struct file *file,
1733 if (!ret) 1731 if (!ret)
1734 ret = ret2; 1732 ret = ret2;
1735 if (pos + len > inode->i_size) { 1733 if (pos + len > inode->i_size) {
1736 vmtruncate(inode, inode->i_size); 1734 ext4_truncate(inode);
1737 /* 1735 /*
1738 * If vmtruncate failed early the inode might still be 1736 * If truncate failed early the inode might still be
1739 * on the orphan list; we need to make sure the inode 1737 * on the orphan list; we need to make sure the inode
1740 * is removed from the orphan list in that case. 1738 * is removed from the orphan list in that case.
1741 */ 1739 */
@@ -2305,15 +2303,9 @@ flush_it:
2305 return; 2303 return;
2306} 2304}
2307 2305
2308static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh) 2306static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2309{ 2307{
2310 /* 2308 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2311 * unmapped buffer is possible for holes.
2312 * delay buffer is possible with delayed allocation.
2313 * We also need to consider unwritten buffer as unmapped.
2314 */
2315 return (!buffer_mapped(bh) || buffer_delay(bh) ||
2316 buffer_unwritten(bh)) && buffer_dirty(bh);
2317} 2309}
2318 2310
2319/* 2311/*
@@ -2398,9 +2390,9 @@ static int __mpage_da_writepage(struct page *page,
2398 * We need to try to allocate 2390 * We need to try to allocate
2399 * unmapped blocks in the same page. 2391 * unmapped blocks in the same page.
2400 * Otherwise we won't make progress 2392 * Otherwise we won't make progress
2401 * with the page in ext4_da_writepage 2393 * with the page in ext4_writepage
2402 */ 2394 */
2403 if (ext4_bh_unmapped_or_delay(NULL, bh)) { 2395 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2404 mpage_add_bh_to_extent(mpd, logical, 2396 mpage_add_bh_to_extent(mpd, logical,
2405 bh->b_size, 2397 bh->b_size,
2406 bh->b_state); 2398 bh->b_state);
@@ -2517,7 +2509,6 @@ static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2517 * so call get_block_wrap with create = 0 2509 * so call get_block_wrap with create = 0
2518 */ 2510 */
2519 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0); 2511 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2520 BUG_ON(create && ret == 0);
2521 if (ret > 0) { 2512 if (ret > 0) {
2522 bh_result->b_size = (ret << inode->i_blkbits); 2513 bh_result->b_size = (ret << inode->i_blkbits);
2523 ret = 0; 2514 ret = 0;
@@ -2525,15 +2516,102 @@ static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2525 return ret; 2516 return ret;
2526} 2517}
2527 2518
2519static int bget_one(handle_t *handle, struct buffer_head *bh)
2520{
2521 get_bh(bh);
2522 return 0;
2523}
2524
2525static int bput_one(handle_t *handle, struct buffer_head *bh)
2526{
2527 put_bh(bh);
2528 return 0;
2529}
2530
2531static int __ext4_journalled_writepage(struct page *page,
2532 struct writeback_control *wbc,
2533 unsigned int len)
2534{
2535 struct address_space *mapping = page->mapping;
2536 struct inode *inode = mapping->host;
2537 struct buffer_head *page_bufs;
2538 handle_t *handle = NULL;
2539 int ret = 0;
2540 int err;
2541
2542 page_bufs = page_buffers(page);
2543 BUG_ON(!page_bufs);
2544 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2545 /* As soon as we unlock the page, it can go away, but we have
2546 * references to buffers so we are safe */
2547 unlock_page(page);
2548
2549 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2550 if (IS_ERR(handle)) {
2551 ret = PTR_ERR(handle);
2552 goto out;
2553 }
2554
2555 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2556 do_journal_get_write_access);
2557
2558 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2559 write_end_fn);
2560 if (ret == 0)
2561 ret = err;
2562 err = ext4_journal_stop(handle);
2563 if (!ret)
2564 ret = err;
2565
2566 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2567 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2568out:
2569 return ret;
2570}
2571
2528/* 2572/*
2573 * Note that we don't need to start a transaction unless we're journaling data
2574 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2575 * need to file the inode to the transaction's list in ordered mode because if
2576 * we are writing back data added by write(), the inode is already there and if
2577 * we are writing back data modified via mmap(), noone guarantees in which
2578 * transaction the data will hit the disk. In case we are journaling data, we
2579 * cannot start transaction directly because transaction start ranks above page
2580 * lock so we have to do some magic.
2581 *
2529 * This function can get called via... 2582 * This function can get called via...
2530 * - ext4_da_writepages after taking page lock (have journal handle) 2583 * - ext4_da_writepages after taking page lock (have journal handle)
2531 * - journal_submit_inode_data_buffers (no journal handle) 2584 * - journal_submit_inode_data_buffers (no journal handle)
2532 * - shrink_page_list via pdflush (no journal handle) 2585 * - shrink_page_list via pdflush (no journal handle)
2533 * - grab_page_cache when doing write_begin (have journal handle) 2586 * - grab_page_cache when doing write_begin (have journal handle)
2587 *
2588 * We don't do any block allocation in this function. If we have page with
2589 * multiple blocks we need to write those buffer_heads that are mapped. This
2590 * is important for mmaped based write. So if we do with blocksize 1K
2591 * truncate(f, 1024);
2592 * a = mmap(f, 0, 4096);
2593 * a[0] = 'a';
2594 * truncate(f, 4096);
2595 * we have in the page first buffer_head mapped via page_mkwrite call back
2596 * but other bufer_heads would be unmapped but dirty(dirty done via the
2597 * do_wp_page). So writepage should write the first block. If we modify
2598 * the mmap area beyond 1024 we will again get a page_fault and the
2599 * page_mkwrite callback will do the block allocation and mark the
2600 * buffer_heads mapped.
