aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2008-07-15 11:36:38 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2008-07-15 11:36:38 -0400
commit8d2567a620ae8c24968a2bdc1c906c724fac1f6a (patch)
tree8e228abbadbe760e3f015d30c2e1180a67eeb8f9
parentbcf559e385ba099996c90469c817d2eb38aba418 (diff)
parent49f1487b2e41bd8439ea39a4f15b4064e823cc54 (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: (61 commits) ext4: Documention update for new ordered mode and delayed allocation ext4: do not set extents feature from the kernel ext4: Don't allow nonextenst mount option for large filesystem ext4: Enable delalloc by default. ext4: delayed allocation i_blocks fix for stat ext4: fix delalloc i_disksize early update issue ext4: Handle page without buffers in ext4_*_writepage() ext4: Add ordered mode support for delalloc ext4: Invert lock ordering of page_lock and transaction start in delalloc mm: Add range_cont mode for writeback ext4: delayed allocation ENOSPC handling percpu_counter: new function percpu_counter_sum_and_set ext4: Add delayed allocation support in data=writeback mode vfs: add hooks for ext4's delayed allocation support jbd2: Remove data=ordered mode support using jbd buffer heads ext4: Use new framework for data=ordered mode in JBD2 jbd2: Implement data=ordered mode handling via inodes vfs: export filemap_fdatawrite_range() ext4: Fix lock inversion in ext4_ext_truncate() ext4: Invert the locking order of page_lock and transaction start ...
-rw-r--r--Documentation/filesystems/ext4.txt125
-rw-r--r--fs/buffer.c19
-rw-r--r--fs/ext4/balloc.c209
-rw-r--r--fs/ext4/dir.c17
-rw-r--r--fs/ext4/ext4.h61
-rw-r--r--fs/ext4/ext4_extents.h1
-rw-r--r--fs/ext4/ext4_i.h10
-rw-r--r--fs/ext4/ext4_jbd2.h21
-rw-r--r--fs/ext4/ext4_sb.h5
-rw-r--r--fs/ext4/extents.c111
-rw-r--r--fs/ext4/file.c20
-rw-r--r--fs/ext4/fsync.c4
-rw-r--r--fs/ext4/group.h2
-rw-r--r--fs/ext4/ialloc.c113
-rw-r--r--fs/ext4/inode.c1591
-rw-r--r--fs/ext4/mballoc.c451
-rw-r--r--fs/ext4/namei.c45
-rw-r--r--fs/ext4/resize.c52
-rw-r--r--fs/ext4/super.c142
-rw-r--r--fs/ext4/xattr.c2
-rw-r--r--fs/ext4/xattr_trusted.c4
-rw-r--r--fs/ext4/xattr_user.c4
-rw-r--r--fs/jbd2/checkpoint.c1
-rw-r--r--fs/jbd2/commit.c294
-rw-r--r--fs/jbd2/journal.c53
-rw-r--r--fs/jbd2/transaction.c365
-rw-r--r--fs/mpage.c14
-rw-r--r--include/linux/fs.h2
-rw-r--r--include/linux/jbd2.h73
-rw-r--r--include/linux/mpage.h10
-rw-r--r--include/linux/percpu_counter.h12
-rw-r--r--include/linux/writeback.h1
-rw-r--r--lib/percpu_counter.c7
-rw-r--r--mm/filemap.c3
-rw-r--r--mm/page-writeback.c3
35 files changed, 2805 insertions, 1042 deletions
diff --git a/Documentation/filesystems/ext4.txt b/Documentation/filesystems/ext4.txt
index 0c5086db8352..80e193d82e2e 100644
--- a/Documentation/filesystems/ext4.txt
+++ b/Documentation/filesystems/ext4.txt
@@ -13,72 +13,93 @@ Mailing list: linux-ext4@vger.kernel.org
131. Quick usage instructions: 131. Quick usage instructions:
14=========================== 14===========================
15 15
16 - Grab updated e2fsprogs from 16 - Compile and install the latest version of e2fsprogs (as of this
17 ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs-interim/ 17 writing version 1.41) from:
18 This is a patchset on top of e2fsprogs-1.39, which can be found at 18
19 http://sourceforge.net/project/showfiles.php?group_id=2406
20
21 or
22
19 ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/ 23 ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/
20 24
21 - It's still mke2fs -j /dev/hda1 25 or grab the latest git repository from:
26
27 git://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git
28
29 - Create a new filesystem using the ext4dev filesystem type:
30
31 # mke2fs -t ext4dev /dev/hda1
32
33 Or configure an existing ext3 filesystem to support extents and set
34 the test_fs flag to indicate that it's ok for an in-development
35 filesystem to touch this filesystem:
22 36
23 - mount /dev/hda1 /wherever -t ext4dev 37 # tune2fs -O extents -E test_fs /dev/hda1
24 38
25 - To enable extents, 39 If the filesystem was created with 128 byte inodes, it can be
40 converted to use 256 byte for greater efficiency via:
26 41
27 mount /dev/hda1 /wherever -t ext4dev -o extents 42 # tune2fs -I 256 /dev/hda1
28 43
29 - The filesystem is compatible with the ext3 driver until you add a file 44 (Note: we currently do not have tools to convert an ext4dev
30 which has extents (ie: `mount -o extents', then create a file). 45 filesystem back to ext3; so please do not do try this on production
46 filesystems.)
31 47
32 NOTE: The "extents" mount flag is temporary. It will soon go away and 48 - Mounting:
33 extents will be enabled by the "-o extents" flag to mke2fs or tune2fs 49
50 # mount -t ext4dev /dev/hda1 /wherever
34 51
35 - When comparing performance with other filesystems, remember that 52 - When comparing performance with other filesystems, remember that
36 ext3/4 by default offers higher data integrity guarantees than most. So 53 ext3/4 by default offers higher data integrity guarantees than most.
37 when comparing with a metadata-only journalling filesystem, use `mount -o 54 So when comparing with a metadata-only journalling filesystem, such
38 data=writeback'. And you might as well use `mount -o nobh' too along 55 as ext3, use `mount -o data=writeback'. And you might as well use
39 with it. Making the journal larger than the mke2fs default often helps 56 `mount -o nobh' too along with it. Making the journal larger than
40 performance with metadata-intensive workloads. 57 the mke2fs default often helps performance with metadata-intensive
58 workloads.
41 59
422. Features 602. Features
43=========== 61===========
44 62
452.1 Currently available 632.1 Currently available
46 64
47* ability to use filesystems > 16TB 65* ability to use filesystems > 16TB (e2fsprogs support not available yet)
48* extent format reduces metadata overhead (RAM, IO for access, transactions) 66* extent format reduces metadata overhead (RAM, IO for access, transactions)
49* extent format more robust in face of on-disk corruption due to magics, 67* extent format more robust in face of on-disk corruption due to magics,
50* internal redunancy in tree 68* internal redunancy in tree
51 69* improved file allocation (multi-block alloc)
522.1 Previously available, soon to be enabled by default by "mkefs.ext4": 70* fix 32000 subdirectory limit
53 71* nsec timestamps for mtime, atime, ctime, create time
54* dir_index and resize inode will be on by default 72* inode version field on disk (NFSv4, Lustre)
55* large inodes will be used by default for fast EAs, nsec timestamps, etc 73* reduced e2fsck time via uninit_bg feature
74* journal checksumming for robustness, performance
75* persistent file preallocation (e.g for streaming media, databases)
76* ability to pack bitmaps and inode tables into larger virtual groups via the
77 flex_bg feature
78* large file support
79* Inode allocation using large virtual block groups via flex_bg
80* delayed allocation
81* large block (up to pagesize) support
82* efficent new ordered mode in JBD2 and ext4(avoid using buffer head to force
83 the ordering)
56 84
572.2 Candidate features for future inclusion 852.2 Candidate features for future inclusion
58 86
59There are several under discussion, whether they all make it in is 87* Online defrag (patches available but not well tested)
60partly a function of how much time everyone has to work on them: 88* reduced mke2fs time via lazy itable initialization in conjuction with
89 the uninit_bg feature (capability to do this is available in e2fsprogs
90 but a kernel thread to do lazy zeroing of unused inode table blocks
91 after filesystem is first mounted is required for safety)
61 92
62* improved file allocation (multi-block alloc, delayed alloc; basically done) 93There are several others under discussion, whether they all make it in is
63* fix 32000 subdirectory limit (patch exists, needs some e2fsck work) 94partly a function of how much time everyone has to work on them. Features like
64* nsec timestamps for mtime, atime, ctime, create time (patch exists, 95metadata checksumming have been discussed and planned for a bit but no patches
65 needs some e2fsck work) 96exist yet so I'm not sure they're in the near-term roadmap.
66* inode version field on disk (NFSv4, Lustre; prototype exists)
67* reduced mke2fs/e2fsck time via uninitialized groups (prototype exists)
68* journal checksumming for robustness, performance (prototype exists)
69* persistent file preallocation (e.g for streaming media, databases)
70 97
71Features like metadata checksumming have been discussed and planned for 98The big performance win will come with mballoc, delalloc and flex_bg
72a bit but no patches exist yet so I'm not sure they're in the near-term 99grouping of bitmaps and inode tables. Some test results available here:
73roadmap.
74 100
75The big performance win will come with mballoc and delalloc. CFS has 101 - http://www.bullopensource.org/ext4/20080530/ffsb-write-2.6.26-rc2.html
76been using mballoc for a few years already with Lustre, and IBM + Bull 102 - http://www.bullopensource.org/ext4/20080530/ffsb-readwrite-2.6.26-rc2.html
77did a lot of benchmarking on it. The reason it isn't in the first set of
78patches is partly a manageability issue, and partly because it doesn't
79directly affect the on-disk format (outside of much better allocation)
80so it isn't critical to get into the first round of changes. I believe
81Alex is working on a new set of patches right now.
82 103
833. Options 1043. Options
84========== 105==========
@@ -222,9 +243,11 @@ stripe=n Number of filesystem blocks that mballoc will try
222 to use for allocation size and alignment. For RAID5/6 243 to use for allocation size and alignment. For RAID5/6
223 systems this should be the number of data 244 systems this should be the number of data
224 disks * RAID chunk size in file system blocks. 245 disks * RAID chunk size in file system blocks.
225 246delalloc (*) Deferring block allocation until write-out time.
247nodelalloc Disable delayed allocation. Blocks are allocation
248 when data is copied from user to page cache.
226Data Mode 249Data Mode
227--------- 250=========
228There are 3 different data modes: 251There are 3 different data modes:
229 252
230* writeback mode 253* writeback mode
@@ -236,10 +259,10 @@ typically provide the best ext4 performance.
236 259
237* ordered mode 260* ordered mode
238In data=ordered mode, ext4 only officially journals metadata, but it logically 261In data=ordered mode, ext4 only officially journals metadata, but it logically
239groups metadata and data blocks into a single unit called a transaction. When 262groups metadata information related to data changes with the data blocks into a
240it's time to write the new metadata out to disk, the associated data blocks 263single unit called a transaction. When it's time to write the new metadata
241are written first. In general, this mode performs slightly slower than 264out to disk, the associated data blocks are written first. In general,
242writeback but significantly faster than journal mode. 265this mode performs slightly slower than writeback but significantly faster than journal mode.
243 266
244* journal mode 267* journal mode
245data=journal mode provides full data and metadata journaling. All new data is 268data=journal mode provides full data and metadata journaling. All new data is
@@ -247,7 +270,8 @@ written to the journal first, and then to its final location.
247In the event of a crash, the journal can be replayed, bringing both data and 270In the event of a crash, the journal can be replayed, bringing both data and
248metadata into a consistent state. This mode is the slowest except when data 271metadata into a consistent state. This mode is the slowest except when data
249needs to be read from and written to disk at the same time where it 272needs to be read from and written to disk at the same time where it
250outperforms all others modes. 273outperforms all others modes. Curently ext4 does not have delayed
274allocation support if this data journalling mode is selected.
251 275
252References 276References
253========== 277==========
@@ -256,7 +280,8 @@ kernel source: <file:fs/ext4/>
256 <file:fs/jbd2/> 280 <file:fs/jbd2/>
257 281
258programs: http://e2fsprogs.sourceforge.net/ 282programs: http://e2fsprogs.sourceforge.net/
259 http://ext2resize.sourceforge.net
260 283
261useful links: http://fedoraproject.org/wiki/ext3-devel 284useful links: http://fedoraproject.org/wiki/ext3-devel
262 http://www.bullopensource.org/ext4/ 285 http://www.bullopensource.org/ext4/
286 http://ext4.wiki.kernel.org/index.php/Main_Page
287 http://fedoraproject.org/wiki/Features/Ext4
diff --git a/fs/buffer.c b/fs/buffer.c
index 0f51c0f7c266..5fa1512cd9a2 100644
--- a/fs/buffer.c
+++ b/fs/buffer.c
@@ -1691,11 +1691,13 @@ static int __block_write_full_page(struct inode *inode, struct page *page,
1691 */ 1691 */
1692 clear_buffer_dirty(bh); 1692 clear_buffer_dirty(bh);
1693 set_buffer_uptodate(bh); 1693 set_buffer_uptodate(bh);
1694 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) { 1694 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1695 buffer_dirty(bh)) {
1695 WARN_ON(bh->b_size != blocksize); 1696 WARN_ON(bh->b_size != blocksize);
1696 err = get_block(inode, block, bh, 1); 1697 err = get_block(inode, block, bh, 1);
1697 if (err) 1698 if (err)
1698 goto recover; 1699 goto recover;
1700 clear_buffer_delay(bh);
1699 if (buffer_new(bh)) { 1701 if (buffer_new(bh)) {
1700 /* blockdev mappings never come here */ 1702 /* blockdev mappings never come here */
1701 clear_buffer_new(bh); 1703 clear_buffer_new(bh);
@@ -1774,7 +1776,8 @@ recover:
1774 bh = head; 1776 bh = head;
1775 /* Recovery: lock and submit the mapped buffers */ 1777 /* Recovery: lock and submit the mapped buffers */
1776 do { 1778 do {
1777 if (buffer_mapped(bh) && buffer_dirty(bh)) { 1779 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1780 !buffer_delay(bh)) {
1778 lock_buffer(bh); 1781 lock_buffer(bh);
1779 mark_buffer_async_write(bh); 1782 mark_buffer_async_write(bh);
1780 } else { 1783 } else {
@@ -2061,6 +2064,7 @@ int generic_write_end(struct file *file, struct address_space *mapping,
2061 struct page *page, void *fsdata) 2064 struct page *page, void *fsdata)
2062{ 2065{
2063 struct inode *inode = mapping->host; 2066 struct inode *inode = mapping->host;
2067 int i_size_changed = 0;
2064 2068
2065 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); 2069 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2066 2070
@@ -2073,12 +2077,21 @@ int generic_write_end(struct file *file, struct address_space *mapping,
2073 */ 2077 */
2074 if (pos+copied > inode->i_size) { 2078 if (pos+copied > inode->i_size) {
2075 i_size_write(inode, pos+copied); 2079 i_size_write(inode, pos+copied);
2076 mark_inode_dirty(inode); 2080 i_size_changed = 1;
2077 } 2081 }
2078 2082
2079 unlock_page(page); 2083 unlock_page(page);
2080 page_cache_release(page); 2084 page_cache_release(page);
2081 2085
2086 /*
2087 * Don't mark the inode dirty under page lock. First, it unnecessarily
2088 * makes the holding time of page lock longer. Second, it forces lock
2089 * ordering of page lock and transaction start for journaling
2090 * filesystems.
2091 */
2092 if (i_size_changed)
2093 mark_inode_dirty(inode);
2094
2082 return copied; 2095 return copied;
2083} 2096}
2084EXPORT_SYMBOL(generic_write_end); 2097EXPORT_SYMBOL(generic_write_end);
diff --git a/fs/ext4/balloc.c b/fs/ext4/balloc.c
index 9cc80b9cc8d8..495ab21b9832 100644
--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -47,7 +47,7 @@ static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
47 ext4_group_t block_group) 47 ext4_group_t block_group)
48{ 48{
49 ext4_group_t actual_group; 49 ext4_group_t actual_group;
50 ext4_get_group_no_and_offset(sb, block, &actual_group, 0); 50 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
51 if (actual_group == block_group) 51 if (actual_group == block_group)
52 return 1; 52 return 1;
53 return 0; 53 return 0;
@@ -121,12 +121,7 @@ unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
121 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); 121 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
122 } 122 }
123 } else { /* For META_BG_BLOCK_GROUPS */ 123 } else { /* For META_BG_BLOCK_GROUPS */
124 int group_rel = (block_group - 124 bit_max += ext4_bg_num_gdb(sb, block_group);
125 le32_to_cpu(sbi->s_es->s_first_meta_bg)) %
126 EXT4_DESC_PER_BLOCK(sb);
127 if (group_rel == 0 || group_rel == 1 ||
128 (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1))
129 bit_max += 1;
130 } 125 }
131 126
132 if (block_group == sbi->s_groups_count - 1) { 127 if (block_group == sbi->s_groups_count - 1) {
@@ -295,7 +290,7 @@ err_out:
295 return 0; 290 return 0;
296} 291}
297/** 292/**
298 * read_block_bitmap() 293 * ext4_read_block_bitmap()
299 * @sb: super block 294 * @sb: super block
300 * @block_group: given block group 295 * @block_group: given block group
301 * 296 *
@@ -305,7 +300,7 @@ err_out:
305 * Return buffer_head on success or NULL in case of failure. 300 * Return buffer_head on success or NULL in case of failure.
306 */ 301 */
307struct buffer_head * 302struct buffer_head *
308read_block_bitmap(struct super_block *sb, ext4_group_t block_group) 303ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
309{ 304{
310 struct ext4_group_desc * desc; 305 struct ext4_group_desc * desc;
311 struct buffer_head * bh = NULL; 306 struct buffer_head * bh = NULL;
@@ -409,8 +404,7 @@ restart:
409 prev = rsv; 404 prev = rsv;
410 } 405 }
411 printk("Window map complete.\n"); 406 printk("Window map complete.\n");
412 if (bad) 407 BUG_ON(bad);
413 BUG();
414} 408}
415#define rsv_window_dump(root, verbose) \ 409#define rsv_window_dump(root, verbose) \
416 __rsv_window_dump((root), (verbose), __func__) 410 __rsv_window_dump((root), (verbose), __func__)
@@ -694,7 +688,7 @@ do_more:
694 count -= overflow; 688 count -= overflow;
695 } 689 }
696 brelse(bitmap_bh); 690 brelse(bitmap_bh);
697 bitmap_bh = read_block_bitmap(sb, block_group); 691 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
698 if (!bitmap_bh) 692 if (!bitmap_bh)
699 goto error_return; 693 goto error_return;
700 desc = ext4_get_group_desc (sb, block_group, &gd_bh); 694 desc = ext4_get_group_desc (sb, block_group, &gd_bh);
@@ -810,6 +804,13 @@ do_more:
810 spin_unlock(sb_bgl_lock(sbi, block_group)); 804 spin_unlock(sb_bgl_lock(sbi, block_group));
811 percpu_counter_add(&sbi->s_freeblocks_counter, count); 805 percpu_counter_add(&sbi->s_freeblocks_counter, count);
812 806
807 if (sbi->s_log_groups_per_flex) {
808 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
809 spin_lock(sb_bgl_lock(sbi, flex_group));
810 sbi->s_flex_groups[flex_group].free_blocks += count;
811 spin_unlock(sb_bgl_lock(sbi, flex_group));
812 }
813
813 /* We dirtied the bitmap block */ 814 /* We dirtied the bitmap block */
814 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block"); 815 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
815 err = ext4_journal_dirty_metadata(handle, bitmap_bh); 816 err = ext4_journal_dirty_metadata(handle, bitmap_bh);
@@ -1598,23 +1599,35 @@ out:
1598 1599
1599/** 1600/**
1600 * ext4_has_free_blocks() 1601 * ext4_has_free_blocks()
1601 * @sbi: in-core super block structure. 1602 * @sbi: in-core super block structure.