2601 *
2602 * We redirty the page if we have any buffer_heads that is either delay or
2603 * unwritten in the page.
2604 *
2605 * We can get recursively called as show below.
2606 *
2607 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2608 * ext4_writepage()
2609 *
2610 * But since we don't do any block allocation we should not deadlock.
2611 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2534 */ 2612 */
2535static int ext4_da_writepage(struct page *page, 2613static int ext4_writepage(struct page *page,
2536 struct writeback_control *wbc) 2614 struct writeback_control *wbc)
2537{ 2615{
2538 int ret = 0; 2616 int ret = 0;
2539 loff_t size; 2617 loff_t size;
@@ -2541,7 +2619,7 @@ static int ext4_da_writepage(struct page *page,
2541 struct buffer_head *page_bufs; 2619 struct buffer_head *page_bufs;
2542 struct inode *inode = page->mapping->host; 2620 struct inode *inode = page->mapping->host;
2543 2621
2544 trace_ext4_da_writepage(inode, page); 2622 trace_ext4_writepage(inode, page);
2545 size = i_size_read(inode); 2623 size = i_size_read(inode);
2546 if (page->index == size >> PAGE_CACHE_SHIFT) 2624 if (page->index == size >> PAGE_CACHE_SHIFT)
2547 len = size & ~PAGE_CACHE_MASK; 2625 len = size & ~PAGE_CACHE_MASK;
@@ -2551,7 +2629,7 @@ static int ext4_da_writepage(struct page *page,
2551 if (page_has_buffers(page)) { 2629 if (page_has_buffers(page)) {
2552 page_bufs = page_buffers(page); 2630 page_bufs = page_buffers(page);
2553 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, 2631 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2554 ext4_bh_unmapped_or_delay)) { 2632 ext4_bh_delay_or_unwritten)) {
2555 /* 2633 /*
2556 * We don't want to do block allocation 2634 * We don't want to do block allocation
2557 * So redirty the page and return 2635 * So redirty the page and return
@@ -2578,13 +2656,13 @@ static int ext4_da_writepage(struct page *page,
2578 * all are mapped and non delay. We don't want to 2656 * all are mapped and non delay. We don't want to
2579 * do block allocation here. 2657 * do block allocation here.
2580 */ 2658 */
2581 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, 2659 ret = block_prepare_write(page, 0, len,
2582 noalloc_get_block_write); 2660 noalloc_get_block_write);
2583 if (!ret) { 2661 if (!ret) {
2584 page_bufs = page_buffers(page); 2662 page_bufs = page_buffers(page);
2585 /* check whether all are mapped and non delay */ 2663 /* check whether all are mapped and non delay */
2586 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, 2664 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2587 ext4_bh_unmapped_or_delay)) { 2665 ext4_bh_delay_or_unwritten)) {
2588 redirty_page_for_writepage(wbc, page); 2666 redirty_page_for_writepage(wbc, page);
2589 unlock_page(page); 2667 unlock_page(page);
2590 return 0; 2668 return 0;
@@ -2600,7 +2678,16 @@ static int ext4_da_writepage(struct page *page,
2600 return 0; 2678 return 0;
2601 } 2679 }
2602 /* now mark the buffer_heads as dirty and uptodate */ 2680 /* now mark the buffer_heads as dirty and uptodate */
2603 block_commit_write(page, 0, PAGE_CACHE_SIZE); 2681 block_commit_write(page, 0, len);
2682 }
2683
2684 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2685 /*
2686 * It's mmapped pagecache. Add buffers and journal it. There
2687 * doesn't seem much point in redirtying the page here.
2688 */
2689 ClearPageChecked(page);
2690 return __ext4_journalled_writepage(page, wbc, len);
2604 } 2691 }
2605 2692
2606 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) 2693 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
@@ -2907,7 +2994,7 @@ retry:
2907 * i_size_read because we hold i_mutex. 2994 * i_size_read because we hold i_mutex.
2908 */ 2995 */
2909 if (pos + len > inode->i_size) 2996 if (pos + len > inode->i_size)
2910 vmtruncate(inode, inode->i_size); 2997 ext4_truncate(inode);
2911 } 2998 }
2912 2999
2913 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) 3000 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
@@ -3130,222 +3217,6 @@ static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3130 return generic_block_bmap(mapping, block, ext4_get_block); 3217 return generic_block_bmap(mapping, block, ext4_get_block);
3131} 3218}
3132 3219
3133static int bget_one(handle_t *handle, struct buffer_head *bh)
3134{
3135 get_bh(bh);
3136 return 0;
3137}
3138
3139static int bput_one(handle_t *handle, struct buffer_head *bh)
3140{
3141 put_bh(bh);
3142 return 0;
3143}
3144
3145/*
3146 * Note that we don't need to start a transaction unless we're journaling data
3147 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3148 * need to file the inode to the transaction's list in ordered mode because if
3149 * we are writing back data added by write(), the inode is already there and if
3150 * we are writing back data modified via mmap(), noone guarantees in which
3151 * transaction the data will hit the disk. In case we are journaling data, we
3152 * cannot start transaction directly because transaction start ranks above page
3153 * lock so we have to do some magic.
3154 *
3155 * In all journaling modes block_write_full_page() will start the I/O.
3156 *
3157 * Problem:
3158 *
3159 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3160 * ext4_writepage()
3161 *
3162 * Similar for:
3163 *
3164 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3165 *
3166 * Same applies to ext4_get_block(). We will deadlock on various things like
3167 * lock_journal and i_data_sem
3168 *
3169 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3170 * allocations fail.
3171 *
3172 * 16May01: If we're reentered then journal_current_handle() will be
3173 * non-zero. We simply *return*.