1603 * @nblocks: number of neeed blocks
1602 * 1604 *
1603 * Check if filesystem has at least 1 free block available for allocation. 1605 * Check if filesystem has free blocks available for allocation.
1606 * Return the number of blocks avaible for allocation for this request
1607 * On success, return nblocks
1604 */ 1608 */
1605static int ext4_has_free_blocks(struct ext4_sb_info *sbi) 1609ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
1610 ext4_fsblk_t nblocks)
1606{ 1611{
1607 ext4_fsblk_t free_blocks, root_blocks; 1612 ext4_fsblk_t free_blocks;
1613 ext4_fsblk_t root_blocks = 0;
1608 1614
1609 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter); 1615 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
1610 root_blocks = ext4_r_blocks_count(sbi->s_es); 1616
1611 if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) && 1617 if (!capable(CAP_SYS_RESOURCE) &&
1612 sbi->s_resuid != current->fsuid && 1618 sbi->s_resuid != current->fsuid &&
1613 (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) { 1619 (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
1614 return 0; 1620 root_blocks = ext4_r_blocks_count(sbi->s_es);
1615 } 1621#ifdef CONFIG_SMP
1616 return 1; 1622 if (free_blocks - root_blocks < FBC_BATCH)
1617} 1623 free_blocks =
1624 percpu_counter_sum_and_set(&sbi->s_freeblocks_counter);
1625#endif
1626 if (free_blocks - root_blocks < nblocks)
1627 return free_blocks - root_blocks;
1628 return nblocks;
1629 }
1630
1618 1631
1619/** 1632/**
1620 * ext4_should_retry_alloc() 1633 * ext4_should_retry_alloc()
@@ -1630,7 +1643,7 @@ static int ext4_has_free_blocks(struct ext4_sb_info *sbi)
1630 */ 1643 */
1631int ext4_should_retry_alloc(struct super_block *sb, int *retries) 1644int ext4_should_retry_alloc(struct super_block *sb, int *retries)
1632{ 1645{
1633 if (!ext4_has_free_blocks(EXT4_SB(sb)) || (*retries)++ > 3) 1646 if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3)
1634 return 0; 1647 return 0;
1635 1648
1636 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); 1649 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
@@ -1639,20 +1652,24 @@ int ext4_should_retry_alloc(struct super_block *sb, int *retries)
1639} 1652}
1640 1653
1641/** 1654/**
1642 * ext4_new_blocks_old() -- core block(s) allocation function 1655 * ext4_old_new_blocks() -- core block bitmap based block allocation function
1656 *
1643 * @handle: handle to this transaction 1657 * @handle: handle to this transaction
1644 * @inode: file inode 1658 * @inode: file inode
1645 * @goal: given target block(filesystem wide) 1659 * @goal: given target block(filesystem wide)
1646 * @count: target number of blocks to allocate 1660 * @count: target number of blocks to allocate
1647 * @errp: error code 1661 * @errp: error code
1648 * 1662 *
1649 * ext4_new_blocks uses a goal block to assist allocation. It tries to 1663 * ext4_old_new_blocks uses a goal block to assist allocation and look up
1650 * allocate block(s) from the block group contains the goal block first. If that 1664 * the block bitmap directly to do block allocation. It tries to
1651 * fails, it will try to allocate block(s) from other block groups without 1665 * allocate block(s) from the block group contains the goal block first. If
1652 * any specific goal block. 1666 * that fails, it will try to allocate block(s) from other block groups
1667 * without any specific goal block.
1668 *
1669 * This function is called when -o nomballoc mount option is enabled
1653 * 1670 *
1654 */ 1671 */
1655ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode, 1672ext4_fsblk_t ext4_old_new_blocks(handle_t *handle, struct inode *inode,
1656 ext4_fsblk_t goal, unsigned long *count, int *errp) 1673 ext4_fsblk_t goal, unsigned long *count, int *errp)
1657{ 1674{
1658 struct buffer_head *bitmap_bh = NULL; 1675 struct buffer_head *bitmap_bh = NULL;
@@ -1676,13 +1693,26 @@ ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode,
1676 ext4_group_t ngroups; 1693 ext4_group_t ngroups;
1677 unsigned long num = *count; 1694 unsigned long num = *count;
1678 1695
1679 *errp = -ENOSPC;
1680 sb = inode->i_sb; 1696 sb = inode->i_sb;
1681 if (!sb) { 1697 if (!sb) {
1698 *errp = -ENODEV;
1682 printk("ext4_new_block: nonexistent device"); 1699 printk("ext4_new_block: nonexistent device");
1683 return 0; 1700 return 0;
1684 } 1701 }
1685 1702
1703 sbi = EXT4_SB(sb);
1704 if (!EXT4_I(inode)->i_delalloc_reserved_flag) {
1705 /*
1706 * With delalloc we already reserved the blocks
1707 */
1708 *count = ext4_has_free_blocks(sbi, *count);
1709 }
1710 if (*count == 0) {
1711 *errp = -ENOSPC;
1712 return 0; /*return with ENOSPC error */
1713 }
1714 num = *count;
1715
1686 /* 1716 /*
1687 * Check quota for allocation of this block. 1717 * Check quota for allocation of this block.
1688 */ 1718 */
@@ -1706,11 +1736,6 @@ ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode,
1706 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0)) 1736 if (block_i && ((windowsz = block_i->rsv_window_node.rsv_goal_size) > 0))
1707 my_rsv = &block_i->rsv_window_node; 1737 my_rsv = &block_i->rsv_window_node;
1708 1738
1709 if (!ext4_has_free_blocks(sbi)) {
1710 *errp = -ENOSPC;
1711 goto out;
1712 }
1713
1714 /* 1739 /*
1715 * First, test whether the goal block is free. 1740 * First, test whether the goal block is free.
1716 */ 1741 */
@@ -1734,7 +1759,7 @@ retry_alloc:
1734 my_rsv = NULL; 1759 my_rsv = NULL;
1735 1760
1736 if (free_blocks > 0) { 1761 if (free_blocks > 0) {
1737 bitmap_bh = read_block_bitmap(sb, group_no); 1762 bitmap_bh = ext4_read_block_bitmap(sb, group_no);
1738 if (!bitmap_bh) 1763 if (!bitmap_bh)
1739 goto io_error; 1764 goto io_error;
1740 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle, 1765 grp_alloc_blk = ext4_try_to_allocate_with_rsv(sb, handle,
@@ -1770,7 +1795,7 @@ retry_alloc:
1770 continue; 1795 continue;
1771 1796
1772 brelse(bitmap_bh); 1797 brelse(bitmap_bh);
1773 bitmap_bh = read_block_bitmap(sb, group_no); 1798 bitmap_bh = ext4_read_block_bitmap(sb, group_no);
1774 if (!bitmap_bh) 1799 if (!bitmap_bh)
1775 goto io_error; 1800 goto io_error;
1776 /* 1801 /*
@@ -1882,7 +1907,15 @@ allocated:
1882 le16_add_cpu(&gdp->bg_free_blocks_count, -num); 1907 le16_add_cpu(&gdp->bg_free_blocks_count, -num);
1883 gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp); 1908 gdp->bg_checksum = ext4_group_desc_csum(sbi, group_no, gdp);
1884 spin_unlock(sb_bgl_lock(sbi, group_no)); 1909 spin_unlock(sb_bgl_lock(sbi, group_no));
1885 percpu_counter_sub(&sbi->s_freeblocks_counter, num); 1910 if (!EXT4_I(inode)->i_delalloc_reserved_flag)
1911 percpu_counter_sub(&sbi->s_freeblocks_counter, num);
1912
1913 if (sbi->s_log_groups_per_flex) {
1914 ext4_group_t flex_group = ext4_flex_group(sbi, group_no);
1915 spin_lock(sb_bgl_lock(sbi, flex_group));
1916 sbi->s_flex_groups[flex_group].free_blocks -= num;
1917 spin_unlock(sb_bgl_lock(sbi, flex_group));
1918 }
1886 1919
1887 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor"); 1920 BUFFER_TRACE(gdp_bh, "journal_dirty_metadata for group descriptor");
1888 err = ext4_journal_dirty_metadata(handle, gdp_bh); 1921 err = ext4_journal_dirty_metadata(handle, gdp_bh);
@@ -1915,46 +1948,104 @@ out:
1915 return 0; 1948 return 0;
1916} 1949}
1917 1950
1918ext4_fsblk_t ext4_new_block(handle_t *handle, struct inode *inode, 1951#define EXT4_META_BLOCK 0x1
1919 ext4_fsblk_t goal, int *errp) 1952
1953static ext4_fsblk_t do_blk_alloc(handle_t *handle, struct inode *inode,
1954 ext4_lblk_t iblock, ext4_fsblk_t goal,
1955 unsigned long *count, int *errp, int flags)
1920{ 1956{
1921 struct ext4_allocation_request ar; 1957 struct ext4_allocation_request ar;
1922 ext4_fsblk_t ret; 1958 ext4_fsblk_t ret;
1923 1959
1924 if (!test_opt(inode->i_sb, MBALLOC)) { 1960 if (!test_opt(inode->i_sb, MBALLOC)) {
1925 unsigned long count = 1; 1961 return ext4_old_new_blocks(handle, inode, goal, count, errp);
1926 ret = ext4_new_blocks_old(handle, inode, goal, &count, errp);
1927 return ret;
1928 } 1962 }
1929 1963
1930 memset(&ar, 0, sizeof(ar)); 1964 memset(&ar, 0, sizeof(ar));
1965 /* Fill with neighbour allocated blocks */
1966
1931 ar.inode = inode; 1967 ar.inode = inode;
1932 ar.goal = goal; 1968 ar.goal = goal;
1933 ar.len = 1; 1969 ar.len = *count;
1970 ar.logical = iblock;
1971
1972 if (S_ISREG(inode->i_mode) && !(flags & EXT4_META_BLOCK))
1973 /* enable in-core preallocation for data block allocation */
1974 ar.flags = EXT4_MB_HINT_DATA;
1975 else
1976 /* disable in-core preallocation for non-regular files */
1977 ar.flags = 0;
1978
1934 ret = ext4_mb_new_blocks(handle, &ar, errp); 1979 ret = ext4_mb_new_blocks(handle, &ar, errp);
1980 *count = ar.len;
1935 return ret; 1981 return ret;
1936} 1982}
1937 1983
1938ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode, 1984/*
1985 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
1986 *
1987 * @handle: handle to this transaction
1988 * @inode: file inode
1989 * @goal: given target block(filesystem wide)
1990 * @count: total number of blocks need
1991 * @errp: error code
1992 *
1993 * Return 1st allocated block numberon success, *count stores total account
1994 * error stores in errp pointer
1995 */
1996ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
1939 ext4_fsblk_t goal, unsigned long *count, int *errp) 1997 ext4_fsblk_t goal, unsigned long *count, int *errp)
1940{ 1998{
1941 struct ext4_allocation_request ar;
1942 ext4_fsblk_t ret; 1999 ext4_fsblk_t ret;
1943 2000 ret = do_blk_alloc(handle, inode, 0, goal,
1944 if (!test_opt(inode->i_sb, MBALLOC)) { 2001 count, errp, EXT4_META_BLOCK);
1945 ret = ext4_new_blocks_old(handle, inode, goal, count, errp); 2002 /*
1946 return ret; 2003 * Account for the allocated meta blocks
2004 */
2005 if (!(*errp)) {
2006 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
2007 EXT4_I(inode)->i_allocated_meta_blocks += *count;
2008 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1947 } 2009 }
1948
1949 memset(&ar, 0, sizeof(ar));
1950 ar.inode = inode;
1951 ar.goal = goal;
1952 ar.len = *count;
1953 ret = ext4_mb_new_blocks(handle, &ar, errp);
1954 *count = ar.len;
1955 return ret; 2010 return ret;
1956} 2011}
1957 2012
2013/*
2014 * ext4_new_meta_block() -- allocate block for meta data (indexing) blocks
2015 *
2016 * @handle: handle to this transaction
2017 * @inode: file inode
2018 * @goal: given target block(filesystem wide)
2019 * @errp: error code
2020 *
2021 * Return allocated block number on success
2022 */
2023ext4_fsblk_t ext4_new_meta_block(handle_t *handle, struct inode *inode,
2024 ext4_fsblk_t goal, int *errp)
2025{
2026 unsigned long count = 1;
2027 return ext4_new_meta_blocks(handle, inode, goal, &count, errp);
2028}
2029
2030/*
2031 * ext4_new_blocks() -- allocate data blocks
2032 *
2033 * @handle: handle to this transaction
2034 * @inode: file inode
2035 * @goal: given target block(filesystem wide)
2036 * @count: total number of blocks need
2037 * @errp: error code
2038 *
2039 * Return 1st allocated block numberon success, *count stores total account
2040 * error stores in errp pointer
2041 */
2042
2043ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
2044 ext4_lblk_t iblock, ext4_fsblk_t goal,
2045 unsigned long *count, int *errp)
2046{
2047 return do_blk_alloc(handle, inode, iblock, goal, count, errp, 0);
2048}
1958 2049
1959/** 2050/**
1960 * ext4_count_free_blocks() -- count filesystem free blocks 2051 * ext4_count_free_blocks() -- count filesystem free blocks
@@ -1986,7 +2077,7 @@ ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
1986 continue; 2077 continue;
1987 desc_count += le16_to_cpu(gdp->bg_free_blocks_count); 2078 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
1988 brelse(bitmap_bh); 2079 brelse(bitmap_bh);
1989 bitmap_bh = read_block_bitmap(sb, i); 2080 bitmap_bh = ext4_read_block_bitmap(sb, i);
1990 if (bitmap_bh == NULL) 2081 if (bitmap_bh == NULL)
1991 continue; 2082 continue;
1992 2083
diff --git a/fs/ext4/dir.c b/fs/ext4/dir.c
index 2bf0331ea194..d3d23d73c08b 100644
--- a/fs/ext4/dir.c
+++ b/fs/ext4/dir.c
@@ -129,7 +129,8 @@ static int ext4_readdir(struct file * filp,
129 struct buffer_head *bh = NULL; 129 struct buffer_head *bh = NULL;
130 130
131 map_bh.b_state = 0; 131 map_bh.b_state = 0;
132 err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh, 0, 0); 132 err = ext4_get_blocks_wrap(NULL, inode, blk, 1, &map_bh,
133 0, 0, 0);
133 if (err > 0) { 134 if (err > 0) {
134 pgoff_t index = map_bh.b_blocknr >> 135 pgoff_t index = map_bh.b_blocknr >>
135 (PAGE_CACHE_SHIFT - inode->i_blkbits); 136 (PAGE_CACHE_SHIFT - inode->i_blkbits);
@@ -272,7 +273,7 @@ static void free_rb_tree_fname(struct rb_root *root)
272 273
273 while (n) { 274 while (n) {
274 /* Do the node's children first */ 275 /* Do the node's children first */
275 if ((n)->rb_left) { 276 if (n->rb_left) {
276 n = n->rb_left; 277 n = n->rb_left;
277 continue; 278 continue;
278 } 279 }
@@ -301,24 +302,18 @@ static void free_rb_tree_fname(struct rb_root *root)
301 parent->rb_right = NULL; 302 parent->rb_right = NULL;
302 n = parent; 303 n = parent;
303 } 304 }
304 root->rb_node = NULL;
305} 305}
306 306
307 307
308static struct dir_private_info *create_dir_info(loff_t pos) 308static struct dir_private_info *ext4_htree_create_dir_info(loff_t pos)
309{ 309{
310 struct dir_private_info *p; 310 struct dir_private_info *p;
311 311
312 p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL); 312 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
313 if (!p) 313 if (!p)
314 return NULL; 314 return NULL;
315 p->root.rb_node = NULL;
316 p->curr_node = NULL;
317 p->extra_fname = NULL;
318 p->last_pos = 0;
319 p->curr_hash = pos2maj_hash(pos); 315 p->curr_hash = pos2maj_hash(pos);
320 p->curr_minor_hash = pos2min_hash(pos); 316 p->curr_minor_hash = pos2min_hash(pos);
321 p->next_hash = 0;
322 return p; 317 return p;
323} 318}
324 319
@@ -433,7 +428,7 @@ static int ext4_dx_readdir(struct file * filp,
433 int ret; 428 int ret;
434 429
435 if (!info) { 430 if (!info) {
436 info = create_dir_info(filp->f_pos); 431 info = ext4_htree_create_dir_info(filp->f_pos);
437 if (!info) 432 if (!info)
438 return -ENOMEM; 433 return -ENOMEM;
439 filp->private_data = info; 434 filp->private_data = info;
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 8158083f7ac0..303e41cf7b14 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -22,7 +22,7 @@
22#include "ext4_i.h" 22#include "ext4_i.h"
23 23
24/* 24/*
25 * The second extended filesystem constants/structures 25 * The fourth extended filesystem constants/structures
26 */ 26 */
27 27
28/* 28/*
@@ -45,7 +45,7 @@
45#define ext4_debug(f, a...) \ 45#define ext4_debug(f, a...) \
46 do { \ 46 do { \
47 printk (KERN_DEBUG "EXT4-fs DEBUG (%s, %d): %s:", \ 47 printk (KERN_DEBUG "EXT4-fs DEBUG (%s, %d): %s:", \
48 __FILE__, __LINE__, __FUNCTION__); \ 48 __FILE__, __LINE__, __func__); \
49 printk (KERN_DEBUG f, ## a); \ 49 printk (KERN_DEBUG f, ## a); \
50 } while (0) 50 } while (0)
51#else 51#else
@@ -74,6 +74,9 @@
74#define EXT4_MB_HINT_GOAL_ONLY 256 74#define EXT4_MB_HINT_GOAL_ONLY 256
75/* goal is meaningful */ 75/* goal is meaningful */
76#define EXT4_MB_HINT_TRY_GOAL 512 76#define EXT4_MB_HINT_TRY_GOAL 512
77/* blocks already pre-reserved by delayed allocation */
78#define EXT4_MB_DELALLOC_RESERVED 1024
79
77 80
78struct ext4_allocation_request { 81struct ext4_allocation_request {
79 /* target inode for block we're allocating */ 82 /* target inode for block we're allocating */
@@ -170,6 +173,15 @@ struct ext4_group_desc
170 __u32 bg_reserved2[3]; 173 __u32 bg_reserved2[3];
171}; 174};
172 175
176/*
177 * Structure of a flex block group info
178 */
179
180struct flex_groups {
181 __u32 free_inodes;
182 __u32 free_blocks;
183};
184
173#define EXT4_BG_INODE_UNINIT 0x0001 /* Inode table/bitmap not in use */ 185#define EXT4_BG_INODE_UNINIT 0x0001 /* Inode table/bitmap not in use */
174#define EXT4_BG_BLOCK_UNINIT 0x0002 /* Block bitmap not in use */ 186#define EXT4_BG_BLOCK_UNINIT 0x0002 /* Block bitmap not in use */
175#define EXT4_BG_INODE_ZEROED 0x0004 /* On-disk itable initialized to zero */ 187#define EXT4_BG_INODE_ZEROED 0x0004 /* On-disk itable initialized to zero */
@@ -527,6 +539,7 @@ do { \
527#define EXT4_MOUNT_JOURNAL_ASYNC_COMMIT 0x1000000 /* Journal Async Commit */ 539#define EXT4_MOUNT_JOURNAL_ASYNC_COMMIT 0x1000000 /* Journal Async Commit */
528#define EXT4_MOUNT_I_VERSION 0x2000000 /* i_version support */ 540#define EXT4_MOUNT_I_VERSION 0x2000000 /* i_version support */
529#define EXT4_MOUNT_MBALLOC 0x4000000 /* Buddy allocation support */ 541#define EXT4_MOUNT_MBALLOC 0x4000000 /* Buddy allocation support */
542#define EXT4_MOUNT_DELALLOC 0x8000000 /* Delalloc support */
530/* Compatibility, for having both ext2_fs.h and ext4_fs.h included at once */ 543/* Compatibility, for having both ext2_fs.h and ext4_fs.h included at once */
531#ifndef _LINUX_EXT2_FS_H 544#ifndef _LINUX_EXT2_FS_H
532#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt 545#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt
@@ -647,7 +660,10 @@ struct ext4_super_block {
647 __le16 s_mmp_interval; /* # seconds to wait in MMP checking */ 660 __le16 s_mmp_interval; /* # seconds to wait in MMP checking */
648 __le64 s_mmp_block; /* Block for multi-mount protection */ 661 __le64 s_mmp_block; /* Block for multi-mount protection */
649 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/ 662 __le32 s_raid_stripe_width; /* blocks on all data disks (N*stride)*/
650 __u32 s_reserved[163]; /* Padding to the end of the block */ 663 __u8 s_log_groups_per_flex; /* FLEX_BG group size */
664 __u8 s_reserved_char_pad2;
665 __le16 s_reserved_pad;
666 __u32 s_reserved[162]; /* Padding to the end of the block */
651}; 667};
652 668
653#ifdef __KERNEL__ 669#ifdef __KERNEL__
@@ -958,12 +974,17 @@ extern ext4_grpblk_t ext4_block_group_offset(struct super_block *sb,
958extern int ext4_bg_has_super(struct super_block *sb, ext4_group_t group); 974extern int ext4_bg_has_super(struct super_block *sb, ext4_group_t group);
959extern unsigned long ext4_bg_num_gdb(struct super_block *sb, 975extern unsigned long ext4_bg_num_gdb(struct super_block *sb,
960 ext4_group_t group); 976 ext4_group_t group);
961extern ext4_fsblk_t ext4_new_block (handle_t *handle, struct inode *inode, 977extern ext4_fsblk_t ext4_new_meta_block(handle_t *handle, struct inode *inode,
962 ext4_fsblk_t goal, int *errp); 978 ext4_fsblk_t goal, int *errp);
963extern ext4_fsblk_t ext4_new_blocks (handle_t *handle, struct inode *inode, 979extern ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
964 ext4_fsblk_t goal, unsigned long *count, int *errp); 980 ext4_fsblk_t goal, unsigned long *count, int *errp);
965extern ext4_fsblk_t ext4_new_blocks_old(handle_t *handle, struct inode *inode, 981extern ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
982 ext4_lblk_t iblock, ext4_fsblk_t goal,
983 unsigned long *count, int *errp);
984extern ext4_fsblk_t ext4_old_new_blocks(handle_t *handle, struct inode *inode,
966 ext4_fsblk_t goal, unsigned long *count, int *errp); 985 ext4_fsblk_t goal, unsigned long *count, int *errp);
986extern ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
987 ext4_fsblk_t nblocks);
967extern void ext4_free_blocks (handle_t *handle, struct inode *inode, 988extern void ext4_free_blocks (handle_t *handle, struct inode *inode,
968 ext4_fsblk_t block, unsigned long count, int metadata); 989 ext4_fsblk_t block, unsigned long count, int metadata);
969extern void ext4_free_blocks_sb (handle_t *handle, struct super_block *sb, 990extern void ext4_free_blocks_sb (handle_t *handle, struct super_block *sb,
@@ -1016,9 +1037,14 @@ extern int __init init_ext4_mballoc(void);
1016extern void exit_ext4_mballoc(void); 1037extern void exit_ext4_mballoc(void);
1017extern void ext4_mb_free_blocks(handle_t *, struct inode *, 1038extern void ext4_mb_free_blocks(handle_t *, struct inode *,
1018 unsigned long, unsigned long, int, unsigned long *); 1039 unsigned long, unsigned long, int, unsigned long *);
1040extern int ext4_mb_add_more_groupinfo(struct super_block *sb,
1041 ext4_group_t i, struct ext4_group_desc *desc);
1042extern void ext4_mb_update_group_info(struct ext4_group_info *grp,
1043 ext4_grpblk_t add);
1019 1044
1020 1045
1021/* inode.c */ 1046/* inode.c */
1047void ext4_da_release_space(struct inode *inode, int used, int to_free);
1022int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, 1048int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
1023 struct buffer_head *bh, ext4_fsblk_t blocknr); 1049 struct buffer_head *bh, ext4_fsblk_t blocknr);
1024struct buffer_head *ext4_getblk(handle_t *, struct inode *, 1050struct buffer_head *ext4_getblk(handle_t *, struct inode *,
@@ -1033,19 +1059,23 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
1033extern struct inode *ext4_iget(struct super_block *, unsigned long); 1059extern struct inode *ext4_iget(struct super_block *, unsigned long);
1034extern int ext4_write_inode (struct inode *, int); 1060extern int ext4_write_inode (struct inode *, int);
1035extern int ext4_setattr (struct dentry *, struct iattr *); 1061extern int ext4_setattr (struct dentry *, struct iattr *);
1062extern int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
1063 struct kstat *stat);
1036extern void ext4_delete_inode (struct inode *); 1064extern void ext4_delete_inode (struct inode *);
1037extern int ext4_sync_inode (handle_t *, struct inode *); 1065extern int ext4_sync_inode (handle_t *, struct inode *);
1038extern void ext4_discard_reservation (struct inode *); 1066extern void ext4_discard_reservation (struct inode *);
1039extern void ext4_dirty_inode(struct inode *); 1067extern void ext4_dirty_inode(struct inode *);
1040extern int ext4_change_inode_journal_flag(struct inode *, int); 1068extern int ext4_change_inode_journal_flag(struct inode *, int);
1041extern int ext4_get_inode_loc(struct inode *, struct ext4_iloc *); 1069extern int ext4_get_inode_loc(struct inode *, struct ext4_iloc *);
1070extern int ext4_can_truncate(struct inode *inode);
1042extern void ext4_truncate (struct inode *); 1071extern void ext4_truncate (struct inode *);
1043extern void ext4_set_inode_flags(struct inode *); 1072extern void ext4_set_inode_flags(struct inode *);
1044extern void ext4_get_inode_flags(struct ext4_inode_info *); 1073extern void ext4_get_inode_flags(struct ext4_inode_info *);
1045extern void ext4_set_aops(struct inode *inode); 1074extern void ext4_set_aops(struct inode *inode);
1046extern int ext4_writepage_trans_blocks(struct inode *); 1075extern int ext4_writepage_trans_blocks(struct inode *);
1047extern int ext4_block_truncate_page(handle_t *handle, struct page *page, 1076extern int ext4_block_truncate_page(handle_t *handle,
1048 struct address_space *mapping, loff_t from); 1077 struct address_space *mapping, loff_t from);
1078extern int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page);
1049 1079
1050/* ioctl.c */ 1080/* ioctl.c */
1051extern long ext4_ioctl(struct file *, unsigned int, unsigned long); 1081extern long ext4_ioctl(struct file *, unsigned int, unsigned long);
@@ -1159,10 +1189,21 @@ struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
1159} 1189}
1160 1190
1161 1191
1192static inline ext4_group_t ext4_flex_group(struct ext4_sb_info *sbi,
1193 ext4_group_t block_group)
1194{
1195 return block_group >> sbi->s_log_groups_per_flex;
1196}
1197
1198static inline unsigned int ext4_flex_bg_size(struct ext4_sb_info *sbi)
1199{
1200 return 1 << sbi->s_log_groups_per_flex;
1201}
1202
1162#define ext4_std_error(sb, errno) \ 1203#define ext4_std_error(sb, errno) \
1163do { \ 1204do { \
1164 if ((errno)) \ 1205 if ((errno)) \
1165 __ext4_std_error((sb), __FUNCTION__, (errno)); \ 1206 __ext4_std_error((sb), __func__, (errno)); \
1166} while (0) 1207} while (0)
1167 1208
1168/* 1209/*
@@ -1191,7 +1232,7 @@ extern int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
1191 ext4_lblk_t iblock, 1232 ext4_lblk_t iblock,
1192 unsigned long max_blocks, struct buffer_head *bh_result, 1233 unsigned long max_blocks, struct buffer_head *bh_result,
1193 int create, int extend_disksize); 1234 int create, int extend_disksize);
1194extern void ext4_ext_truncate(struct inode *, struct page *); 1235extern void ext4_ext_truncate(struct inode *);
1195extern void ext4_ext_init(struct super_block *); 1236extern void ext4_ext_init(struct super_block *);
1196extern void ext4_ext_release(struct super_block *); 1237extern void ext4_ext_release(struct super_block *);
1197extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset, 1238extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset,
@@ -1199,7 +1240,7 @@ extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset,
1199extern int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, 1240extern int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode,
1200 sector_t block, unsigned long max_blocks, 1241 sector_t block, unsigned long max_blocks,
1201 struct buffer_head *bh, int create, 1242 struct buffer_head *bh, int create,
1202 int extend_disksize); 1243 int extend_disksize, int flag);
1203#endif /* __KERNEL__ */ 1244#endif /* __KERNEL__ */
1204 1245
1205#endif /* _EXT4_H */ 1246#endif /* _EXT4_H */
diff --git a/fs/ext4/ext4_extents.h b/fs/ext4/ext4_extents.h
index 75333b595fab..6c166c0a54b7 100644
--- a/fs/ext4/ext4_extents.h
+++ b/fs/ext4/ext4_extents.h
@@ -212,6 +212,7 @@ static inline int ext4_ext_get_actual_len(struct ext4_extent *ext)
212 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN)); 212 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN));
213} 213}
214 214
215extern int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks);
215extern ext4_fsblk_t idx_pblock(struct ext4_extent_idx *); 216extern ext4_fsblk_t idx_pblock(struct ext4_extent_idx *);
216extern void ext4_ext_store_pblock(struct ext4_extent *, ext4_fsblk_t); 217extern void ext4_ext_store_pblock(struct ext4_extent *, ext4_fsblk_t);
217extern int ext4_extent_tree_init(handle_t *, struct inode *); 218extern int ext4_extent_tree_init(handle_t *, struct inode *);
diff --git a/fs/ext4/ext4_i.h b/fs/ext4/ext4_i.h
index 26a4ae255d79..ef7409f0e7e4 100644
--- a/fs/ext4/ext4_i.h
+++ b/fs/ext4/ext4_i.h
@@ -79,7 +79,7 @@ struct ext4_ext_cache {
79}; 79};
80 80
81/* 81/*
82 * third extended file system inode data in memory 82 * fourth extended file system inode data in memory
83 */ 83 */
84struct ext4_inode_info { 84struct ext4_inode_info {
85 __le32 i_data[15]; /* unconverted */ 85 __le32 i_data[15]; /* unconverted */
@@ -150,6 +150,7 @@ struct ext4_inode_info {
150 */ 150 */
151 struct rw_semaphore i_data_sem; 151 struct rw_semaphore i_data_sem;
152 struct inode vfs_inode; 152 struct inode vfs_inode;
153 struct jbd2_inode jinode;
153 154
154 unsigned long i_ext_generation; 155 unsigned long i_ext_generation;
155 struct ext4_ext_cache i_cached_extent; 156 struct ext4_ext_cache i_cached_extent;
@@ -162,6 +163,13 @@ struct ext4_inode_info {
162 /* mballoc */ 163 /* mballoc */
163 struct list_head i_prealloc_list; 164 struct list_head i_prealloc_list;
164 spinlock_t i_prealloc_lock; 165 spinlock_t i_prealloc_lock;
166
167 /* allocation reservation info for delalloc */
168 unsigned long i_reserved_data_blocks;
169 unsigned long i_reserved_meta_blocks;
170 unsigned long i_allocated_meta_blocks;
171 unsigned short i_delalloc_reserved_flag;
172 spinlock_t i_block_reservation_lock;
165}; 173};
166 174
167#endif /* _EXT4_I */ 175#endif /* _EXT4_I */
diff --git a/fs/ext4/ext4_jbd2.h b/fs/ext4/ext4_jbd2.h
index 9255a7d28b24..eb8bc3afe6e9 100644
--- a/fs/ext4/ext4_jbd2.h
+++ b/fs/ext4/ext4_jbd2.h
@@ -142,19 +142,17 @@ int __ext4_journal_dirty_metadata(const char *where,
142 handle_t *handle, struct buffer_head *bh); 142 handle_t *handle, struct buffer_head *bh);
143 143
144#define ext4_journal_get_undo_access(handle, bh) \ 144#define ext4_journal_get_undo_access(handle, bh) \
145 __ext4_journal_get_undo_access(__FUNCTION__, (handle), (bh)) 145 __ext4_journal_get_undo_access(__func__, (handle), (bh))
146#define ext4_journal_get_write_access(handle, bh) \ 146#define ext4_journal_get_write_access(handle, bh) \
147 __ext4_journal_get_write_access(__FUNCTION__, (handle), (bh)) 147 __ext4_journal_get_write_access(__func__, (handle), (bh))
148#define ext4_journal_revoke(handle, blocknr, bh) \ 148#define ext4_journal_revoke(handle, blocknr, bh) \
149 __ext4_journal_revoke(__FUNCTION__, (handle), (blocknr), (bh)) 149 __ext4_journal_revoke(__func__, (handle), (blocknr), (bh))
150#define ext4_journal_get_create_access(handle, bh) \ 150#define ext4_journal_get_create_access(handle, bh) \
151 __ext4_journal_get_create_access(__FUNCTION__, (handle), (bh)) 151 __ext4_journal_get_create_access(__func__, (handle), (bh))
152#define ext4_journal_dirty_metadata(handle, bh) \ 152#define ext4_journal_dirty_metadata(handle, bh) \
153 __ext4_journal_dirty_metadata(__FUNCTION__, (handle), (bh)) 153 __ext4_journal_dirty_metadata(__func__, (handle), (bh))
154#define ext4_journal_forget(handle, bh) \ 154#define ext4_journal_forget(handle, bh) \
155 __ext4_journal_forget(__FUNCTION__, (handle), (bh)) 155 __ext4_journal_forget(__func__, (handle), (bh))
156
157int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh);
158 156
159handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks); 157handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks);
160int __ext4_journal_stop(const char *where, handle_t *handle); 158int __ext4_journal_stop(const char *where, handle_t *handle);
@@ -165,7 +163,7 @@ static inline handle_t *ext4_journal_start(struct inode *inode, int nblocks)
165} 163}
166 164
167#define ext4_journal_stop(handle) \ 165#define ext4_journal_stop(handle) \
168 __ext4_journal_stop(__FUNCTION__, (handle)) 166 __ext4_journal_stop(__func__, (handle))
169 167
170static inline handle_t *ext4_journal_current_handle(void) 168static inline handle_t *ext4_journal_current_handle(void)
171{ 169{
@@ -192,6 +190,11 @@ static inline int ext4_journal_force_commit(journal_t *journal)
192 return jbd2_journal_force_commit(journal); 190 return jbd2_journal_force_commit(journal);
193} 191}
194 192
193static inline int ext4_jbd2_file_inode(handle_t *handle, struct inode *inode)
194{
195 return jbd2_journal_file_inode(handle, &EXT4_I(inode)->jinode);
196}
197
195/* super.c */ 198/* super.c */
196int ext4_force_commit(struct super_block *sb); 199int ext4_force_commit(struct super_block *sb);
197 200
diff --git a/fs/ext4/ext4_sb.h b/fs/ext4/ext4_sb.h
index 5802e69f2191..6300226d5531 100644
--- a/fs/ext4/ext4_sb.h
+++ b/fs/ext4/ext4_sb.h
@@ -25,7 +25,7 @@
25#include <linux/rbtree.h> 25#include <linux/rbtree.h>
26 26
27/* 27/*
28 * third extended-fs super-block data in memory 28 * fourth extended-fs super-block data in memory
29 */ 29 */
30struct ext4_sb_info { 30struct ext4_sb_info {
31 unsigned long s_desc_size; /* Size of a group descriptor in bytes */ 31 unsigned long s_desc_size; /* Size of a group descriptor in bytes */
@@ -143,6 +143,9 @@ struct ext4_sb_info {
143 143
144 /* locality groups */ 144 /* locality groups */
145 struct ext4_locality_group *s_locality_groups; 145 struct ext4_locality_group *s_locality_groups;
146
147 unsigned int s_log_groups_per_flex;
148 struct flex_groups *s_flex_groups;
146}; 149};
147 150
148#endif /* _EXT4_SB */ 151#endif /* _EXT4_SB */
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 47929c4e3dae..42c4c0c892ed 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -92,17 +92,16 @@ static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
92 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff); 92 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
93} 93}
94 94
95static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed) 95static int ext4_ext_journal_restart(handle_t *handle, int needed)
96{ 96{
97 int err; 97 int err;
98 98
99 if (handle->h_buffer_credits > needed) 99 if (handle->h_buffer_credits > needed)
100 return handle; 100 return 0;
101 if (!ext4_journal_extend(handle, needed)) 101 err = ext4_journal_extend(handle, needed);
102 return handle; 102 if (err)
103 err = ext4_journal_restart(handle, needed); 103 return err;
104 104 return ext4_journal_restart(handle, needed);
105 return handle;
106} 105}
107 106
108/* 107/*
@@ -180,15 +179,18 @@ static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
180 return bg_start + colour + block; 179 return bg_start + colour + block;
181} 180}
182 181
182/*
183 * Allocation for a meta data block
184 */
183static ext4_fsblk_t 185static ext4_fsblk_t
184ext4_ext_new_block(handle_t *handle, struct inode *inode, 186ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
185 struct ext4_ext_path *path, 187 struct ext4_ext_path *path,
186 struct ext4_extent *ex, int *err) 188 struct ext4_extent *ex, int *err)
187{ 189{
188 ext4_fsblk_t goal, newblock; 190 ext4_fsblk_t goal, newblock;
189 191
190 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block)); 192 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
191 newblock = ext4_new_block(handle, inode, goal, err); 193 newblock = ext4_new_meta_block(handle, inode, goal, err);
192 return newblock; 194 return newblock;
193} 195}
194 196
@@ -246,6 +248,36 @@ static int ext4_ext_space_root_idx(struct inode *inode)
246 return size; 248 return size;
247} 249}
248 250
251/*
252 * Calculate the number of metadata blocks needed
253 * to allocate @blocks
254 * Worse case is one block per extent
255 */
256int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
257{
258 int lcap, icap, rcap, leafs, idxs, num;
259 int newextents = blocks;
260
261 rcap = ext4_ext_space_root_idx(inode);
262 lcap = ext4_ext_space_block(inode);
263 icap = ext4_ext_space_block_idx(inode);
264
265 /* number of new leaf blocks needed */
266 num = leafs = (newextents + lcap - 1) / lcap;
267
268 /*
269 * Worse case, we need separate index block(s)
270 * to link all new leaf blocks
271 */
272 idxs = (leafs + icap - 1) / icap;
273 do {
274 num += idxs;
275 idxs = (idxs + icap - 1) / icap;
276 } while (idxs > rcap);
277
278 return num;
279}
280
249static int 281static int
250ext4_ext_max_entries(struct inode *inode, int depth) 282ext4_ext_max_entries(struct inode *inode, int depth)
251{ 283{
@@ -524,6 +556,7 @@ ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
524 alloc = 1; 556 alloc = 1;
525 } 557 }
526 path[0].p_hdr = eh; 558 path[0].p_hdr = eh;
559 path[0].p_bh = NULL;
527 560
528 i = depth; 561 i = depth;
529 /* walk through the tree */ 562 /* walk through the tree */
@@ -552,12 +585,14 @@ ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
552 } 585 }
553 586
554 path[ppos].p_depth = i; 587 path[ppos].p_depth = i;
555 path[ppos].p_hdr = eh;
556 path[ppos].p_ext = NULL; 588 path[ppos].p_ext = NULL;
557 path[ppos].p_idx = NULL; 589 path[ppos].p_idx = NULL;
558 590
559 /* find extent */ 591 /* find extent */
560 ext4_ext_binsearch(inode, path + ppos, block); 592 ext4_ext_binsearch(inode, path + ppos, block);
593 /* if not an empty leaf */
594 if (path[ppos].p_ext)
595 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
561 596
562 ext4_ext_show_path(inode, path); 597 ext4_ext_show_path(inode, path);
563 598
@@ -688,7 +723,8 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode,
688 /* allocate all needed blocks */ 723 /* allocate all needed blocks */
689 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at); 724 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
690 for (a = 0; a < depth - at; a++) { 725 for (a = 0; a < depth - at; a++) {
691 newblock = ext4_ext_new_block(handle, inode, path, newext, &err); 726 newblock = ext4_ext_new_meta_block(handle, inode, path,
727 newext, &err);
692 if (newblock == 0) 728 if (newblock == 0)
693 goto cleanup; 729 goto cleanup;
694 ablocks[a] = newblock; 730 ablocks[a] = newblock;
@@ -884,7 +920,7 @@ static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
884 ext4_fsblk_t newblock; 920 ext4_fsblk_t newblock;
885 int err = 0; 921 int err = 0;
886 922
887 newblock = ext4_ext_new_block(handle, inode, path, newext, &err); 923 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
888 if (newblock == 0) 924 if (newblock == 0)
889 return err; 925 return err;
890 926
@@ -981,6 +1017,8 @@ repeat:
981 /* if we found index with free entry, then use that 1017 /* if we found index with free entry, then use that
982 * entry: create all needed subtree and add new leaf */ 1018 * entry: create all needed subtree and add new leaf */
983 err = ext4_ext_split(handle, inode, path, newext, i); 1019 err = ext4_ext_split(handle, inode, path, newext, i);
1020 if (err)
1021 goto out;
984 1022
985 /* refill path */ 1023 /* refill path */
986 ext4_ext_drop_refs(path); 1024 ext4_ext_drop_refs(path);
@@ -1883,11 +1921,9 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
1883 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb); 1921 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1884#endif 1922#endif
1885 1923
1886 handle = ext4_ext_journal_restart(handle, credits); 1924 err = ext4_ext_journal_restart(handle, credits);
1887 if (IS_ERR(handle)) { 1925 if (err)
1888 err = PTR_ERR(handle);
1889 goto out; 1926 goto out;
1890 }
1891 1927
1892 err = ext4_ext_get_access(handle, inode, path + depth); 1928 err = ext4_ext_get_access(handle, inode, path + depth);
1893 if (err) 1929 if (err)
@@ -2529,6 +2565,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2529 int err = 0, depth, ret; 2565 int err = 0, depth, ret;
2530 unsigned long allocated = 0; 2566 unsigned long allocated = 0;
2531 struct ext4_allocation_request ar; 2567 struct ext4_allocation_request ar;
2568 loff_t disksize;
2532 2569
2533 __clear_bit(BH_New, &bh_result->b_state); 2570 __clear_bit(BH_New, &bh_result->b_state);
2534 ext_debug("blocks %u/%lu requested for inode %u\n", 2571 ext_debug("blocks %u/%lu requested for inode %u\n",
@@ -2616,8 +2653,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2616 */ 2653 */
2617 if (allocated > max_blocks) 2654 if (allocated > max_blocks)
2618 allocated = max_blocks; 2655 allocated = max_blocks;
2619 /* mark the buffer unwritten */ 2656 set_buffer_unwritten(bh_result);
2620 __set_bit(BH_Unwritten, &bh_result->b_state);
2621 goto out2; 2657 goto out2;
2622 } 2658 }
2623 2659
@@ -2716,14 +2752,19 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
2716 goto out2; 2752 goto out2;
2717 } 2753 }
2718 2754
2719 if (extend_disksize && inode->i_size > EXT4_I(inode)->i_disksize)
2720 EXT4_I(inode)->i_disksize = inode->i_size;
2721
2722 /* previous routine could use block we allocated */ 2755 /* previous routine could use block we allocated */
2723 newblock = ext_pblock(&newex); 2756 newblock = ext_pblock(&newex);
2724 allocated = ext4_ext_get_actual_len(&newex); 2757 allocated = ext4_ext_get_actual_len(&newex);
2725outnew: 2758outnew:
2726 __set_bit(BH_New, &bh_result->b_state); 2759 if (extend_disksize) {
2760 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2761 if (disksize > i_size_read(inode))
2762 disksize = i_size_read(inode);
2763 if (disksize > EXT4_I(inode)->i_disksize)
2764 EXT4_I(inode)->i_disksize = disksize;
2765 }
2766
2767 set_buffer_new(bh_result);
2727 2768
2728 /* Cache only when it is _not_ an uninitialized extent */ 2769 /* Cache only when it is _not_ an uninitialized extent */
2729 if (create != EXT4_CREATE_UNINITIALIZED_EXT) 2770 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
@@ -2733,7 +2774,7 @@ out:
2733 if (allocated > max_blocks) 2774 if (allocated > max_blocks)
2734 allocated = max_blocks; 2775 allocated = max_blocks;
2735 ext4_ext_show_leaf(inode, path); 2776 ext4_ext_show_leaf(inode, path);
2736 __set_bit(BH_Mapped, &bh_result->b_state); 2777 set_buffer_mapped(bh_result);
2737 bh_result->b_bdev = inode->i_sb->s_bdev; 2778 bh_result->b_bdev = inode->i_sb->s_bdev;
2738 bh_result->b_blocknr = newblock; 2779 bh_result->b_blocknr = newblock;
2739out2: 2780out2:
@@ -2744,7 +2785,7 @@ out2:
2744 return err ? err : allocated; 2785 return err ? err : allocated;
2745} 2786}
2746 2787
2747void ext4_ext_truncate(struct inode * inode, struct page *page) 2788void ext4_ext_truncate(struct inode *inode)
2748{ 2789{
2749 struct address_space *mapping = inode->i_mapping; 2790 struct address_space *mapping = inode->i_mapping;
2750 struct super_block *sb = inode->i_sb; 2791 struct super_block *sb = inode->i_sb;
@@ -2757,18 +2798,14 @@ void ext4_ext_truncate(struct inode * inode, struct page *page)
2757 */ 2798 */
2758 err = ext4_writepage_trans_blocks(inode) + 3; 2799 err = ext4_writepage_trans_blocks(inode) + 3;
2759 handle = ext4_journal_start(inode, err); 2800 handle = ext4_journal_start(inode, err);
2760 if (IS_ERR(handle)) { 2801 if (IS_ERR(handle))
2761 if (page) {
2762 clear_highpage(page);
2763 flush_dcache_page(page);
2764 unlock_page(page);
2765 page_cache_release(page);
2766 }
2767 return; 2802 return;
2768 }
2769 2803
2770 if (page) 2804 if (inode->i_size & (sb->s_blocksize - 1))
2771 ext4_block_truncate_page(handle, page, mapping, inode->i_size); 2805 ext4_block_truncate_page(handle, mapping, inode->i_size);
2806
2807 if (ext4_orphan_add(handle, inode))
2808 goto out_stop;
2772 2809
2773 down_write(&EXT4_I(inode)->i_data_sem); 2810 down_write(&EXT4_I(inode)->i_data_sem);
2774 ext4_ext_invalidate_cache(inode); 2811 ext4_ext_invalidate_cache(inode);
@@ -2780,8 +2817,6 @@ void ext4_ext_truncate(struct inode * inode, struct page *page)
2780 * Probably we need not scan at all, 2817 * Probably we need not scan at all,
2781 * because page truncation is enough. 2818 * because page truncation is enough.
2782 */ 2819 */
2783 if (ext4_orphan_add(handle, inode))
2784 goto out_stop;
2785 2820
2786 /* we have to know where to truncate from in crash case */ 2821 /* we have to know where to truncate from in crash case */
2787 EXT4_I(inode)->i_disksize = inode->i_size; 2822 EXT4_I(inode)->i_disksize = inode->i_size;
@@ -2798,6 +2833,7 @@ void ext4_ext_truncate(struct inode * inode, struct page *page)
2798 handle->h_sync = 1; 2833 handle->h_sync = 1;
2799 2834
2800out_stop: 2835out_stop:
2836 up_write(&EXT4_I(inode)->i_data_sem);
2801 /* 2837 /*
2802 * If this was a simple ftruncate() and the file will remain alive, 2838 * If this was a simple ftruncate() and the file will remain alive,
2803 * then we need to clear up the orphan record which we created above. 2839 * then we need to clear up the orphan record which we created above.
@@ -2808,7 +2844,6 @@ out_stop:
2808 if (inode->i_nlink) 2844 if (inode->i_nlink)
2809 ext4_orphan_del(handle, inode); 2845 ext4_orphan_del(handle, inode);
2810 2846
2811 up_write(&EXT4_I(inode)->i_data_sem);
2812 inode->i_mtime = inode->i_ctime = ext4_current_time(inode); 2847 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2813 ext4_mark_inode_dirty(handle, inode); 2848 ext4_mark_inode_dirty(handle, inode);
2814 ext4_journal_stop(handle); 2849 ext4_journal_stop(handle);
@@ -2911,7 +2946,7 @@ retry:
2911 } 2946 }
2912 ret = ext4_get_blocks_wrap(handle, inode, block, 2947 ret = ext4_get_blocks_wrap(handle, inode, block,
2913 max_blocks, &map_bh, 2948 max_blocks, &map_bh,
2914 EXT4_CREATE_UNINITIALIZED_EXT, 0); 2949 EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
2915 if (ret <= 0) { 2950 if (ret <= 0) {
2916#ifdef EXT4FS_DEBUG 2951#ifdef EXT4FS_DEBUG
2917 WARN_ON(ret <= 0); 2952 WARN_ON(ret <= 0);
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 4159be6366ab..430eb7978db4 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -123,6 +123,23 @@ force_commit:
123 return ret; 123 return ret;
124} 124}
125 125
126static struct vm_operations_struct ext4_file_vm_ops = {
127 .fault = filemap_fault,
128 .page_mkwrite = ext4_page_mkwrite,
129};
130
131static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
132{
133 struct address_space *mapping = file->f_mapping;
134
135 if (!mapping->a_ops->readpage)
136 return -ENOEXEC;
137 file_accessed(file);
138 vma->vm_ops = &ext4_file_vm_ops;
139 vma->vm_flags |= VM_CAN_NONLINEAR;
140 return 0;
141}
142
126const struct file_operations ext4_file_operations = { 143const struct file_operations ext4_file_operations = {
127 .llseek = generic_file_llseek, 144 .llseek = generic_file_llseek,
128 .read = do_sync_read, 145 .read = do_sync_read,
@@ -133,7 +150,7 @@ const struct file_operations ext4_file_operations = {
133#ifdef CONFIG_COMPAT 150#ifdef CONFIG_COMPAT
134 .compat_ioctl = ext4_compat_ioctl, 151 .compat_ioctl = ext4_compat_ioctl,
135#endif 152#endif
136 .mmap = generic_file_mmap, 153 .mmap = ext4_file_mmap,
137 .open = generic_file_open, 154 .open = generic_file_open,
138 .release = ext4_release_file, 155 .release = ext4_release_file,
139 .fsync = ext4_sync_file, 156 .fsync = ext4_sync_file,
@@ -144,6 +161,7 @@ const struct file_operations ext4_file_operations = {
144const struct inode_operations ext4_file_inode_operations = { 161const struct inode_operations ext4_file_inode_operations = {
145 .truncate = ext4_truncate, 162 .truncate = ext4_truncate,
146 .setattr = ext4_setattr, 163 .setattr = ext4_setattr,
164 .getattr = ext4_getattr,
147#ifdef CONFIG_EXT4DEV_FS_XATTR 165#ifdef CONFIG_EXT4DEV_FS_XATTR
148 .setxattr = generic_setxattr, 166 .setxattr = generic_setxattr,
149 .getxattr = generic_getxattr, 167 .getxattr = generic_getxattr,
diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 1c8ba48d4f8d..a45c3737ad31 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -27,6 +27,7 @@
27#include <linux/sched.h> 27#include <linux/sched.h>
28#include <linux/writeback.h> 28#include <linux/writeback.h>
29#include <linux/jbd2.h> 29#include <linux/jbd2.h>
30#include <linux/blkdev.h>
30#include "ext4.h" 31#include "ext4.h"
31#include "ext4_jbd2.h" 32#include "ext4_jbd2.h"
32 33
@@ -45,6 +46,7 @@
45int ext4_sync_file(struct file * file, struct dentry *dentry, int datasync) 46int ext4_sync_file(struct file * file, struct dentry *dentry, int datasync)
46{ 47{
47 struct inode *inode = dentry->d_inode; 48 struct inode *inode = dentry->d_inode;
49 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
48 int ret = 0; 50 int ret = 0;
49 51
50 J_ASSERT(ext4_journal_current_handle() == NULL); 52 J_ASSERT(ext4_journal_current_handle() == NULL);
@@ -85,6 +87,8 @@ int ext4_sync_file(struct file * file, struct dentry *dentry, int datasync)
85 .nr_to_write = 0, /* sys_fsync did this */ 87 .nr_to_write = 0, /* sys_fsync did this */
86 }; 88 };
87 ret = sync_inode(inode, &wbc); 89 ret = sync_inode(inode, &wbc);
90 if (journal && (journal->j_flags & JBD2_BARRIER))
91 blkdev_issue_flush(inode->i_sb->s_bdev, NULL);
88 } 92 }
89out: 93out:
90 return ret; 94 return ret;
diff --git a/fs/ext4/group.h b/fs/ext4/group.h
index 7eb0604e7eea..c2c0a8d06d0e 100644
--- a/fs/ext4/group.h
+++ b/fs/ext4/group.h
@@ -13,7 +13,7 @@ extern __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 group,
13 struct ext4_group_desc *gdp); 13 struct ext4_group_desc *gdp);
14extern int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 group, 14extern int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 group,
15 struct ext4_group_desc *gdp); 15 struct ext4_group_desc *gdp);
16struct buffer_head *read_block_bitmap(struct super_block *sb, 16struct buffer_head *ext4_read_block_bitmap(struct super_block *sb,
17 ext4_group_t block_group); 17 ext4_group_t block_group);
18extern unsigned ext4_init_block_bitmap(struct super_block *sb, 18extern unsigned ext4_init_block_bitmap(struct super_block *sb,
19 struct buffer_head *bh, 19 struct buffer_head *bh,
diff --git a/fs/ext4/ialloc.c b/fs/ext4/ialloc.c
index c6efbab0c801..a92eb305344f 100644
--- a/fs/ext4/ialloc.c
+++ b/fs/ext4/ialloc.c
@@ -157,6 +157,7 @@ void ext4_free_inode (handle_t *handle, struct inode * inode)
157 struct ext4_super_block * es; 157 struct ext4_super_block * es;
158 struct ext4_sb_info *sbi; 158 struct ext4_sb_info *sbi;
159 int fatal = 0, err; 159 int fatal = 0, err;
160 ext4_group_t flex_group;
160 161
161 if (atomic_read(&inode->i_count) > 1) { 162 if (atomic_read(&inode->i_count) > 1) {
162 printk ("ext4_free_inode: inode has count=%d\n", 163 printk ("ext4_free_inode: inode has count=%d\n",
@@ -232,6 +233,12 @@ void ext4_free_inode (handle_t *handle, struct inode * inode)
232 if (is_directory) 233 if (is_directory)
233 percpu_counter_dec(&sbi->s_dirs_counter); 234 percpu_counter_dec(&sbi->s_dirs_counter);
234 235
236 if (sbi->s_log_groups_per_flex) {
237 flex_group = ext4_flex_group(sbi, block_group);
238 spin_lock(sb_bgl_lock(sbi, flex_group));
239 sbi->s_flex_groups[flex_group].free_inodes++;
240 spin_unlock(sb_bgl_lock(sbi, flex_group));
241 }
235 } 242 }
236 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); 243 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
237 err = ext4_journal_dirty_metadata(handle, bh2); 244 err = ext4_journal_dirty_metadata(handle, bh2);
@@ -286,6 +293,80 @@ static int find_group_dir(struct super_block *sb, struct inode *parent,
286 return ret; 293 return ret;
287} 294}
288 295
296#define free_block_ratio 10
297
298static int find_group_flex(struct super_block *sb, struct inode *parent,
299 ext4_group_t *best_group)
300{
301 struct ext4_sb_info *sbi = EXT4_SB(sb);
302 struct ext4_group_desc *desc;
303 struct buffer_head *bh;
304 struct flex_groups *flex_group = sbi->s_flex_groups;
305 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
306 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
307 ext4_group_t ngroups = sbi->s_groups_count;
308 int flex_size = ext4_flex_bg_size(sbi);
309 ext4_group_t best_flex = parent_fbg_group;
310 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
311 int flexbg_free_blocks;
312 int flex_freeb_ratio;
313 ext4_group_t n_fbg_groups;
314 ext4_group_t i;
315
316 n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >>
317 sbi->s_log_groups_per_flex;
318
319find_close_to_parent:
320 flexbg_free_blocks = flex_group[best_flex].free_blocks;
321 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
322 if (flex_group[best_flex].free_inodes &&
323 flex_freeb_ratio > free_block_ratio)
324 goto found_flexbg;
325
326 if (best_flex && best_flex == parent_fbg_group) {
327 best_flex--;
328 goto find_close_to_parent;
329 }
330
331 for (i = 0; i < n_fbg_groups; i++) {
332 if (i == parent_fbg_group || i == parent_fbg_group - 1)
333 continue;
334
335 flexbg_free_blocks = flex_group[i].free_blocks;
336 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
337
338 if (flex_freeb_ratio > free_block_ratio &&
339 flex_group[i].free_inodes) {
340 best_flex = i;
341 goto found_flexbg;
342 }
343
344 if (best_flex < 0 ||
345 (flex_group[i].free_blocks >
346 flex_group[best_flex].free_blocks &&
347 flex_group[i].free_inodes))
348 best_flex = i;
349 }
350
351 if (!flex_group[best_flex].free_inodes ||
352 !flex_group[best_flex].free_blocks)
353 return -1;
354
355found_flexbg:
356 for (i = best_flex * flex_size; i < ngroups &&
357 i < (best_flex + 1) * flex_size; i++) {
358 desc = ext4_get_group_desc(sb, i, &bh);
359 if (le16_to_cpu(desc->bg_free_inodes_count)) {
360 *best_group = i;
361 goto out;
362 }
363 }
364
365 return -1;
366out:
367 return 0;
368}
369
289/* 370/*
290 * Orlov's allocator for directories. 371 * Orlov's allocator for directories.