3174 *
3175 * 1 July 2001: @@@ FIXME:
3176 * In journalled data mode, a data buffer may be metadata against the
3177 * current transaction. But the same file is part of a shared mapping
3178 * and someone does a writepage() on it.
3179 *
3180 * We will move the buffer onto the async_data list, but *after* it has
3181 * been dirtied. So there's a small window where we have dirty data on
3182 * BJ_Metadata.
3183 *
3184 * Note that this only applies to the last partial page in the file. The
3185 * bit which block_write_full_page() uses prepare/commit for. (That's
3186 * broken code anyway: it's wrong for msync()).
3187 *
3188 * It's a rare case: affects the final partial page, for journalled data
3189 * where the file is subject to bith write() and writepage() in the same
3190 * transction. To fix it we'll need a custom block_write_full_page().
3191 * We'll probably need that anyway for journalling writepage() output.
3192 *
3193 * We don't honour synchronous mounts for writepage(). That would be
3194 * disastrous. Any write() or metadata operation will sync the fs for
3195 * us.
3196 *
3197 */
3198static int __ext4_normal_writepage(struct page *page,
3199 struct writeback_control *wbc)
3200{
3201 struct inode *inode = page->mapping->host;
3202
3203 if (test_opt(inode->i_sb, NOBH))
3204 return nobh_writepage(page, noalloc_get_block_write, wbc);
3205 else
3206 return block_write_full_page(page, noalloc_get_block_write,
3207 wbc);
3208}
3209
3210static int ext4_normal_writepage(struct page *page,
3211 struct writeback_control *wbc)
3212{
3213 struct inode *inode = page->mapping->host;
3214 loff_t size = i_size_read(inode);
3215 loff_t len;
3216
3217 trace_ext4_normal_writepage(inode, page);
3218 J_ASSERT(PageLocked(page));
3219 if (page->index == size >> PAGE_CACHE_SHIFT)
3220 len = size & ~PAGE_CACHE_MASK;
3221 else
3222 len = PAGE_CACHE_SIZE;
3223
3224 if (page_has_buffers(page)) {
3225 /* if page has buffers it should all be mapped
3226 * and allocated. If there are not buffers attached
3227 * to the page we know the page is dirty but it lost
3228 * buffers. That means that at some moment in time
3229 * after write_begin() / write_end() has been called
3230 * all buffers have been clean and thus they must have been
3231 * written at least once. So they are all mapped and we can
3232 * happily proceed with mapping them and writing the page.
3233 */
3234 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3235 ext4_bh_unmapped_or_delay));
3236 }
3237
3238 if (!ext4_journal_current_handle())
3239 return __ext4_normal_writepage(page, wbc);
3240
3241 redirty_page_for_writepage(wbc, page);
3242 unlock_page(page);
3243 return 0;
3244}
3245
3246static int __ext4_journalled_writepage(struct page *page,
3247 struct writeback_control *wbc)
3248{
3249 struct address_space *mapping = page->mapping;
3250 struct inode *inode = mapping->host;
3251 struct buffer_head *page_bufs;
3252 handle_t *handle = NULL;
3253 int ret = 0;
3254 int err;
3255
3256 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3257 noalloc_get_block_write);
3258 if (ret != 0)
3259 goto out_unlock;
3260
3261 page_bufs = page_buffers(page);
3262 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3263 bget_one);
3264 /* As soon as we unlock the page, it can go away, but we have
3265 * references to buffers so we are safe */
3266 unlock_page(page);
3267
3268 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3269 if (IS_ERR(handle)) {
3270 ret = PTR_ERR(handle);
3271 goto out;
3272 }
3273
3274 ret = walk_page_buffers(handle, page_bufs, 0,
3275 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3276
3277 err = walk_page_buffers(handle, page_bufs, 0,
3278 PAGE_CACHE_SIZE, NULL, write_end_fn);
3279 if (ret == 0)
3280 ret = err;
3281 err = ext4_journal_stop(handle);
3282 if (!ret)
3283 ret = err;
3284
3285 walk_page_buffers(handle, page_bufs, 0,
3286 PAGE_CACHE_SIZE, NULL, bput_one);
3287 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3288 goto out;
3289
3290out_unlock:
3291 unlock_page(page);
3292out:
3293 return ret;
3294}
3295
3296static int ext4_journalled_writepage(struct page *page,
3297 struct writeback_control *wbc)
3298{
3299 struct inode *inode = page->mapping->host;
3300 loff_t size = i_size_read(inode);
3301 loff_t len;
3302
3303 trace_ext4_journalled_writepage(inode, page);
3304 J_ASSERT(PageLocked(page));
3305 if (page->index == size >> PAGE_CACHE_SHIFT)
3306 len = size & ~PAGE_CACHE_MASK;
3307 else
3308 len = PAGE_CACHE_SIZE;
3309
3310 if (page_has_buffers(page)) {
3311 /* if page has buffers it should all be mapped
3312 * and allocated. If there are not buffers attached
3313 * to the page we know the page is dirty but it lost
3314 * buffers. That means that at some moment in time
3315 * after write_begin() / write_end() has been called
3316 * all buffers have been clean and thus they must have been
3317 * written at least once. So they are all mapped and we can
3318 * happily proceed with mapping them and writing the page.
3319 */
3320 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3321 ext4_bh_unmapped_or_delay));
3322 }
3323
3324 if (ext4_journal_current_handle())
3325 goto no_write;
3326
3327 if (PageChecked(page)) {
3328 /*
3329 * It's mmapped pagecache. Add buffers and journal it. There
3330 * doesn't seem much point in redirtying the page here.
3331 */
3332 ClearPageChecked(page);
3333 return __ext4_journalled_writepage(page, wbc);
3334 } else {
3335 /*
3336 * It may be a page full of checkpoint-mode buffers. We don't
3337 * really know unless we go poke around in the buffer_heads.