291 * 372 *
@@ -501,6 +582,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
501 struct inode *ret; 582 struct inode *ret;
502 ext4_group_t i; 583 ext4_group_t i;
503 int free = 0; 584 int free = 0;
585 ext4_group_t flex_group;
504 586
505 /* Cannot create files in a deleted directory */ 587 /* Cannot create files in a deleted directory */
506 if (!dir || !dir->i_nlink) 588 if (!dir || !dir->i_nlink)
@@ -514,6 +596,12 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
514 596
515 sbi = EXT4_SB(sb); 597 sbi = EXT4_SB(sb);
516 es = sbi->s_es; 598 es = sbi->s_es;
599
600 if (sbi->s_log_groups_per_flex) {
601 ret2 = find_group_flex(sb, dir, &group);
602 goto got_group;
603 }
604
517 if (S_ISDIR(mode)) { 605 if (S_ISDIR(mode)) {
518 if (test_opt (sb, OLDALLOC)) 606 if (test_opt (sb, OLDALLOC))
519 ret2 = find_group_dir(sb, dir, &group); 607 ret2 = find_group_dir(sb, dir, &group);
@@ -522,6 +610,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
522 } else 610 } else
523 ret2 = find_group_other(sb, dir, &group); 611 ret2 = find_group_other(sb, dir, &group);
524 612
613got_group:
525 err = -ENOSPC; 614 err = -ENOSPC;
526 if (ret2 == -1) 615 if (ret2 == -1)
527 goto out; 616 goto out;
@@ -600,7 +689,7 @@ got:
600 /* We may have to initialize the block bitmap if it isn't already */ 689 /* We may have to initialize the block bitmap if it isn't already */
601 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && 690 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
602 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 691 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
603 struct buffer_head *block_bh = read_block_bitmap(sb, group); 692 struct buffer_head *block_bh = ext4_read_block_bitmap(sb, group);
604 693
605 BUFFER_TRACE(block_bh, "get block bitmap access"); 694 BUFFER_TRACE(block_bh, "get block bitmap access");
606 err = ext4_journal_get_write_access(handle, block_bh); 695 err = ext4_journal_get_write_access(handle, block_bh);
@@ -676,6 +765,13 @@ got:
676 percpu_counter_inc(&sbi->s_dirs_counter); 765 percpu_counter_inc(&sbi->s_dirs_counter);
677 sb->s_dirt = 1; 766 sb->s_dirt = 1;
678 767
768 if (sbi->s_log_groups_per_flex) {
769 flex_group = ext4_flex_group(sbi, group);
770 spin_lock(sb_bgl_lock(sbi, flex_group));
771 sbi->s_flex_groups[flex_group].free_inodes--;
772 spin_unlock(sb_bgl_lock(sbi, flex_group));
773 }
774
679 inode->i_uid = current->fsuid; 775 inode->i_uid = current->fsuid;
680 if (test_opt (sb, GRPID)) 776 if (test_opt (sb, GRPID))
681 inode->i_gid = dir->i_gid; 777 inode->i_gid = dir->i_gid;
@@ -740,14 +836,10 @@ got:
740 goto fail_free_drop; 836 goto fail_free_drop;
741 837
742 if (test_opt(sb, EXTENTS)) { 838 if (test_opt(sb, EXTENTS)) {
743 /* set extent flag only for diretory, file and normal symlink*/ 839 /* set extent flag only for directory, file and normal symlink*/
744 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 840 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
745 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; 841 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
746 ext4_ext_tree_init(handle, inode); 842 ext4_ext_tree_init(handle, inode);
747 err = ext4_update_incompat_feature(handle, sb,
748 EXT4_FEATURE_INCOMPAT_EXTENTS);
749 if (err)
750 goto fail_free_drop;
751 } 843 }
752 } 844 }
753 845
@@ -817,6 +909,14 @@ struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
817 if (IS_ERR(inode)) 909 if (IS_ERR(inode))
818 goto iget_failed; 910 goto iget_failed;
819 911
912 /*
913 * If the orphans has i_nlinks > 0 then it should be able to be
914 * truncated, otherwise it won't be removed from the orphan list
915 * during processing and an infinite loop will result.
916 */
917 if (inode->i_nlink && !ext4_can_truncate(inode))
918 goto bad_orphan;
919
820 if (NEXT_ORPHAN(inode) > max_ino) 920 if (NEXT_ORPHAN(inode) > max_ino)
821 goto bad_orphan; 921 goto bad_orphan;
822 brelse(bitmap_bh); 922 brelse(bitmap_bh);
@@ -838,6 +938,7 @@ bad_orphan:
838 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", 938 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
839 NEXT_ORPHAN(inode)); 939 NEXT_ORPHAN(inode));
840 printk(KERN_NOTICE "max_ino=%lu\n", max_ino); 940 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
941 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
841 /* Avoid freeing blocks if we got a bad deleted inode */ 942 /* Avoid freeing blocks if we got a bad deleted inode */
842 if (inode->i_nlink == 0) 943 if (inode->i_nlink == 0)
843 inode->i_blocks = 0; 944 inode->i_blocks = 0;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 8d9707746413..8ca2763df091 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -32,12 +32,23 @@
32#include <linux/string.h> 32#include <linux/string.h>
33#include <linux/buffer_head.h> 33#include <linux/buffer_head.h>
34#include <linux/writeback.h> 34#include <linux/writeback.h>
35#include <linux/pagevec.h>
35#include <linux/mpage.h> 36#include <linux/mpage.h>
36#include <linux/uio.h> 37#include <linux/uio.h>
37#include <linux/bio.h> 38#include <linux/bio.h>
38#include "ext4_jbd2.h" 39#include "ext4_jbd2.h"
39#include "xattr.h" 40#include "xattr.h"
40#include "acl.h" 41#include "acl.h"
42#include "ext4_extents.h"
43
44static inline int ext4_begin_ordered_truncate(struct inode *inode,
45 loff_t new_size)
46{
47 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
48 new_size);
49}
50
51static void ext4_invalidatepage(struct page *page, unsigned long offset);
41 52
42/* 53/*
43 * Test whether an inode is a fast symlink. 54 * Test whether an inode is a fast symlink.
@@ -181,6 +192,8 @@ void ext4_delete_inode (struct inode * inode)
181{ 192{
182 handle_t *handle; 193 handle_t *handle;
183 194
195 if (ext4_should_order_data(inode))
196 ext4_begin_ordered_truncate(inode, 0);
184 truncate_inode_pages(&inode->i_data, 0); 197 truncate_inode_pages(&inode->i_data, 0);
185 198
186 if (is_bad_inode(inode)) 199 if (is_bad_inode(inode))
@@ -508,11 +521,12 @@ static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
508 * direct blocks 521 * direct blocks
509 */ 522 */
510static int ext4_alloc_blocks(handle_t *handle, struct inode *inode, 523static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
511 ext4_fsblk_t goal, int indirect_blks, int blks, 524 ext4_lblk_t iblock, ext4_fsblk_t goal,
512 ext4_fsblk_t new_blocks[4], int *err) 525 int indirect_blks, int blks,
526 ext4_fsblk_t new_blocks[4], int *err)
513{ 527{
514 int target, i; 528 int target, i;
515 unsigned long count = 0; 529 unsigned long count = 0, blk_allocated = 0;
516 int index = 0; 530 int index = 0;
517 ext4_fsblk_t current_block = 0; 531 ext4_fsblk_t current_block = 0;
518 int ret = 0; 532 int ret = 0;
@@ -525,12 +539,13 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
525 * the first direct block of this branch. That's the 539 * the first direct block of this branch. That's the
526 * minimum number of blocks need to allocate(required) 540 * minimum number of blocks need to allocate(required)
527 */ 541 */
528 target = blks + indirect_blks; 542 /* first we try to allocate the indirect blocks */
529 543 target = indirect_blks;
530 while (1) { 544 while (target > 0) {
531 count = target; 545 count = target;
532 /* allocating blocks for indirect blocks and direct blocks */ 546 /* allocating blocks for indirect blocks and direct blocks */
533 current_block = ext4_new_blocks(handle,inode,goal,&count,err); 547 current_block = ext4_new_meta_blocks(handle, inode,
548 goal, &count, err);
534 if (*err) 549 if (*err)
535 goto failed_out; 550 goto failed_out;
536 551
@@ -540,16 +555,48 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
540 new_blocks[index++] = current_block++; 555 new_blocks[index++] = current_block++;
541 count--; 556 count--;
542 } 557 }
543 558 if (count > 0) {
544 if (count > 0) 559 /*
560 * save the new block number
561 * for the first direct block
562 */
563 new_blocks[index] = current_block;
564 printk(KERN_INFO "%s returned more blocks than "
565 "requested\n", __func__);
566 WARN_ON(1);
545 break; 567 break;
568 }
546 } 569 }
547 570
548 /* save the new block number for the first direct block */ 571 target = blks - count ;
549 new_blocks[index] = current_block; 572 blk_allocated = count;
550 573 if (!target)
574 goto allocated;
575 /* Now allocate data blocks */
576 count = target;
577 /* allocating blocks for data blocks */
578 current_block = ext4_new_blocks(handle, inode, iblock,
579 goal, &count, err);
580 if (*err && (target == blks)) {
581 /*
582 * if the allocation failed and we didn't allocate
583 * any blocks before
584 */
585 goto failed_out;
586 }
587 if (!*err) {
588 if (target == blks) {
589 /*
590 * save the new block number
591 * for the first direct block
592 */
593 new_blocks[index] = current_block;
594 }
595 blk_allocated += count;
596 }
597allocated:
551 /* total number of blocks allocated for direct blocks */ 598 /* total number of blocks allocated for direct blocks */
552 ret = count; 599 ret = blk_allocated;
553 *err = 0; 600 *err = 0;
554 return ret; 601 return ret;
555failed_out: 602failed_out:
@@ -584,8 +631,9 @@ failed_out:
584 * as described above and return 0. 631 * as described above and return 0.
585 */ 632 */
586static int ext4_alloc_branch(handle_t *handle, struct inode *inode, 633static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
587 int indirect_blks, int *blks, ext4_fsblk_t goal, 634 ext4_lblk_t iblock, int indirect_blks,
588 ext4_lblk_t *offsets, Indirect *branch) 635 int *blks, ext4_fsblk_t goal,
636 ext4_lblk_t *offsets, Indirect *branch)
589{ 637{
590 int blocksize = inode->i_sb->s_blocksize; 638 int blocksize = inode->i_sb->s_blocksize;
591 int i, n = 0; 639 int i, n = 0;
@@ -595,7 +643,7 @@ static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
595 ext4_fsblk_t new_blocks[4]; 643 ext4_fsblk_t new_blocks[4];
596 ext4_fsblk_t current_block; 644 ext4_fsblk_t current_block;
597 645
598 num = ext4_alloc_blocks(handle, inode, goal, indirect_blks, 646 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
599 *blks, new_blocks, &err); 647 *blks, new_blocks, &err);
600 if (err) 648 if (err)
601 return err; 649 return err;
@@ -799,6 +847,7 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
799 struct ext4_inode_info *ei = EXT4_I(inode); 847 struct ext4_inode_info *ei = EXT4_I(inode);
800 int count = 0; 848 int count = 0;
801 ext4_fsblk_t first_block = 0; 849 ext4_fsblk_t first_block = 0;
850 loff_t disksize;
802 851
803 852
804 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)); 853 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
@@ -855,8 +904,9 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
855 /* 904 /*
856 * Block out ext4_truncate while we alter the tree 905 * Block out ext4_truncate while we alter the tree
857 */ 906 */
858 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal, 907 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
859 offsets + (partial - chain), partial); 908 &count, goal,
909 offsets + (partial - chain), partial);
860 910
861 /* 911 /*
862 * The ext4_splice_branch call will free and forget any buffers 912 * The ext4_splice_branch call will free and forget any buffers
@@ -873,8 +923,13 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
873 * protect it if you're about to implement concurrent 923 * protect it if you're about to implement concurrent
874 * ext4_get_block() -bzzz 924 * ext4_get_block() -bzzz
875 */ 925 */
876 if (!err && extend_disksize && inode->i_size > ei->i_disksize) 926 if (!err && extend_disksize) {
877 ei->i_disksize = inode->i_size; 927 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
928 if (disksize > i_size_read(inode))
929 disksize = i_size_read(inode);
930 if (disksize > ei->i_disksize)
931 ei->i_disksize = disksize;
932 }
878 if (err) 933 if (err)
879 goto cleanup; 934 goto cleanup;
880 935
@@ -934,7 +989,7 @@ out:
934 */ 989 */
935int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block, 990int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
936 unsigned long max_blocks, struct buffer_head *bh, 991 unsigned long max_blocks, struct buffer_head *bh,
937 int create, int extend_disksize) 992 int create, int extend_disksize, int flag)
938{ 993{
939 int retval; 994 int retval;
940 995
@@ -975,6 +1030,15 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
975 * with create == 1 flag. 1030 * with create == 1 flag.
976 */ 1031 */
977 down_write((&EXT4_I(inode)->i_data_sem)); 1032 down_write((&EXT4_I(inode)->i_data_sem));
1033
1034 /*
1035 * if the caller is from delayed allocation writeout path
1036 * we have already reserved fs blocks for allocation
1037 * let the underlying get_block() function know to
1038 * avoid double accounting
1039 */
1040 if (flag)
1041 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
978 /* 1042 /*
979 * We need to check for EXT4 here because migrate 1043 * We need to check for EXT4 here because migrate
980 * could have changed the inode type in between 1044 * could have changed the inode type in between
@@ -996,6 +1060,18 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
996 ~EXT4_EXT_MIGRATE; 1060 ~EXT4_EXT_MIGRATE;
997 } 1061 }
998 } 1062 }
1063
1064 if (flag) {
1065 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1066 /*
1067 * Update reserved blocks/metadata blocks
1068 * after successful block allocation
1069 * which were deferred till now
1070 */
1071 if ((retval > 0) && buffer_delay(bh))
1072 ext4_da_release_space(inode, retval, 0);
1073 }
1074
999 up_write((&EXT4_I(inode)->i_data_sem)); 1075 up_write((&EXT4_I(inode)->i_data_sem));
1000 return retval; 1076 return retval;
1001} 1077}
@@ -1021,7 +1097,7 @@ static int ext4_get_block(struct inode *inode, sector_t iblock,
1021 } 1097 }
1022 1098
1023 ret = ext4_get_blocks_wrap(handle, inode, iblock, 1099 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1024 max_blocks, bh_result, create, 0); 1100 max_blocks, bh_result, create, 0, 0);
1025 if (ret > 0) { 1101 if (ret > 0) {
1026 bh_result->b_size = (ret << inode->i_blkbits); 1102 bh_result->b_size = (ret << inode->i_blkbits);
1027 ret = 0; 1103 ret = 0;
@@ -1047,7 +1123,7 @@ struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1047 dummy.b_blocknr = -1000; 1123 dummy.b_blocknr = -1000;
1048 buffer_trace_init(&dummy.b_history); 1124 buffer_trace_init(&dummy.b_history);
1049 err = ext4_get_blocks_wrap(handle, inode, block, 1, 1125 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1050 &dummy, create, 1); 1126 &dummy, create, 1, 0);
1051 /* 1127 /*
1052 * ext4_get_blocks_handle() returns number of blocks 1128 * ext4_get_blocks_handle() returns number of blocks
1053 * mapped. 0 in case of a HOLE. 1129 * mapped. 0 in case of a HOLE.
@@ -1203,19 +1279,20 @@ static int ext4_write_begin(struct file *file, struct address_space *mapping,
1203 to = from + len; 1279 to = from + len;
1204 1280
1205retry: 1281retry:
1206 page = __grab_cache_page(mapping, index);
1207 if (!page)
1208 return -ENOMEM;
1209 *pagep = page;
1210
1211 handle = ext4_journal_start(inode, needed_blocks); 1282 handle = ext4_journal_start(inode, needed_blocks);
1212 if (IS_ERR(handle)) { 1283 if (IS_ERR(handle)) {
1213 unlock_page(page);
1214 page_cache_release(page);
1215 ret = PTR_ERR(handle); 1284 ret = PTR_ERR(handle);
1216 goto out; 1285 goto out;
1217 } 1286 }
1218 1287
1288 page = __grab_cache_page(mapping, index);
1289 if (!page) {
1290 ext4_journal_stop(handle);
1291 ret = -ENOMEM;
1292 goto out;
1293 }
1294 *pagep = page;
1295
1219 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, 1296 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1220 ext4_get_block); 1297 ext4_get_block);
1221 1298
@@ -1225,8 +1302,8 @@ retry:
1225 } 1302 }
1226 1303
1227 if (ret) { 1304 if (ret) {
1228 ext4_journal_stop(handle);
1229 unlock_page(page); 1305 unlock_page(page);
1306 ext4_journal_stop(handle);
1230 page_cache_release(page); 1307 page_cache_release(page);
1231 } 1308 }
1232 1309
@@ -1236,15 +1313,6 @@ out:
1236 return ret; 1313 return ret;
1237} 1314}
1238 1315
1239int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1240{
1241 int err = jbd2_journal_dirty_data(handle, bh);
1242 if (err)
1243 ext4_journal_abort_handle(__func__, __func__,
1244 bh, handle, err);
1245 return err;
1246}
1247
1248/* For write_end() in data=journal mode */ 1316/* For write_end() in data=journal mode */
1249static int write_end_fn(handle_t *handle, struct buffer_head *bh) 1317static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1250{ 1318{
@@ -1255,29 +1323,6 @@ static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1255} 1323}
1256 1324
1257/* 1325/*
1258 * Generic write_end handler for ordered and writeback ext4 journal modes.
1259 * We can't use generic_write_end, because that unlocks the page and we need to
1260 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1261 * after block_write_end.
1262 */
1263static int ext4_generic_write_end(struct file *file,
1264 struct address_space *mapping,
1265 loff_t pos, unsigned len, unsigned copied,
1266 struct page *page, void *fsdata)
1267{
1268 struct inode *inode = file->f_mapping->host;
1269
1270 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1271
1272 if (pos+copied > inode->i_size) {
1273 i_size_write(inode, pos+copied);
1274 mark_inode_dirty(inode);
1275 }
1276
1277 return copied;
1278}
1279
1280/*
1281 * We need to pick up the new inode size which generic_commit_write gave us 1326 * We need to pick up the new inode size which generic_commit_write gave us
1282 * `file' can be NULL - eg, when called from page_symlink(). 1327 * `file' can be NULL - eg, when called from page_symlink().