3338 * But block_write_full_page will do the right thing.
3339 */
3340 return block_write_full_page(page, noalloc_get_block_write,
3341 wbc);
3342 }
3343no_write:
3344 redirty_page_for_writepage(wbc, page);
3345 unlock_page(page);
3346 return 0;
3347}
3348
3349static int ext4_readpage(struct file *file, struct page *page) 3220static int ext4_readpage(struct file *file, struct page *page)
3350{ 3221{
3351 return mpage_readpage(page, ext4_get_block); 3222 return mpage_readpage(page, ext4_get_block);
@@ -3492,7 +3363,7 @@ static int ext4_journalled_set_page_dirty(struct page *page)
3492static const struct address_space_operations ext4_ordered_aops = { 3363static const struct address_space_operations ext4_ordered_aops = {
3493 .readpage = ext4_readpage, 3364 .readpage = ext4_readpage,
3494 .readpages = ext4_readpages, 3365 .readpages = ext4_readpages,
3495 .writepage = ext4_normal_writepage, 3366 .writepage = ext4_writepage,
3496 .sync_page = block_sync_page, 3367 .sync_page = block_sync_page,
3497 .write_begin = ext4_write_begin, 3368 .write_begin = ext4_write_begin,
3498 .write_end = ext4_ordered_write_end, 3369 .write_end = ext4_ordered_write_end,
@@ -3507,7 +3378,7 @@ static const struct address_space_operations ext4_ordered_aops = {
3507static const struct address_space_operations ext4_writeback_aops = { 3378static const struct address_space_operations ext4_writeback_aops = {
3508 .readpage = ext4_readpage, 3379 .readpage = ext4_readpage,
3509 .readpages = ext4_readpages, 3380 .readpages = ext4_readpages,
3510 .writepage = ext4_normal_writepage, 3381 .writepage = ext4_writepage,
3511 .sync_page = block_sync_page, 3382 .sync_page = block_sync_page,
3512 .write_begin = ext4_write_begin, 3383 .write_begin = ext4_write_begin,
3513 .write_end = ext4_writeback_write_end, 3384 .write_end = ext4_writeback_write_end,
@@ -3522,7 +3393,7 @@ static const struct address_space_operations ext4_writeback_aops = {
3522static const struct address_space_operations ext4_journalled_aops = { 3393static const struct address_space_operations ext4_journalled_aops = {
3523 .readpage = ext4_readpage, 3394 .readpage = ext4_readpage,
3524 .readpages = ext4_readpages, 3395 .readpages = ext4_readpages,
3525 .writepage = ext4_journalled_writepage, 3396 .writepage = ext4_writepage,
3526 .sync_page = block_sync_page, 3397 .sync_page = block_sync_page,
3527 .write_begin = ext4_write_begin, 3398 .write_begin = ext4_write_begin,
3528 .write_end = ext4_journalled_write_end, 3399 .write_end = ext4_journalled_write_end,
@@ -3536,7 +3407,7 @@ static const struct address_space_operations ext4_journalled_aops = {
3536static const struct address_space_operations ext4_da_aops = { 3407static const struct address_space_operations ext4_da_aops = {
3537 .readpage = ext4_readpage, 3408 .readpage = ext4_readpage,
3538 .readpages = ext4_readpages, 3409 .readpages = ext4_readpages,
3539 .writepage = ext4_da_writepage, 3410 .writepage = ext4_writepage,
3540 .writepages = ext4_da_writepages, 3411 .writepages = ext4_da_writepages,
3541 .sync_page = block_sync_page, 3412 .sync_page = block_sync_page,
3542 .write_begin = ext4_da_write_begin, 3413 .write_begin = ext4_da_write_begin,
@@ -3583,7 +3454,8 @@ int ext4_block_truncate_page(handle_t *handle,
3583 struct page *page; 3454 struct page *page;
3584 int err = 0; 3455 int err = 0;
3585 3456
3586 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT); 3457 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3458 mapping_gfp_mask(mapping) & ~__GFP_FS);
3587 if (!page) 3459 if (!page)
3588 return -EINVAL; 3460 return -EINVAL;
3589 3461
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index 24a6abb2aef5..7050a9cd04a4 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -191,7 +191,7 @@ setversion_out:
191 case EXT4_IOC_GROUP_EXTEND: { 191 case EXT4_IOC_GROUP_EXTEND: {
192 ext4_fsblk_t n_blocks_count; 192 ext4_fsblk_t n_blocks_count;
193 struct super_block *sb = inode->i_sb; 193 struct super_block *sb = inode->i_sb;
194 int err, err2; 194 int err, err2=0;
195 195
196 if (!capable(CAP_SYS_RESOURCE)) 196 if (!capable(CAP_SYS_RESOURCE))
197 return -EPERM; 197 return -EPERM;
@@ -204,9 +204,11 @@ setversion_out:
204 return err; 204 return err;
205 205
206 err = ext4_group_extend(sb, EXT4_SB(sb)->s_es, n_blocks_count); 206 err = ext4_group_extend(sb, EXT4_SB(sb)->s_es, n_blocks_count);
207 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 207 if (EXT4_SB(sb)->s_journal) {
208 err2 = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 208 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
209 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 209 err2 = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
210 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
211 }
210 if (err == 0) 212 if (err == 0)
211 err = err2; 213 err = err2;
212 mnt_drop_write(filp->f_path.mnt); 214 mnt_drop_write(filp->f_path.mnt);
@@ -251,7 +253,7 @@ setversion_out:
251 case EXT4_IOC_GROUP_ADD: { 253 case EXT4_IOC_GROUP_ADD: {
252 struct ext4_new_group_data input; 254 struct ext4_new_group_data input;
253 struct super_block *sb = inode->i_sb; 255 struct super_block *sb = inode->i_sb;
254 int err, err2; 256 int err, err2=0;
255 257
256 if (!capable(CAP_SYS_RESOURCE)) 258 if (!capable(CAP_SYS_RESOURCE))
257 return -EPERM; 259 return -EPERM;
@@ -265,9 +267,11 @@ setversion_out:
265 return err; 267 return err;
266 268
267 err = ext4_group_add(sb, &input); 269 err = ext4_group_add(sb, &input);
268 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal); 270 if (EXT4_SB(sb)->s_journal) {
269 err2 = jbd2_journal_flush(EXT4_SB(sb)->s_journal); 271 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
270 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal); 272 err2 = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
273 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
274 }
271 if (err == 0) 275 if (err == 0)
272 err = err2; 276 err = err2;
273 mnt_drop_write(filp->f_path.mnt); 277 mnt_drop_write(filp->f_path.mnt);
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 519a0a686d94..cd258463e2a9 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -657,7 +657,8 @@ static void ext4_mb_mark_free_simple(struct super_block *sb,
657 } 657 }
658} 658}
659 659
660static void ext4_mb_generate_buddy(struct super_block *sb, 660static noinline_for_stack
661void ext4_mb_generate_buddy(struct super_block *sb,
661 void *buddy, void *bitmap, ext4_group_t group) 662 void *buddy, void *bitmap, ext4_group_t group)
662{ 663{
663 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 664 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
@@ -1480,7 +1481,8 @@ static void ext4_mb_measure_extent(struct ext4_allocation_context *ac,
1480 ext4_mb_check_limits(ac, e4b, 0); 1481 ext4_mb_check_limits(ac, e4b, 0);
1481} 1482}
1482 1483
1483static int ext4_mb_try_best_found(struct ext4_allocation_context *ac, 1484static noinline_for_stack
1485int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1484 struct ext4_buddy *e4b) 1486 struct ext4_buddy *e4b)
1485{ 1487{
1486 struct ext4_free_extent ex = ac->ac_b_ex; 1488 struct ext4_free_extent ex = ac->ac_b_ex;
@@ -1507,7 +1509,8 @@ static int ext4_mb_try_best_found(struct ext4_allocation_context *ac,
1507 return 0; 1509 return 0;
1508} 1510}
1509 1511
1510static int ext4_mb_find_by_goal(struct ext4_allocation_context *ac, 1512static noinline_for_stack
1513int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1511 struct ext4_buddy *e4b) 1514 struct ext4_buddy *e4b)
1512{ 1515{
1513 ext4_group_t group = ac->ac_g_ex.fe_group; 1516 ext4_group_t group = ac->ac_g_ex.fe_group;
@@ -1566,7 +1569,8 @@ static int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
1566 * The routine scans buddy structures (not bitmap!) from given order 1569 * The routine scans buddy structures (not bitmap!) from given order
1567 * to max order and tries to find big enough chunk to satisfy the req 1570 * to max order and tries to find big enough chunk to satisfy the req
1568 */ 1571 */
1569static void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac, 1572static noinline_for_stack
1573void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1570 struct ext4_buddy *e4b) 1574 struct ext4_buddy *e4b)
1571{ 1575{
1572 struct super_block *sb = ac->ac_sb; 1576 struct super_block *sb = ac->ac_sb;
@@ -1609,7 +1613,8 @@ static void ext4_mb_simple_scan_group(struct ext4_allocation_context *ac,
1609 * In order to optimize scanning, caller must pass number of 1613 * In order to optimize scanning, caller must pass number of
1610 * free blocks in the group, so the routine can know upper limit. 1614 * free blocks in the group, so the routine can know upper limit.
1611 */ 1615 */
1612static void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac, 1616static noinline_for_stack
1617void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1613 struct ext4_buddy *e4b) 1618 struct ext4_buddy *e4b)
1614{ 1619{
1615 struct super_block *sb = ac->ac_sb; 1620 struct super_block *sb = ac->ac_sb;
@@ -1668,7 +1673,8 @@ static void ext4_mb_complex_scan_group(struct ext4_allocation_context *ac,
1668 * we try to find stripe-aligned chunks for stripe-size requests 1673 * we try to find stripe-aligned chunks for stripe-size requests
1669 * XXX should do so at least for multiples of stripe size as well 1674 * XXX should do so at least for multiples of stripe size as well
1670 */ 1675 */
1671static void ext4_mb_scan_aligned(struct ext4_allocation_context *ac, 1676static noinline_for_stack
1677void ext4_mb_scan_aligned(struct ext4_allocation_context *ac,
1672 struct ext4_buddy *e4b) 1678 struct ext4_buddy *e4b)
1673{ 1679{
1674 struct super_block *sb = ac->ac_sb; 1680 struct super_block *sb = ac->ac_sb;
@@ -1831,7 +1837,8 @@ void ext4_mb_put_buddy_cache_lock(struct super_block *sb,
1831 1837
1832} 1838}
1833 1839
1834static int ext4_mb_init_group(struct super_block *sb, ext4_group_t group) 1840static noinline_for_stack
1841int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
1835{ 1842{
1836 1843
1837 int ret; 1844 int ret;
@@ -2902,7 +2909,11 @@ int __init init_ext4_mballoc(void)
2902 2909
2903void exit_ext4_mballoc(void) 2910void exit_ext4_mballoc(void)
2904{ 2911{
2905 /* XXX: synchronize_rcu(); */ 2912 /*
2913 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
2914 * before destroying the slab cache.