1283 * 1328 *
@@ -1290,15 +1335,14 @@ static int ext4_ordered_write_end(struct file *file,
1290 struct page *page, void *fsdata) 1335 struct page *page, void *fsdata)
1291{ 1336{
1292 handle_t *handle = ext4_journal_current_handle(); 1337 handle_t *handle = ext4_journal_current_handle();
1293 struct inode *inode = file->f_mapping->host; 1338 struct inode *inode = mapping->host;
1294 unsigned from, to; 1339 unsigned from, to;
1295 int ret = 0, ret2; 1340 int ret = 0, ret2;
1296 1341
1297 from = pos & (PAGE_CACHE_SIZE - 1); 1342 from = pos & (PAGE_CACHE_SIZE - 1);
1298 to = from + len; 1343 to = from + len;
1299 1344
1300 ret = walk_page_buffers(handle, page_buffers(page), 1345 ret = ext4_jbd2_file_inode(handle, inode);
1301 from, to, NULL, ext4_journal_dirty_data);
1302 1346
1303 if (ret == 0) { 1347 if (ret == 0) {
1304 /* 1348 /*
@@ -1311,7 +1355,7 @@ static int ext4_ordered_write_end(struct file *file,
1311 new_i_size = pos + copied; 1355 new_i_size = pos + copied;
1312 if (new_i_size > EXT4_I(inode)->i_disksize) 1356 if (new_i_size > EXT4_I(inode)->i_disksize)
1313 EXT4_I(inode)->i_disksize = new_i_size; 1357 EXT4_I(inode)->i_disksize = new_i_size;
1314 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, 1358 ret2 = generic_write_end(file, mapping, pos, len, copied,
1315 page, fsdata); 1359 page, fsdata);
1316 copied = ret2; 1360 copied = ret2;
1317 if (ret2 < 0) 1361 if (ret2 < 0)
@@ -1320,8 +1364,6 @@ static int ext4_ordered_write_end(struct file *file,
1320 ret2 = ext4_journal_stop(handle); 1364 ret2 = ext4_journal_stop(handle);
1321 if (!ret) 1365 if (!ret)
1322 ret = ret2; 1366 ret = ret2;
1323 unlock_page(page);
1324 page_cache_release(page);
1325 1367
1326 return ret ? ret : copied; 1368 return ret ? ret : copied;
1327} 1369}
@@ -1332,7 +1374,7 @@ static int ext4_writeback_write_end(struct file *file,
1332 struct page *page, void *fsdata) 1374 struct page *page, void *fsdata)
1333{ 1375{
1334 handle_t *handle = ext4_journal_current_handle(); 1376 handle_t *handle = ext4_journal_current_handle();
1335 struct inode *inode = file->f_mapping->host; 1377 struct inode *inode = mapping->host;
1336 int ret = 0, ret2; 1378 int ret = 0, ret2;
1337 loff_t new_i_size; 1379 loff_t new_i_size;
1338 1380
@@ -1340,7 +1382,7 @@ static int ext4_writeback_write_end(struct file *file,
1340 if (new_i_size > EXT4_I(inode)->i_disksize) 1382 if (new_i_size > EXT4_I(inode)->i_disksize)
1341 EXT4_I(inode)->i_disksize = new_i_size; 1383 EXT4_I(inode)->i_disksize = new_i_size;
1342 1384
1343 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, 1385 ret2 = generic_write_end(file, mapping, pos, len, copied,
1344 page, fsdata); 1386 page, fsdata);
1345 copied = ret2; 1387 copied = ret2;
1346 if (ret2 < 0) 1388 if (ret2 < 0)
@@ -1349,8 +1391,6 @@ static int ext4_writeback_write_end(struct file *file,
1349 ret2 = ext4_journal_stop(handle); 1391 ret2 = ext4_journal_stop(handle);
1350 if (!ret) 1392 if (!ret)
1351 ret = ret2; 1393 ret = ret2;
1352 unlock_page(page);
1353 page_cache_release(page);
1354 1394
1355 return ret ? ret : copied; 1395 return ret ? ret : copied;
1356} 1396}
@@ -1389,14 +1429,965 @@ static int ext4_journalled_write_end(struct file *file,
1389 ret = ret2; 1429 ret = ret2;
1390 } 1430 }
1391 1431
1432 unlock_page(page);
1392 ret2 = ext4_journal_stop(handle); 1433 ret2 = ext4_journal_stop(handle);
1393 if (!ret) 1434 if (!ret)
1394 ret = ret2; 1435 ret = ret2;
1395 unlock_page(page);
1396 page_cache_release(page); 1436 page_cache_release(page);
1397 1437
1398 return ret ? ret : copied; 1438 return ret ? ret : copied;
1399} 1439}
1440/*
1441 * Calculate the number of metadata blocks need to reserve
1442 * to allocate @blocks for non extent file based file
1443 */
1444static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1445{
1446 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1447 int ind_blks, dind_blks, tind_blks;
1448
1449 /* number of new indirect blocks needed */
1450 ind_blks = (blocks + icap - 1) / icap;
1451
1452 dind_blks = (ind_blks + icap - 1) / icap;
1453
1454 tind_blks = 1;
1455
1456 return ind_blks + dind_blks + tind_blks;
1457}
1458
1459/*
1460 * Calculate the number of metadata blocks need to reserve
1461 * to allocate given number of blocks
1462 */
1463static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1464{
1465 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1466 return ext4_ext_calc_metadata_amount(inode, blocks);
1467
1468 return ext4_indirect_calc_metadata_amount(inode, blocks);
1469}
1470
1471static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1472{
1473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1474 unsigned long md_needed, mdblocks, total = 0;
1475
1476 /*
1477 * recalculate the amount of metadata blocks to reserve
1478 * in order to allocate nrblocks
1479 * worse case is one extent per block
1480 */
1481 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1482 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1483 mdblocks = ext4_calc_metadata_amount(inode, total);
1484 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1485
1486 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1487 total = md_needed + nrblocks;
1488
1489 if (ext4_has_free_blocks(sbi, total) < total) {
1490 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1491 return -ENOSPC;
1492 }
1493
1494 /* reduce fs free blocks counter */
1495 percpu_counter_sub(&sbi->s_freeblocks_counter, total);
1496
1497 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1498 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1499
1500 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1501 return 0; /* success */
1502}
1503
1504void ext4_da_release_space(struct inode *inode, int used, int to_free)
1505{
1506 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1507 int total, mdb, mdb_free, release;
1508
1509 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1510 /* recalculate the number of metablocks still need to be reserved */
1511 total = EXT4_I(inode)->i_reserved_data_blocks - used - to_free;
1512 mdb = ext4_calc_metadata_amount(inode, total);
1513
1514 /* figure out how many metablocks to release */
1515 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1516 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1517
1518 /* Account for allocated meta_blocks */
1519 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1520
1521 release = to_free + mdb_free;
1522
1523 /* update fs free blocks counter for truncate case */
1524 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1525
1526 /* update per-inode reservations */
1527 BUG_ON(used + to_free > EXT4_I(inode)->i_reserved_data_blocks);
1528 EXT4_I(inode)->i_reserved_data_blocks -= (used + to_free);
1529
1530 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1531 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1532 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1533 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1534}
1535
1536static void ext4_da_page_release_reservation(struct page *page,
1537 unsigned long offset)
1538{
1539 int to_release = 0;
1540 struct buffer_head *head, *bh;
1541 unsigned int curr_off = 0;
1542
1543 head = page_buffers(page);
1544 bh = head;
1545 do {
1546 unsigned int next_off = curr_off + bh->b_size;
1547
1548 if ((offset <= curr_off) && (buffer_delay(bh))) {
1549 to_release++;
1550 clear_buffer_delay(bh);
1551 }
1552 curr_off = next_off;
1553 } while ((bh = bh->b_this_page) != head);
1554 ext4_da_release_space(page->mapping->host, 0, to_release);
1555}
1556
1557/*
1558 * Delayed allocation stuff
1559 */
1560
1561struct mpage_da_data {
1562 struct inode *inode;
1563 struct buffer_head lbh; /* extent of blocks */
1564 unsigned long first_page, next_page; /* extent of pages */
1565 get_block_t *get_block;
1566 struct writeback_control *wbc;
1567};
1568
1569/*
1570 * mpage_da_submit_io - walks through extent of pages and try to write
1571 * them with __mpage_writepage()
1572 *
1573 * @mpd->inode: inode
1574 * @mpd->first_page: first page of the extent
1575 * @mpd->next_page: page after the last page of the extent
1576 * @mpd->get_block: the filesystem's block mapper function
1577 *
1578 * By the time mpage_da_submit_io() is called we expect all blocks
1579 * to be allocated. this may be wrong if allocation failed.
1580 *
1581 * As pages are already locked by write_cache_pages(), we can't use it
1582 */
1583static int mpage_da_submit_io(struct mpage_da_data *mpd)
1584{
1585 struct address_space *mapping = mpd->inode->i_mapping;
1586 struct mpage_data mpd_pp = {
1587 .bio = NULL,
1588 .last_block_in_bio = 0,
1589 .get_block = mpd->get_block,
1590 .use_writepage = 1,
1591 };
1592 int ret = 0, err, nr_pages, i;
1593 unsigned long index, end;
1594 struct pagevec pvec;
1595
1596 BUG_ON(mpd->next_page <= mpd->first_page);
1597
1598 pagevec_init(&pvec, 0);
1599 index = mpd->first_page;
1600 end = mpd->next_page - 1;
1601
1602 while (index <= end) {
1603 /* XXX: optimize tail */
1604 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1605 if (nr_pages == 0)
1606 break;
1607 for (i = 0; i < nr_pages; i++) {
1608 struct page *page = pvec.pages[i];
1609
1610 index = page->index;
1611 if (index > end)
1612 break;
1613 index++;
1614
1615 err = __mpage_writepage(page, mpd->wbc, &mpd_pp);
1616
1617 /*
1618 * In error case, we have to continue because
1619 * remaining pages are still locked
1620 * XXX: unlock and re-dirty them?
1621 */
1622 if (ret == 0)
1623 ret = err;
1624 }
1625 pagevec_release(&pvec);
1626 }
1627 if (mpd_pp.bio)
1628 mpage_bio_submit(WRITE, mpd_pp.bio);
1629
1630 return ret;
1631}
1632
1633/*
1634 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1635 *
1636 * @mpd->inode - inode to walk through
1637 * @exbh->b_blocknr - first block on a disk
1638 * @exbh->b_size - amount of space in bytes
1639 * @logical - first logical block to start assignment with
1640 *
1641 * the function goes through all passed space and put actual disk
1642 * block numbers into buffer heads, dropping BH_Delay
1643 */
1644static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1645 struct buffer_head *exbh)
1646{
1647 struct inode *inode = mpd->inode;
1648 struct address_space *mapping = inode->i_mapping;
1649 int blocks = exbh->b_size >> inode->i_blkbits;
1650 sector_t pblock = exbh->b_blocknr, cur_logical;
1651 struct buffer_head *head, *bh;
1652 unsigned long index, end;
1653 struct pagevec pvec;
1654 int nr_pages, i;
1655
1656 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1657 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1658 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1659
1660 pagevec_init(&pvec, 0);
1661
1662 while (index <= end) {
1663 /* XXX: optimize tail */
1664 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1665 if (nr_pages == 0)
1666 break;
1667 for (i = 0; i < nr_pages; i++) {
1668 struct page *page = pvec.pages[i];
1669
1670 index = page->index;
1671 if (index > end)
1672 break;
1673 index++;
1674
1675 BUG_ON(!PageLocked(page));
1676 BUG_ON(PageWriteback(page));
1677 BUG_ON(!page_has_buffers(page));
1678
1679 bh = page_buffers(page);
1680 head = bh;
1681
1682 /* skip blocks out of the range */
1683 do {
1684 if (cur_logical >= logical)
1685 break;
1686 cur_logical++;
1687 } while ((bh = bh->b_this_page) != head);
1688
1689 do {
1690 if (cur_logical >= logical + blocks)
1691 break;
1692 if (buffer_delay(bh)) {
1693 bh->b_blocknr = pblock;
1694 clear_buffer_delay(bh);
1695 } else if (buffer_mapped(bh))
1696 BUG_ON(bh->b_blocknr != pblock);
1697
1698 cur_logical++;
1699 pblock++;
1700 } while ((bh = bh->b_this_page) != head);
1701 }
1702 pagevec_release(&pvec);
1703 }
1704}
1705
1706
1707/*
1708 * __unmap_underlying_blocks - just a helper function to unmap
1709 * set of blocks described by @bh
1710 */
1711static inline void __unmap_underlying_blocks(struct inode *inode,
1712 struct buffer_head *bh)
1713{
1714 struct block_device *bdev = inode->i_sb->s_bdev;
1715 int blocks, i;
1716
1717 blocks = bh->b_size >> inode->i_blkbits;
1718 for (i = 0; i < blocks; i++)
1719 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1720}
1721
1722/*
1723 * mpage_da_map_blocks - go through given space
1724 *
1725 * @mpd->lbh - bh describing space
1726 * @mpd->get_block - the filesystem's block mapper function
1727 *
1728 * The function skips space we know is already mapped to disk blocks.
1729 *
1730 * The function ignores errors ->get_block() returns, thus real
1731 * error handling is postponed to __mpage_writepage()
1732 */
1733static void mpage_da_map_blocks(struct mpage_da_data *mpd)
1734{
1735 struct buffer_head *lbh = &mpd->lbh;
1736 int err = 0, remain = lbh->b_size;
1737 sector_t next = lbh->b_blocknr;
1738 struct buffer_head new;
1739
1740 /*
1741 * We consider only non-mapped and non-allocated blocks
1742 */
1743 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1744 return;
1745
1746 while (remain) {
1747 new.b_state = lbh->b_state;
1748 new.b_blocknr = 0;
1749 new.b_size = remain;
1750 err = mpd->get_block(mpd->inode, next, &new, 1);
1751 if (err) {
1752 /*
1753 * Rather than implement own error handling
1754 * here, we just leave remaining blocks
1755 * unallocated and try again with ->writepage()
1756 */
1757 break;
1758 }
1759 BUG_ON(new.b_size == 0);
1760
1761 if (buffer_new(&new))
1762 __unmap_underlying_blocks(mpd->inode, &new);
1763
1764 /*
1765 * If blocks are delayed marked, we need to
1766 * put actual blocknr and drop delayed bit
1767 */
1768 if (buffer_delay(lbh))
1769 mpage_put_bnr_to_bhs(mpd, next, &new);
1770
1771 /* go for the remaining blocks */
1772 next += new.b_size >> mpd->inode->i_blkbits;
1773 remain -= new.b_size;
1774 }
1775}
1776
1777#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | (1 << BH_Delay))
1778
1779/*
1780 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1781 *
1782 * @mpd->lbh - extent of blocks
1783 * @logical - logical number of the block in the file
1784 * @bh - bh of the block (used to access block's state)
1785 *
1786 * the function is used to collect contig. blocks in same state
1787 */
1788static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1789 sector_t logical, struct buffer_head *bh)
1790{
1791 struct buffer_head *lbh = &mpd->lbh;
1792 sector_t next;
1793
1794 next = lbh->b_blocknr + (lbh->b_size >> mpd->inode->i_blkbits);
1795
1796 /*
1797 * First block in the extent
1798 */
1799 if (lbh->b_size == 0) {
1800 lbh->b_blocknr = logical;
1801 lbh->b_size = bh->b_size;
1802 lbh->b_state = bh->b_state & BH_FLAGS;
1803 return;
1804 }
1805
1806 /*
1807 * Can we merge the block to our big extent?
1808 */
1809 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1810 lbh->b_size += bh->b_size;
1811 return;
1812 }
1813
1814 /*
1815 * We couldn't merge the block to our extent, so we
1816 * need to flush current extent and start new one
1817 */
1818 mpage_da_map_blocks(mpd);
1819
1820 /*
1821 * Now start a new extent
1822 */
1823 lbh->b_size = bh->b_size;
1824 lbh->b_state = bh->b_state & BH_FLAGS;
1825 lbh->b_blocknr = logical;
1826}
1827
1828/*
1829 * __mpage_da_writepage - finds extent of pages and blocks
1830 *
1831 * @page: page to consider
1832 * @wbc: not used, we just follow rules
1833 * @data: context
1834 *
1835 * The function finds extents of pages and scan them for all blocks.
1836 */
1837static int __mpage_da_writepage(struct page *page,
1838 struct writeback_control *wbc, void *data)
1839{
1840 struct mpage_da_data *mpd = data;
1841 struct inode *inode = mpd->inode;
1842 struct buffer_head *bh, *head, fake;
1843 sector_t logical;
1844
1845 /*
1846 * Can we merge this page to current extent?
1847 */
1848 if (mpd->next_page != page->index) {
1849 /*
1850 * Nope, we can't. So, we map non-allocated blocks
1851 * and start IO on them using __mpage_writepage()
1852 */
1853 if (mpd->next_page != mpd->first_page) {
1854 mpage_da_map_blocks(mpd);
1855 mpage_da_submit_io(mpd);
1856 }
1857
1858 /*
1859 * Start next extent of pages ...
1860 */
1861 mpd->first_page = page->index;
1862
1863 /*
1864 * ... and blocks
1865 */
1866 mpd->lbh.b_size = 0;
1867 mpd->lbh.b_state = 0;
1868 mpd->lbh.b_blocknr = 0;
1869 }
1870
1871 mpd->next_page = page->index + 1;
1872 logical = (sector_t) page->index <<
1873 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1874
1875 if (!page_has_buffers(page)) {
1876 /*
1877 * There is no attached buffer heads yet (mmap?)
1878 * we treat the page asfull of dirty blocks
1879 */
1880 bh = &fake;
1881 bh->b_size = PAGE_CACHE_SIZE;
1882 bh->b_state = 0;
1883 set_buffer_dirty(bh);
1884 set_buffer_uptodate(bh);
1885 mpage_add_bh_to_extent(mpd, logical, bh);
1886 } else {
1887 /*
1888 * Page with regular buffer heads, just add all dirty ones
1889 */
1890 head = page_buffers(page);
1891 bh = head;
1892 do {
1893 BUG_ON(buffer_locked(bh));
1894 if (buffer_dirty(bh))
1895 mpage_add_bh_to_extent(mpd, logical, bh);
1896 logical++;
1897 } while ((bh = bh->b_this_page) != head);
1898 }
1899
1900 return 0;
1901}
1902
1903/*
1904 * mpage_da_writepages - walk the list of dirty pages of the given
1905 * address space, allocates non-allocated blocks, maps newly-allocated
1906 * blocks to existing bhs and issue IO them
1907 *
1908 * @mapping: address space structure to write
1909 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1910 * @get_block: the filesystem's block mapper function.
1911 *
1912 * This is a library function, which implements the writepages()
1913 * address_space_operation.
1914 *
1915 * In order to avoid duplication of logic that deals with partial pages,
1916 * multiple bio per page, etc, we find non-allocated blocks, allocate
1917 * them with minimal calls to ->get_block() and re-use __mpage_writepage()
1918 *
1919 * It's important that we call __mpage_writepage() only once for each
1920 * involved page, otherwise we'd have to implement more complicated logic
1921 * to deal with pages w/o PG_lock or w/ PG_writeback and so on.
1922 *
1923 * See comments to mpage_writepages()
1924 */
1925static int mpage_da_writepages(struct address_space *mapping,
1926 struct writeback_control *wbc,
1927 get_block_t get_block)
1928{
1929 struct mpage_da_data mpd;
1930 int ret;
1931
1932 if (!get_block)
1933 return generic_writepages(mapping, wbc);
1934
1935 mpd.wbc = wbc;
1936 mpd.inode = mapping->host;
1937 mpd.lbh.b_size = 0;
1938 mpd.lbh.b_state = 0;
1939 mpd.lbh.b_blocknr = 0;
1940 mpd.first_page = 0;
1941 mpd.next_page = 0;
1942 mpd.get_block = get_block;
1943
1944 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
1945
1946 /*
1947 * Handle last extent of pages
1948 */
1949 if (mpd.next_page != mpd.first_page) {
1950 mpage_da_map_blocks(&mpd);
1951 mpage_da_submit_io(&mpd);
1952 }
1953
1954 return ret;
1955}
1956
1957/*
1958 * this is a special callback for ->write_begin() only
1959 * it's intention is to return mapped block or reserve space
1960 */
1961static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1962 struct buffer_head *bh_result, int create)
1963{
1964 int ret = 0;
1965
1966 BUG_ON(create == 0);
1967 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1968
1969 /*
1970 * first, we need to know whether the block is allocated already
1971 * preallocated blocks are unmapped but should treated
1972 * the same as allocated blocks.
1973 */
1974 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
1975 if ((ret == 0) && !buffer_delay(bh_result)) {
1976 /* the block isn't (pre)allocated yet, let's reserve space */
1977 /*
1978 * XXX: __block_prepare_write() unmaps passed block,
1979 * is it OK?
1980 */
1981 ret = ext4_da_reserve_space(inode, 1);
1982 if (ret)
1983 /* not enough space to reserve */
1984 return ret;
1985
1986 map_bh(bh_result, inode->i_sb, 0);
1987 set_buffer_new(bh_result);
1988 set_buffer_delay(bh_result);
1989 } else if (ret > 0) {
1990 bh_result->b_size = (ret << inode->i_blkbits);
1991 ret = 0;
1992 }
1993
1994 return ret;
1995}
1996#define EXT4_DELALLOC_RSVED 1
1997static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
1998 struct buffer_head *bh_result, int create)
1999{
2000 int ret;
2001 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2002 loff_t disksize = EXT4_I(inode)->i_disksize;
2003 handle_t *handle = NULL;
2004
2005 handle = ext4_journal_current_handle();
2006 if (!handle) {
2007 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2008 bh_result, 0, 0, 0);
2009 BUG_ON(!ret);
2010 } else {
2011 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2012 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2013 }
2014
2015 if (ret > 0) {
2016 bh_result->b_size = (ret << inode->i_blkbits);
2017
2018 /*
2019 * Update on-disk size along with block allocation
2020 * we don't use 'extend_disksize' as size may change
2021 * within already allocated block -bzzz
2022 */
2023 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2024 if (disksize > i_size_read(inode))
2025 disksize = i_size_read(inode);
2026 if (disksize > EXT4_I(inode)->i_disksize) {
2027 /*
2028 * XXX: replace with spinlock if seen contended -bzzz
2029 */
2030 down_write(&EXT4_I(inode)->i_data_sem);
2031 if (disksize > EXT4_I(inode)->i_disksize)
2032 EXT4_I(inode)->i_disksize = disksize;
2033 up_write(&EXT4_I(inode)->i_data_sem);
2034
2035 if (EXT4_I(inode)->i_disksize == disksize) {
2036 ret = ext4_mark_inode_dirty(handle, inode);
2037 return ret;
2038 }
2039 }
2040 ret = 0;
2041 }
2042 return ret;
2043}
2044
2045static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2046{
2047 /*
2048 * unmapped buffer is possible for holes.
2049 * delay buffer is possible with delayed allocation
2050 */
2051 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2052}
2053
2054static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2055 struct buffer_head *bh_result, int create)
2056{
2057 int ret = 0;
2058 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2059
2060 /*
2061 * we don't want to do block allocation in writepage
2062 * so call get_block_wrap with create = 0
2063 */
2064 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2065 bh_result, 0, 0, 0);
2066 if (ret > 0) {
2067 bh_result->b_size = (ret << inode->i_blkbits);
2068 ret = 0;
2069 }
2070 return ret;
2071}
2072
2073/*
2074 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2075 * get called via journal_submit_inode_data_buffers (no journal handle)
2076 * get called via shrink_page_list via pdflush (no journal handle)
2077 * or grab_page_cache when doing write_begin (have journal handle)
2078 */
2079static int ext4_da_writepage(struct page *page,
2080 struct writeback_control *wbc)
2081{
2082 int ret = 0;
2083 loff_t size;
2084 unsigned long len;
2085 struct buffer_head *page_bufs;
2086 struct inode *inode = page->mapping->host;
2087
2088 size = i_size_read(inode);
2089 if (page->index == size >> PAGE_CACHE_SHIFT)
2090 len = size & ~PAGE_CACHE_MASK;
2091 else
2092 len = PAGE_CACHE_SIZE;
2093
2094 if (page_has_buffers(page)) {
2095 page_bufs = page_buffers(page);
2096 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2097 ext4_bh_unmapped_or_delay)) {
2098 /*
2099 * We don't want to do block allocation
2100 * So redirty the page and return
2101 * We may reach here when we do a journal commit
2102 * via journal_submit_inode_data_buffers.
2103 * If we don't have mapping block we just ignore
2104 * them. We can also reach here via shrink_page_list
2105 */
2106 redirty_page_for_writepage(wbc, page);
2107 unlock_page(page);
2108 return 0;
2109 }
2110 } else {
2111 /*
2112 * The test for page_has_buffers() is subtle:
2113 * We know the page is dirty but it lost buffers. That means
2114 * that at some moment in time after write_begin()/write_end()
2115 * has been called all buffers have been clean and thus they
2116 * must have been written at least once. So they are all
2117 * mapped and we can happily proceed with mapping them
2118 * and writing the page.
2119 *
2120 * Try to initialize the buffer_heads and check whether
2121 * all are mapped and non delay. We don't want to
2122 * do block allocation here.
2123 */
2124 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2125 ext4_normal_get_block_write);
2126 if (!ret) {
2127 page_bufs = page_buffers(page);
2128 /* check whether all are mapped and non delay */
2129 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2130 ext4_bh_unmapped_or_delay)) {
2131 redirty_page_for_writepage(wbc, page);
2132 unlock_page(page);
2133 return 0;
2134 }
2135 } else {
2136 /*
2137 * We can't do block allocation here
2138 * so just redity the page and unlock
2139 * and return
2140 */
2141 redirty_page_for_writepage(wbc, page);
2142 unlock_page(page);
2143 return 0;
2144 }
2145 }
2146
2147 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2148 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2149 else
2150 ret = block_write_full_page(page,
2151 ext4_normal_get_block_write,
2152 wbc);
2153
2154 return ret;
2155}
2156
2157/*
2158 * For now just follow the DIO way to estimate the max credits
2159 * needed to write out EXT4_MAX_WRITEBACK_PAGES.
2160 * todo: need to calculate the max credits need for
2161 * extent based files, currently the DIO credits is based on
2162 * indirect-blocks mapping way.
2163 *
2164 * Probably should have a generic way to calculate credits
2165 * for DIO, writepages, and truncate
2166 */
2167#define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS
2168#define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS
2169
2170static int ext4_da_writepages(struct address_space *mapping,
2171 struct writeback_control *wbc)
2172{
2173 struct inode *inode = mapping->host;
2174 handle_t *handle = NULL;
2175 int needed_blocks;
2176 int ret = 0;
2177 long to_write;
2178 loff_t range_start = 0;
2179
2180 /*
2181 * No pages to write? This is mainly a kludge to avoid starting
2182 * a transaction for special inodes like journal inode on last iput()
2183 * because that could violate lock ordering on umount
2184 */
2185 if (!mapping->nrpages)
2186 return 0;
2187
2188 /*
2189 * Estimate the worse case needed credits to write out
2190 * EXT4_MAX_BUF_BLOCKS pages
2191 */
2192 needed_blocks = EXT4_MAX_WRITEBACK_CREDITS;
2193
2194 to_write = wbc->nr_to_write;
2195 if (!wbc->range_cyclic) {
2196 /*
2197 * If range_cyclic is not set force range_cont
2198 * and save the old writeback_index
2199 */
2200 wbc->range_cont = 1;
2201 range_start = wbc->range_start;
2202 }
2203
2204 while (!ret && to_write) {
2205 /* start a new transaction*/
2206 handle = ext4_journal_start(inode, needed_blocks);
2207 if (IS_ERR(handle)) {
2208 ret = PTR_ERR(handle);
2209 goto out_writepages;
2210 }
2211 if (ext4_should_order_data(inode)) {
2212 /*
2213 * With ordered mode we need to add
2214 * the inode to the journal handle
2215 * when we do block allocation.
2216 */
2217 ret = ext4_jbd2_file_inode(handle, inode);
2218 if (ret) {
2219 ext4_journal_stop(handle);
2220 goto out_writepages;
2221 }
2222
2223 }
2224 /*
2225 * set the max dirty pages could be write at a time
2226 * to fit into the reserved transaction credits
2227 */
2228 if (wbc->nr_to_write > EXT4_MAX_WRITEBACK_PAGES)
2229 wbc->nr_to_write = EXT4_MAX_WRITEBACK_PAGES;
2230
2231 to_write -= wbc->nr_to_write;
2232 ret = mpage_da_writepages(mapping, wbc,
2233 ext4_da_get_block_write);
2234 ext4_journal_stop(handle);
2235 if (wbc->nr_to_write) {
2236 /*
2237 * There is no more writeout needed
2238 * or we requested for a noblocking writeout
2239 * and we found the device congested
2240 */
2241 to_write += wbc->nr_to_write;
2242 break;
2243 }
2244 wbc->nr_to_write = to_write;
2245 }
2246
2247out_writepages:
2248 wbc->nr_to_write = to_write;
2249 if (range_start)
2250 wbc->range_start = range_start;
2251 return ret;
2252}
2253
2254static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2255 loff_t pos, unsigned len, unsigned flags,
2256 struct page **pagep, void **fsdata)
2257{
2258 int ret, retries = 0;
2259 struct page *page;
2260 pgoff_t index;
2261 unsigned from, to;
2262 struct inode *inode = mapping->host;
2263 handle_t *handle;
2264
2265 index = pos >> PAGE_CACHE_SHIFT;
2266 from = pos & (PAGE_CACHE_SIZE - 1);
2267 to = from + len;
2268
2269retry:
2270 /*
2271 * With delayed allocation, we don't log the i_disksize update
2272 * if there is delayed block allocation. But we still need
2273 * to journalling the i_disksize update if writes to the end
2274 * of file which has an already mapped buffer.
2275 */
2276 handle = ext4_journal_start(inode, 1);
2277 if (IS_ERR(handle)) {
2278 ret = PTR_ERR(handle);
2279 goto out;
2280 }
2281
2282 page = __grab_cache_page(mapping, index);
2283 if (!page)
2284 return -ENOMEM;
2285 *pagep = page;
2286
2287 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2288 ext4_da_get_block_prep);
2289 if (ret < 0) {
2290 unlock_page(page);
2291 ext4_journal_stop(handle);
2292 page_cache_release(page);
2293 }
2294
2295 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2296 goto retry;
2297out:
2298 return ret;
2299}
2300
2301/*
2302 * Check if we should update i_disksize
2303 * when write to the end of file but not require block allocation
2304 */
2305static int ext4_da_should_update_i_disksize(struct page *page,
2306 unsigned long offset)
2307{
2308 struct buffer_head *bh;
2309 struct inode *inode = page->mapping->host;
2310 unsigned int idx;
2311 int i;
2312
2313 bh = page_buffers(page);
2314 idx = offset >> inode->i_blkbits;
2315
2316 for (i=0; i < idx; i++)
2317 bh = bh->b_this_page;
2318
2319 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2320 return 0;
2321 return 1;
2322}
2323
2324static int ext4_da_write_end(struct file *file,
2325 struct address_space *mapping,
2326 loff_t pos, unsigned len, unsigned copied,
2327 struct page *page, void *fsdata)
2328{
2329 struct inode *inode = mapping->host;
2330 int ret = 0, ret2;
2331 handle_t *handle = ext4_journal_current_handle();
2332 loff_t new_i_size;
2333 unsigned long start, end;
2334
2335 start = pos & (PAGE_CACHE_SIZE - 1);
2336 end = start + copied -1;
2337
2338 /*
2339 * generic_write_end() will run mark_inode_dirty() if i_size
2340 * changes. So let's piggyback the i_disksize mark_inode_dirty
2341 * into that.
2342 */
2343
2344 new_i_size = pos + copied;
2345 if (new_i_size > EXT4_I(inode)->i_disksize) {
2346 if (ext4_da_should_update_i_disksize(page, end)) {
2347 down_write(&EXT4_I(inode)->i_data_sem);
2348 if (new_i_size > EXT4_I(inode)->i_disksize) {
2349 /*
2350 * Updating i_disksize when extending file
2351 * without needing block allocation
2352 */
2353 if (ext4_should_order_data(inode))
2354 ret = ext4_jbd2_file_inode(handle,
2355 inode);
2356
2357 EXT4_I(inode)->i_disksize = new_i_size;
2358 }
2359 up_write(&EXT4_I(inode)->i_data_sem);
2360 }
2361 }
2362 ret2 = generic_write_end(file, mapping, pos, len, copied,
2363 page, fsdata);
2364 copied = ret2;
2365 if (ret2 < 0)
2366 ret = ret2;
2367 ret2 = ext4_journal_stop(handle);
2368 if (!ret)
2369 ret = ret2;
2370
2371 return ret ? ret : copied;
2372}
2373
2374static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2375{
2376 /*
2377 * Drop reserved blocks
2378 */
2379 BUG_ON(!PageLocked(page));
2380 if (!page_has_buffers(page))
2381 goto out;
2382
2383 ext4_da_page_release_reservation(page, offset);
2384
2385out:
2386 ext4_invalidatepage(page, offset);
2387
2388 return;
2389}
2390
1400 2391
1401/* 2392/*
1402 * bmap() is special. It gets used by applications such as lilo and by 2393 * bmap() is special. It gets used by applications such as lilo and by
@@ -1418,6 +2409,16 @@ static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
1418 journal_t *journal; 2409 journal_t *journal;
1419 int err; 2410 int err;
1420 2411
2412 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2413 test_opt(inode->i_sb, DELALLOC)) {
2414 /*
2415 * With delalloc we want to sync the file
2416 * so that we can make sure we allocate
2417 * blocks for file
2418 */
2419 filemap_write_and_wait(mapping);
2420 }
2421
1421 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { 2422 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
1422 /* 2423 /*
1423 * This is a REALLY heavyweight approach, but the use of 2424 * This is a REALLY heavyweight approach, but the use of
@@ -1462,21 +2463,17 @@ static int bput_one(handle_t *handle, struct buffer_head *bh)
1462 return 0; 2463 return 0;
1463} 2464}
1464 2465
1465static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1466{
1467 if (buffer_mapped(bh))
1468 return ext4_journal_dirty_data(handle, bh);
1469 return 0;
1470}
1471
1472/* 2466/*
1473 * Note that we always start a transaction even if we're not journalling 2467 * Note that we don't need to start a transaction unless we're journaling data
1474 * data. This is to preserve ordering: any hole instantiation within 2468 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1475 * __block_write_full_page -> ext4_get_block() should be journalled 2469 * need to file the inode to the transaction's list in ordered mode because if
1476 * along with the data so we don't crash and then get metadata which 2470 * we are writing back data added by write(), the inode is already there and if
1477 * refers to old data. 2471 * we are writing back data modified via mmap(), noone guarantees in which
2472 * transaction the data will hit the disk. In case we are journaling data, we
2473 * cannot start transaction directly because transaction start ranks above page
2474 * lock so we have to do some magic.
1478 * 2475 *
1479 * In all journalling modes block_write_full_page() will start the I/O. 2476 * In all journaling modes block_write_full_page() will start the I/O.
1480 * 2477 *
1481 * Problem: 2478 * Problem:
1482 * 2479 *
@@ -1518,105 +2515,103 @@ static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1518 * disastrous. Any write() or metadata operation will sync the fs for 2515 * disastrous. Any write() or metadata operation will sync the fs for
1519 * us. 2516 * us.
1520 * 2517 *
1521 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1522 * we don't need to open a transaction here.
1523 */ 2518 */
1524static int ext4_ordered_writepage(struct page *page, 2519static int __ext4_normal_writepage(struct page *page,
1525 struct writeback_control *wbc) 2520 struct writeback_control *wbc)
1526{ 2521{
1527 struct inode *inode = page->mapping->host; 2522 struct inode *inode = page->mapping->host;
1528 struct buffer_head *page_bufs;
1529 handle_t *handle = NULL;
1530 int ret = 0;
1531 int err;
1532
1533 J_ASSERT(PageLocked(page));
1534
1535 /*
1536 * We give up here if we're reentered, because it might be for a
1537 * different filesystem.
1538 */
1539 if (ext4_journal_current_handle())
1540 goto out_fail;
1541 2523
1542 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); 2524 if (test_opt(inode->i_sb, NOBH))
2525 return nobh_writepage(page,
2526 ext4_normal_get_block_write, wbc);
2527 else
2528 return block_write_full_page(page,
2529 ext4_normal_get_block_write,
2530 wbc);
2531}
1543 2532
1544 if (IS_ERR(handle)) { 2533static int ext4_normal_writepage(struct page *page,
1545 ret = PTR_ERR(handle); 2534 struct writeback_control *wbc)
1546 goto out_fail; 2535{
1547 } 2536 struct inode *inode = page->mapping->host;
2537 loff_t size = i_size_read(inode);
2538 loff_t len;
1548 2539
1549 if (!page_has_buffers(page)) { 2540 J_ASSERT(PageLocked(page));
1550 create_empty_buffers(page, inode->i_sb->s_blocksize, 2541 if (page->index == size >> PAGE_CACHE_SHIFT)
1551 (1 << BH_Dirty)|(1 << BH_Uptodate)); 2542 len = size & ~PAGE_CACHE_MASK;
2543 else
2544 len = PAGE_CACHE_SIZE;
2545
2546 if (page_has_buffers(page)) {
2547 /* if page has buffers it should all be mapped
2548 * and allocated. If there are not buffers attached
2549 * to the page we know the page is dirty but it lost
2550 * buffers. That means that at some moment in time
2551 * after write_begin() / write_end() has been called
2552 * all buffers have been clean and thus they must have been
2553 * written at least once. So they are all mapped and we can
2554 * happily proceed with mapping them and writing the page.