2915 */
2916 rcu_barrier();
2906 kmem_cache_destroy(ext4_pspace_cachep); 2917 kmem_cache_destroy(ext4_pspace_cachep);
2907 kmem_cache_destroy(ext4_ac_cachep); 2918 kmem_cache_destroy(ext4_ac_cachep);
2908 kmem_cache_destroy(ext4_free_ext_cachep); 2919 kmem_cache_destroy(ext4_free_ext_cachep);
@@ -3457,7 +3468,8 @@ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
3457 * used in in-core bitmap. buddy must be generated from this bitmap 3468 * used in in-core bitmap. buddy must be generated from this bitmap
3458 * Need to be called with ext4 group lock held 3469 * Need to be called with ext4 group lock held
3459 */ 3470 */
3460static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap, 3471static noinline_for_stack
3472void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
3461 ext4_group_t group) 3473 ext4_group_t group)
3462{ 3474{
3463 struct ext4_group_info *grp = ext4_get_group_info(sb, group); 3475 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
@@ -4215,14 +4227,9 @@ ext4_mb_initialize_context(struct ext4_allocation_context *ac,
4215 ext4_get_group_no_and_offset(sb, goal, &group, &block); 4227 ext4_get_group_no_and_offset(sb, goal, &group, &block);
4216 4228
4217 /* set up allocation goals */ 4229 /* set up allocation goals */
4230 memset(ac, 0, sizeof(struct ext4_allocation_context));
4218 ac->ac_b_ex.fe_logical = ar->logical; 4231 ac->ac_b_ex.fe_logical = ar->logical;
4219 ac->ac_b_ex.fe_group = 0;
4220 ac->ac_b_ex.fe_start = 0;
4221 ac->ac_b_ex.fe_len = 0;
4222 ac->ac_status = AC_STATUS_CONTINUE; 4232 ac->ac_status = AC_STATUS_CONTINUE;
4223 ac->ac_groups_scanned = 0;
4224 ac->ac_ex_scanned = 0;
4225 ac->ac_found = 0;
4226 ac->ac_sb = sb; 4233 ac->ac_sb = sb;
4227 ac->ac_inode = ar->inode; 4234 ac->ac_inode = ar->inode;
4228 ac->ac_o_ex.fe_logical = ar->logical; 4235 ac->ac_o_ex.fe_logical = ar->logical;
@@ -4233,15 +4240,7 @@ ext4_mb_initialize_context(struct ext4_allocation_context *ac,
4233 ac->ac_g_ex.fe_group = group; 4240 ac->ac_g_ex.fe_group = group;
4234 ac->ac_g_ex.fe_start = block; 4241 ac->ac_g_ex.fe_start = block;
4235 ac->ac_g_ex.fe_len = len; 4242 ac->ac_g_ex.fe_len = len;
4236 ac->ac_f_ex.fe_len = 0;
4237 ac->ac_flags = ar->flags; 4243 ac->ac_flags = ar->flags;
4238 ac->ac_2order = 0;
4239 ac->ac_criteria = 0;
4240 ac->ac_pa = NULL;
4241 ac->ac_bitmap_page = NULL;
4242 ac->ac_buddy_page = NULL;
4243 ac->alloc_semp = NULL;
4244 ac->ac_lg = NULL;
4245 4244
4246 /* we have to define context: we'll we work with a file or 4245 /* we have to define context: we'll we work with a file or
4247 * locality group. this is a policy, actually */ 4246 * locality group. this is a policy, actually */
@@ -4509,10 +4508,7 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
4509 } 4508 }
4510 4509
4511 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS); 4510 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4512 if (ac) { 4511 if (!ac) {
4513 ac->ac_sb = sb;
4514 ac->ac_inode = ar->inode;
4515 } else {
4516 ar->len = 0; 4512 ar->len = 0;
4517 *errp = -ENOMEM; 4513 *errp = -ENOMEM;
4518 goto out1; 4514 goto out1;
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 18bfd5dab642..e378cb383979 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -297,6 +297,7 @@ int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
297 unsigned int new_offset; 297 unsigned int new_offset;
298 struct buffer_head *bh_in = jh2bh(jh_in); 298 struct buffer_head *bh_in = jh2bh(jh_in);
299 struct jbd2_buffer_trigger_type *triggers; 299 struct jbd2_buffer_trigger_type *triggers;
300 journal_t *journal = transaction->t_journal;
300 301
301 /* 302 /*
302 * The buffer really shouldn't be locked: only the current committing 303 * The buffer really shouldn't be locked: only the current committing
@@ -310,6 +311,11 @@ int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
310 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); 311 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
311 312
312 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); 313 new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
314 /* keep subsequent assertions sane */
315 new_bh->b_state = 0;
316 init_buffer(new_bh, NULL, NULL);
317 atomic_set(&new_bh->b_count, 1);
318 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
313 319
314 /* 320 /*
315 * If a new transaction has already done a buffer copy-out, then 321 * If a new transaction has already done a buffer copy-out, then
@@ -388,14 +394,6 @@ repeat:
388 kunmap_atomic(mapped_data, KM_USER0); 394 kunmap_atomic(mapped_data, KM_USER0);
389 } 395 }
390 396
391 /* keep subsequent assertions sane */
392 new_bh->b_state = 0;
393 init_buffer(new_bh, NULL, NULL);
394 atomic_set(&new_bh->b_count, 1);
395 jbd_unlock_bh_state(bh_in);
396
397 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
398
399 set_bh_page(new_bh, new_page, new_offset); 397 set_bh_page(new_bh, new_page, new_offset);
400 new_jh->b_transaction = NULL; 398 new_jh->b_transaction = NULL;
401 new_bh->b_size = jh2bh(jh_in)->b_size; 399 new_bh->b_size = jh2bh(jh_in)->b_size;
@@ -412,7 +410,11 @@ repeat:
412 * copying is moved to the transaction's shadow queue. 410 * copying is moved to the transaction's shadow queue.
413 */ 411 */
414 JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); 412 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
415 jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); 413 spin_lock(&journal->j_list_lock);
414 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
415 spin_unlock(&journal->j_list_lock);
416 jbd_unlock_bh_state(bh_in);
417
416 JBUFFER_TRACE(new_jh, "file as BJ_IO"); 418 JBUFFER_TRACE(new_jh, "file as BJ_IO");
417 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO); 419 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
418 420
@@ -2410,6 +2412,7 @@ const char *jbd2_dev_to_name(dev_t device)
2410 int i = hash_32(device, CACHE_SIZE_BITS); 2412 int i = hash_32(device, CACHE_SIZE_BITS);
2411 char *ret; 2413 char *ret;
2412 struct block_device *bd; 2414 struct block_device *bd;
2415 static struct devname_cache *new_dev;
2413 2416
2414 rcu_read_lock(); 2417 rcu_read_lock();
2415 if (devcache[i] && devcache[i]->device == device) { 2418 if (devcache[i] && devcache[i]->device == device) {
@@ -2419,20 +2422,20 @@ const char *jbd2_dev_to_name(dev_t device)
2419 } 2422 }
2420 rcu_read_unlock(); 2423 rcu_read_unlock();
2421 2424
2425 new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2426 if (!new_dev)
2427 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2422 spin_lock(&devname_cache_lock); 2428 spin_lock(&devname_cache_lock);
2423 if (devcache[i]) { 2429 if (devcache[i]) {
2424 if (devcache[i]->device == device) { 2430 if (devcache[i]->device == device) {
2431 kfree(new_dev);
2425 ret = devcache[i]->devname; 2432 ret = devcache[i]->devname;
2426 spin_unlock(&devname_cache_lock); 2433 spin_unlock(&devname_cache_lock);
2427 return ret; 2434 return ret;
2428 } 2435 }
2429 call_rcu(&devcache[i]->rcu, free_devcache); 2436 call_rcu(&devcache[i]->rcu, free_devcache);
2430 } 2437 }
2431 devcache[i] = kmalloc(sizeof(struct devname_cache), GFP_KERNEL); 2438 devcache[i] = new_dev;
2432 if (!devcache[i]) {
2433 spin_unlock(&devname_cache_lock);
2434 return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2435 }
2436 devcache[i]->device = device; 2439 devcache[i]->device = device;
2437 bd = bdget(device); 2440 bd = bdget(device);
2438 if (bd) { 2441 if (bd) {
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index 494501edba6b..6213ac728f30 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -499,34 +499,15 @@ void jbd2_journal_unlock_updates (journal_t *journal)
499 wake_up(&journal->j_wait_transaction_locked); 499 wake_up(&journal->j_wait_transaction_locked);
500} 500}
501 501
502/* 502static void warn_dirty_buffer(struct buffer_head *bh)
503 * Report any unexpected dirty buffers which turn up. Normally those
504 * indicate an error, but they can occur if the user is running (say)
505 * tune2fs to modify the live filesystem, so we need the option of
506 * continuing as gracefully as possible. #
507 *
508 * The caller should already hold the journal lock and
509 * j_list_lock spinlock: most callers will need those anyway
510 * in order to probe the buffer's journaling state safely.
511 */
512static void jbd_unexpected_dirty_buffer(struct journal_head *jh)
513{ 503{
514 int jlist; 504 char b[BDEVNAME_SIZE];
515
516 /* If this buffer is one which might reasonably be dirty
517 * --- ie. data, or not part of this journal --- then
518 * we're OK to leave it alone, but otherwise we need to
519 * move the dirty bit to the journal's own internal
520 * JBDDirty bit. */
521 jlist = jh->b_jlist;
522 505
523 if (jlist == BJ_Metadata || jlist == BJ_Reserved || 506 printk(KERN_WARNING
524 jlist == BJ_Shadow || jlist == BJ_Forget) { 507 "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
525 struct buffer_head *bh = jh2bh(jh); 508 "There's a risk of filesystem corruption in case of system "
526 509 "crash.\n",
527 if (test_clear_buffer_dirty(bh)) 510 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
528 set_buffer_jbddirty(bh);
529 }
530} 511}
531 512
532/* 513/*
@@ -593,14 +574,16 @@ repeat:
593 if (jh->b_next_transaction) 574 if (jh->b_next_transaction)
594 J_ASSERT_JH(jh, jh->b_next_transaction == 575 J_ASSERT_JH(jh, jh->b_next_transaction ==
595 transaction); 576 transaction);
577 warn_dirty_buffer(bh);
596 } 578 }
597 /* 579 /*
598 * In any case we need to clean the dirty flag and we must 580 * In any case we need to clean the dirty flag and we must
599 * do it under the buffer lock to be sure we don't race 581 * do it under the buffer lock to be sure we don't race
600 * with running write-out. 582 * with running write-out.
601 */ 583 */
602 JBUFFER_TRACE(jh, "Unexpected dirty buffer"); 584 JBUFFER_TRACE(jh, "Journalling dirty buffer");
603 jbd_unexpected_dirty_buffer(jh); 585 clear_buffer_dirty(bh);
586 set_buffer_jbddirty(bh);
604 } 587 }
605 588
606 unlock_buffer(bh); 589 unlock_buffer(bh);
@@ -843,6 +826,15 @@ int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
843 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); 826 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
844 827
845 if (jh->b_transaction == NULL) { 828 if (jh->b_transaction == NULL) {
829 /*
830 * Previous jbd2_journal_forget() could have left the buffer
831 * with jbddirty bit set because it was being committed. When
832 * the commit finished, we've filed the buffer for
833 * checkpointing and marked it dirty. Now we are reallocating
834 * the buffer so the transaction freeing it must have
835 * committed and so it's safe to clear the dirty bit.
836 */
837 clear_buffer_dirty(jh2bh(jh));
846 jh->b_transaction = transaction; 838 jh->b_transaction = transaction;
847 839
848 /* first access by this transaction */ 840 /* first access by this transaction */
@@ -1644,8 +1636,13 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1644 1636
1645 if (jh->b_cp_transaction) { 1637 if (jh->b_cp_transaction) {
1646 JBUFFER_TRACE(jh, "on running+cp transaction"); 1638 JBUFFER_TRACE(jh, "on running+cp transaction");
1639 /*
1640 * We don't want to write the buffer anymore, clear the
1641 * bit so that we don't confuse checks in
1642 * __journal_file_buffer
1643 */
1644 clear_buffer_dirty(bh);
1647 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); 1645 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1648 clear_buffer_jbddirty(bh);
1649 may_free = 0; 1646 may_free = 0;
1650 } else { 1647 } else {
1651 JBUFFER_TRACE(jh, "on running transaction"); 1648 JBUFFER_TRACE(jh, "on running transaction");
@@ -1896,12 +1893,17 @@ void __jbd2_journal_file_buffer(struct journal_head *jh,
1896 if (jh->b_transaction && jh->b_jlist == jlist) 1893 if (jh->b_transaction && jh->b_jlist == jlist)
1897 return; 1894 return;
1898 1895
1899 /* The following list of buffer states needs to be consistent
1900 * with __jbd_unexpected_dirty_buffer()'s handling of dirty
1901 * state. */
1902
1903 if (jlist == BJ_Metadata || jlist == BJ_Reserved || 1896 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
1904 jlist == BJ_Shadow || jlist == BJ_Forget) { 1897 jlist == BJ_Shadow || jlist == BJ_Forget) {
1898 /*
1899 * For metadata buffers, we track dirty bit in buffer_jbddirty
1900 * instead of buffer_dirty. We should not see a dirty bit set
1901 * here because we clear it in do_get_write_access but e.g.
1902 * tune2fs can modify the sb and set the dirty bit at any time
1903 * so we try to gracefully handle that.
1904 */
1905 if (buffer_dirty(bh))
1906 warn_dirty_buffer(bh);
1905 if (test_clear_buffer_dirty(bh) || 1907 if (test_clear_buffer_dirty(bh) ||
1906 test_clear_buffer_jbddirty(bh)) 1908 test_clear_buffer_jbddirty(bh))
1907 was_dirty = 1; 1909 was_dirty = 1;
diff --git a/include/trace/events/ext4.h b/include/trace/events/ext4.h
index acf4cc9cd36d..dfbc9b0edc88 100644
--- a/include/trace/events/ext4.h
+++ b/include/trace/events/ext4.h
@@ -34,7 +34,8 @@ TRACE_EVENT(ext4_free_inode,
34 34
35 TP_printk("dev %s ino %lu mode %d uid %u gid %u blocks %llu", 35 TP_printk("dev %s ino %lu mode %d uid %u gid %u blocks %llu",
36 jbd2_dev_to_name(__entry->dev), __entry->ino, __entry->mode, 36 jbd2_dev_to_name(__entry->dev), __entry->ino, __entry->mode,
37 __entry->uid, __entry->gid, __entry->blocks) 37 __entry->uid, __entry->gid,
38 (unsigned long long) __entry->blocks)
38); 39);
39 40
40TRACE_EVENT(ext4_request_inode, 41TRACE_EVENT(ext4_request_inode,
@@ -189,7 +190,7 @@ TRACE_EVENT(ext4_journalled_write_end,
189 __entry->copied) 190 __entry->copied)
190); 191);
191 192
192TRACE_EVENT(ext4_da_writepage, 193TRACE_EVENT(ext4_writepage,
193 TP_PROTO(struct inode *inode, struct page *page), 194 TP_PROTO(struct inode *inode, struct page *page),
194 195
195 TP_ARGS(inode, page), 196 TP_ARGS(inode, page),
@@ -341,49 +342,6 @@ TRACE_EVENT(ext4_da_write_end,
341 __entry->copied) 342 __entry->copied)
342); 343);
343 344
344TRACE_EVENT(ext4_normal_writepage,
345 TP_PROTO(struct inode *inode, struct page *page),
346
347 TP_ARGS(inode, page),
348
349 TP_STRUCT__entry(
350 __field( dev_t, dev )
351 __field( ino_t, ino )
352 __field( pgoff_t, index )
353 ),
354
355 TP_fast_assign(
356 __entry->dev = inode->i_sb->s_dev;
357 __entry->ino = inode->i_ino;
358 __entry->index = page->index;
359 ),
360
361 TP_printk("dev %s ino %lu page_index %lu",
362 jbd2_dev_to_name(__entry->dev), __entry->ino, __entry->index)
363);
364
365TRACE_EVENT(ext4_journalled_writepage,
366 TP_PROTO(struct inode *inode, struct page *page),
367
368 TP_ARGS(inode, page),
369
370 TP_STRUCT__entry(
371 __field( dev_t, dev )
372 __field( ino_t, ino )
373 __field( pgoff_t, index )
374
375 ),
376
377 TP_fast_assign(
378 __entry->dev = inode->i_sb->s_dev;
379 __entry->ino = inode->i_ino;
380 __entry->index = page->index;
381 ),
382
383 TP_printk("dev %s ino %lu page_index %lu",
384 jbd2_dev_to_name(__entry->dev), __entry->ino, __entry->index)
385);
386
387TRACE_EVENT(ext4_discard_blocks, 345TRACE_EVENT(ext4_discard_blocks,
388 TP_PROTO(struct super_block *sb, unsigned long long blk, 346 TP_PROTO(struct super_block *sb, unsigned long long blk,
389 unsigned long long count), 347 unsigned long long count),