2555 */
2556 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2557 ext4_bh_unmapped_or_delay));
1552 } 2558 }
1553 page_bufs = page_buffers(page);
1554 walk_page_buffers(handle, page_bufs, 0,
1555 PAGE_CACHE_SIZE, NULL, bget_one);
1556
1557 ret = block_write_full_page(page, ext4_get_block, wbc);
1558 2559
1559 /* 2560 if (!ext4_journal_current_handle())
1560 * The page can become unlocked at any point now, and 2561 return __ext4_normal_writepage(page, wbc);
1561 * truncate can then come in and change things. So we
1562 * can't touch *page from now on. But *page_bufs is
1563 * safe due to elevated refcount.
1564 */
1565 2562
1566 /*
1567 * And attach them to the current transaction. But only if
1568 * block_write_full_page() succeeded. Otherwise they are unmapped,
1569 * and generally junk.
1570 */
1571 if (ret == 0) {
1572 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1573 NULL, jbd2_journal_dirty_data_fn);
1574 if (!ret)
1575 ret = err;
1576 }
1577 walk_page_buffers(handle, page_bufs, 0,
1578 PAGE_CACHE_SIZE, NULL, bput_one);
1579 err = ext4_journal_stop(handle);
1580 if (!ret)
1581 ret = err;
1582 return ret;
1583
1584out_fail:
1585 redirty_page_for_writepage(wbc, page); 2563 redirty_page_for_writepage(wbc, page);
1586 unlock_page(page); 2564 unlock_page(page);
1587 return ret; 2565 return 0;
1588} 2566}
1589 2567
1590static int ext4_writeback_writepage(struct page *page, 2568static int __ext4_journalled_writepage(struct page *page,
1591 struct writeback_control *wbc) 2569 struct writeback_control *wbc)
1592{ 2570{
1593 struct inode *inode = page->mapping->host; 2571 struct address_space *mapping = page->mapping;
2572 struct inode *inode = mapping->host;
2573 struct buffer_head *page_bufs;
1594 handle_t *handle = NULL; 2574 handle_t *handle = NULL;
1595 int ret = 0; 2575 int ret = 0;
1596 int err; 2576 int err;
1597 2577
1598 if (ext4_journal_current_handle()) 2578 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1599 goto out_fail; 2579 ext4_normal_get_block_write);
2580 if (ret != 0)
2581 goto out_unlock;
2582
2583 page_bufs = page_buffers(page);
2584 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2585 bget_one);
2586 /* As soon as we unlock the page, it can go away, but we have
2587 * references to buffers so we are safe */
2588 unlock_page(page);
1600 2589
1601 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); 2590 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1602 if (IS_ERR(handle)) { 2591 if (IS_ERR(handle)) {
1603 ret = PTR_ERR(handle); 2592 ret = PTR_ERR(handle);
1604 goto out_fail; 2593 goto out;
1605 } 2594 }
1606 2595
1607 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode)) 2596 ret = walk_page_buffers(handle, page_bufs, 0,
1608 ret = nobh_writepage(page, ext4_get_block, wbc); 2597 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1609 else
1610 ret = block_write_full_page(page, ext4_get_block, wbc);
1611 2598
2599 err = walk_page_buffers(handle, page_bufs, 0,
2600 PAGE_CACHE_SIZE, NULL, write_end_fn);
2601 if (ret == 0)
2602 ret = err;
1612 err = ext4_journal_stop(handle); 2603 err = ext4_journal_stop(handle);
1613 if (!ret) 2604 if (!ret)
1614 ret = err; 2605 ret = err;
1615 return ret;
1616 2606
1617out_fail: 2607 walk_page_buffers(handle, page_bufs, 0,
1618 redirty_page_for_writepage(wbc, page); 2608 PAGE_CACHE_SIZE, NULL, bput_one);
2609 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2610 goto out;
2611
2612out_unlock:
1619 unlock_page(page); 2613 unlock_page(page);
2614out:
1620 return ret; 2615 return ret;
1621} 2616}
1622 2617
@@ -1624,59 +2619,53 @@ static int ext4_journalled_writepage(struct page *page,
1624 struct writeback_control *wbc) 2619 struct writeback_control *wbc)
1625{ 2620{
1626 struct inode *inode = page->mapping->host; 2621 struct inode *inode = page->mapping->host;
1627 handle_t *handle = NULL; 2622 loff_t size = i_size_read(inode);
1628 int ret = 0; 2623 loff_t len;
1629 int err;
1630 2624
1631 if (ext4_journal_current_handle()) 2625 J_ASSERT(PageLocked(page));
1632 goto no_write; 2626 if (page->index == size >> PAGE_CACHE_SHIFT)
2627 len = size & ~PAGE_CACHE_MASK;
2628 else
2629 len = PAGE_CACHE_SIZE;
2630
2631 if (page_has_buffers(page)) {
2632 /* if page has buffers it should all be mapped
2633 * and allocated. If there are not buffers attached
2634 * to the page we know the page is dirty but it lost
2635 * buffers. That means that at some moment in time
2636 * after write_begin() / write_end() has been called
2637 * all buffers have been clean and thus they must have been
2638 * written at least once. So they are all mapped and we can
2639 * happily proceed with mapping them and writing the page.
2640 */
2641 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2642 ext4_bh_unmapped_or_delay));
2643 }
1633 2644
1634 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); 2645 if (ext4_journal_current_handle())
1635 if (IS_ERR(handle)) {
1636 ret = PTR_ERR(handle);
1637 goto no_write; 2646 goto no_write;
1638 }
1639 2647
1640 if (!page_has_buffers(page) || PageChecked(page)) { 2648 if (PageChecked(page)) {
1641 /* 2649 /*
1642 * It's mmapped pagecache. Add buffers and journal it. There 2650 * It's mmapped pagecache. Add buffers and journal it. There
1643 * doesn't seem much point in redirtying the page here. 2651 * doesn't seem much point in redirtying the page here.
1644 */ 2652 */
1645 ClearPageChecked(page); 2653 ClearPageChecked(page);
1646 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, 2654 return __ext4_journalled_writepage(page, wbc);
1647 ext4_get_block);
1648 if (ret != 0) {
1649 ext4_journal_stop(handle);
1650 goto out_unlock;
1651 }
1652 ret = walk_page_buffers(handle, page_buffers(page), 0,
1653 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1654
1655 err = walk_page_buffers(handle, page_buffers(page), 0,
1656 PAGE_CACHE_SIZE, NULL, write_end_fn);
1657 if (ret == 0)
1658 ret = err;
1659 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1660 unlock_page(page);
1661 } else { 2655 } else {
1662 /* 2656 /*
1663 * It may be a page full of checkpoint-mode buffers. We don't 2657 * It may be a page full of checkpoint-mode buffers. We don't
1664 * really know unless we go poke around in the buffer_heads. 2658 * really know unless we go poke around in the buffer_heads.
1665 * But block_write_full_page will do the right thing. 2659 * But block_write_full_page will do the right thing.
1666 */ 2660 */
1667 ret = block_write_full_page(page, ext4_get_block, wbc); 2661 return block_write_full_page(page,
2662 ext4_normal_get_block_write,
2663 wbc);
1668 } 2664 }
1669 err = ext4_journal_stop(handle);
1670 if (!ret)
1671 ret = err;
1672out:
1673 return ret;
1674
1675no_write: 2665no_write:
1676 redirty_page_for_writepage(wbc, page); 2666 redirty_page_for_writepage(wbc, page);
1677out_unlock:
1678 unlock_page(page); 2667 unlock_page(page);
1679 goto out; 2668 return 0;
1680} 2669}
1681 2670
1682static int ext4_readpage(struct file *file, struct page *page) 2671static int ext4_readpage(struct file *file, struct page *page)
@@ -1819,7 +2808,7 @@ static int ext4_journalled_set_page_dirty(struct page *page)
1819static const struct address_space_operations ext4_ordered_aops = { 2808static const struct address_space_operations ext4_ordered_aops = {
1820 .readpage = ext4_readpage, 2809 .readpage = ext4_readpage,
1821 .readpages = ext4_readpages, 2810 .readpages = ext4_readpages,
1822 .writepage = ext4_ordered_writepage, 2811 .writepage = ext4_normal_writepage,
1823 .sync_page = block_sync_page, 2812 .sync_page = block_sync_page,
1824 .write_begin = ext4_write_begin, 2813 .write_begin = ext4_write_begin,
1825 .write_end = ext4_ordered_write_end, 2814 .write_end = ext4_ordered_write_end,
@@ -1833,7 +2822,7 @@ static const struct address_space_operations ext4_ordered_aops = {
1833static const struct address_space_operations ext4_writeback_aops = { 2822static const struct address_space_operations ext4_writeback_aops = {
1834 .readpage = ext4_readpage, 2823 .readpage = ext4_readpage,
1835 .readpages = ext4_readpages, 2824 .readpages = ext4_readpages,
1836 .writepage = ext4_writeback_writepage, 2825 .writepage = ext4_normal_writepage,
1837 .sync_page = block_sync_page, 2826 .sync_page = block_sync_page,
1838 .write_begin = ext4_write_begin, 2827 .write_begin = ext4_write_begin,
1839 .write_end = ext4_writeback_write_end, 2828 .write_end = ext4_writeback_write_end,
@@ -1857,10 +2846,31 @@ static const struct address_space_operations ext4_journalled_aops = {
1857 .releasepage = ext4_releasepage, 2846 .releasepage = ext4_releasepage,
1858}; 2847};
1859 2848
2849static const struct address_space_operations ext4_da_aops = {
2850 .readpage = ext4_readpage,
2851 .readpages = ext4_readpages,
2852 .writepage = ext4_da_writepage,
2853 .writepages = ext4_da_writepages,
2854 .sync_page = block_sync_page,
2855 .write_begin = ext4_da_write_begin,
2856 .write_end = ext4_da_write_end,
2857 .bmap = ext4_bmap,
2858 .invalidatepage = ext4_da_invalidatepage,
2859 .releasepage = ext4_releasepage,
2860 .direct_IO = ext4_direct_IO,
2861 .migratepage = buffer_migrate_page,
2862};
2863
1860void ext4_set_aops(struct inode *inode) 2864void ext4_set_aops(struct inode *inode)
1861{ 2865{
1862 if (ext4_should_order_data(inode)) 2866 if (ext4_should_order_data(inode) &&
2867 test_opt(inode->i_sb, DELALLOC))
2868 inode->i_mapping->a_ops = &ext4_da_aops;
2869 else if (ext4_should_order_data(inode))
1863 inode->i_mapping->a_ops = &ext4_ordered_aops; 2870 inode->i_mapping->a_ops = &ext4_ordered_aops;
2871 else if (ext4_should_writeback_data(inode) &&
2872 test_opt(inode->i_sb, DELALLOC))
2873 inode->i_mapping->a_ops = &ext4_da_aops;
1864 else if (ext4_should_writeback_data(inode)) 2874 else if (ext4_should_writeback_data(inode))
1865 inode->i_mapping->a_ops = &ext4_writeback_aops; 2875 inode->i_mapping->a_ops = &ext4_writeback_aops;
1866 else 2876 else
@@ -1873,7 +2883,7 @@ void ext4_set_aops(struct inode *inode)
1873 * This required during truncate. We need to physically zero the tail end 2883 * This required during truncate. We need to physically zero the tail end
1874 * of that block so it doesn't yield old data if the file is later grown. 2884 * of that block so it doesn't yield old data if the file is later grown.
1875 */ 2885 */
1876int ext4_block_truncate_page(handle_t *handle, struct page *page, 2886int ext4_block_truncate_page(handle_t *handle,
1877 struct address_space *mapping, loff_t from) 2887 struct address_space *mapping, loff_t from)
1878{ 2888{
1879 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; 2889 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
@@ -1882,8 +2892,13 @@ int ext4_block_truncate_page(handle_t *handle, struct page *page,
1882 ext4_lblk_t iblock; 2892 ext4_lblk_t iblock;
1883 struct inode *inode = mapping->host; 2893 struct inode *inode = mapping->host;
1884 struct buffer_head *bh; 2894 struct buffer_head *bh;
2895 struct page *page;
1885 int err = 0; 2896 int err = 0;
1886 2897
2898 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
2899 if (!page)
2900 return -EINVAL;
2901
1887 blocksize = inode->i_sb->s_blocksize; 2902 blocksize = inode->i_sb->s_blocksize;
1888 length = blocksize - (offset & (blocksize - 1)); 2903 length = blocksize - (offset & (blocksize - 1));
1889 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); 2904 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
@@ -1956,7 +2971,7 @@ int ext4_block_truncate_page(handle_t *handle, struct page *page,
1956 err = ext4_journal_dirty_metadata(handle, bh); 2971 err = ext4_journal_dirty_metadata(handle, bh);
1957 } else { 2972 } else {
1958 if (ext4_should_order_data(inode)) 2973 if (ext4_should_order_data(inode))
1959 err = ext4_journal_dirty_data(handle, bh); 2974 err = ext4_jbd2_file_inode(handle, inode);
1960 mark_buffer_dirty(bh); 2975 mark_buffer_dirty(bh);
1961 } 2976 }
1962 2977
@@ -2179,7 +3194,21 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
2179 3194
2180 if (this_bh) { 3195 if (this_bh) {
2181 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); 3196 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2182 ext4_journal_dirty_metadata(handle, this_bh); 3197
3198 /*
3199 * The buffer head should have an attached journal head at this
3200 * point. However, if the data is corrupted and an indirect
3201 * block pointed to itself, it would have been detached when
3202 * the block was cleared. Check for this instead of OOPSing.
3203 */
3204 if (bh2jh(this_bh))
3205 ext4_journal_dirty_metadata(handle, this_bh);
3206 else
3207 ext4_error(inode->i_sb, __func__,
3208 "circular indirect block detected, "
3209 "inode=%lu, block=%llu",
3210 inode->i_ino,
3211 (unsigned long long) this_bh->b_blocknr);
2183 } 3212 }
2184} 3213}
2185 3214
@@ -2305,6 +3334,19 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
2305 } 3334 }
2306} 3335}
2307 3336
3337int ext4_can_truncate(struct inode *inode)
3338{
3339 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3340 return 0;
3341 if (S_ISREG(inode->i_mode))
3342 return 1;
3343 if (S_ISDIR(inode->i_mode))
3344 return 1;
3345 if (S_ISLNK(inode->i_mode))
3346 return !ext4_inode_is_fast_symlink(inode);
3347 return 0;
3348}
3349
2308/* 3350/*
2309 * ext4_truncate() 3351 * ext4_truncate()
2310 * 3352 *
@@ -2347,51 +3389,25 @@ void ext4_truncate(struct inode *inode)
2347 int n; 3389 int n;
2348 ext4_lblk_t last_block; 3390 ext4_lblk_t last_block;
2349 unsigned blocksize = inode->i_sb->s_blocksize; 3391 unsigned blocksize = inode->i_sb->s_blocksize;
2350 struct page *page;
2351 3392
2352 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 3393 if (!ext4_can_truncate(inode))
2353 S_ISLNK(inode->i_mode)))
2354 return;
2355 if (ext4_inode_is_fast_symlink(inode))
2356 return;
2357 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2358 return; 3394 return;
2359 3395
2360 /*
2361 * We have to lock the EOF page here, because lock_page() nests
2362 * outside jbd2_journal_start().
2363 */
2364 if ((inode->i_size & (blocksize - 1)) == 0) {
2365 /* Block boundary? Nothing to do */
2366 page = NULL;
2367 } else {
2368 page = grab_cache_page(mapping,
2369 inode->i_size >> PAGE_CACHE_SHIFT);
2370 if (!page)
2371 return;
2372 }
2373
2374 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) { 3396 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
2375 ext4_ext_truncate(inode, page); 3397 ext4_ext_truncate(inode);
2376 return; 3398 return;
2377 } 3399 }
2378 3400
2379 handle = start_transaction(inode); 3401 handle = start_transaction(inode);
2380 if (IS_ERR(handle)) { 3402 if (IS_ERR(handle))
2381 if (page) {
2382 clear_highpage(page);
2383 flush_dcache_page(page);
2384 unlock_page(page);
2385 page_cache_release(page);
2386 }
2387 return; /* AKPM: return what? */ 3403 return; /* AKPM: return what? */
2388 }
2389 3404
2390 last_block = (inode->i_size + blocksize-1) 3405 last_block = (inode->i_size + blocksize-1)
2391 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb); 3406 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
2392 3407
2393 if (page) 3408 if (inode->i_size & (blocksize - 1))
2394 ext4_block_truncate_page(handle, page, mapping, inode->i_size); 3409 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3410 goto out_stop;
2395 3411
2396 n = ext4_block_to_path(inode, last_block, offsets, NULL); 3412 n = ext4_block_to_path(inode, last_block, offsets, NULL);
2397 if (n == 0) 3413 if (n == 0)
@@ -2410,6 +3426,11 @@ void ext4_truncate(struct inode *inode)
2410 goto out_stop; 3426 goto out_stop;
2411 3427
2412 /* 3428 /*
3429 * From here we block out all ext4_get_block() callers who want to
3430 * modify the block allocation tree.
3431 */
3432 down_write(&ei->i_data_sem);
3433 /*
2413 * The orphan list entry will now protect us from any crash which 3434 * The orphan list entry will now protect us from any crash which
2414 * occurs before the truncate completes, so it is now safe to propagate 3435 * occurs before the truncate completes, so it is now safe to propagate
2415 * the new, shorter inode size (held for now in i_size) into the 3436 * the new, shorter inode size (held for now in i_size) into the
@@ -2418,12 +3439,6 @@ void ext4_truncate(struct inode *inode)
2418 */ 3439 */
2419 ei->i_disksize = inode->i_size; 3440 ei->i_disksize = inode->i_size;
2420 3441
2421 /*
2422 * From here we block out all ext4_get_block() callers who want to
2423 * modify the block allocation tree.
2424 */
2425 down_write(&ei->i_data_sem);
2426
2427 if (n == 1) { /* direct blocks */ 3442 if (n == 1) { /* direct blocks */
2428 ext4_free_data(handle, inode, NULL, i_data+offsets[0], 3443 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2429 i_data + EXT4_NDIR_BLOCKS); 3444 i_data + EXT4_NDIR_BLOCKS);
@@ -3107,7 +4122,14 @@ int ext4_write_inode(struct inode *inode, int wait)
3107 * be freed, so we have a strong guarantee that no future commit will 4122 * be freed, so we have a strong guarantee that no future commit will
3108 * leave these blocks visible to the user.) 4123 * leave these blocks visible to the user.)
3109 * 4124 *
3110 * Called with inode->sem down. 4125 * Another thing we have to assure is that if we are in ordered mode
4126 * and inode is still attached to the committing transaction, we must
4127 * we start writeout of all the dirty pages which are being truncated.
4128 * This way we are sure that all the data written in the previous
4129 * transaction are already on disk (truncate waits for pages under
4130 * writeback).
4131 *
4132 * Called with inode->i_mutex down.
3111 */ 4133 */
3112int ext4_setattr(struct dentry *dentry, struct iattr *attr) 4134int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3113{ 4135{
@@ -3173,6 +4195,22 @@ int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3173 if (!error) 4195 if (!error)
3174 error = rc; 4196 error = rc;
3175 ext4_journal_stop(handle); 4197 ext4_journal_stop(handle);
4198
4199 if (ext4_should_order_data(inode)) {
4200 error = ext4_begin_ordered_truncate(inode,
4201 attr->ia_size);
4202 if (error) {
4203 /* Do as much error cleanup as possible */
4204 handle = ext4_journal_start(inode, 3);
4205 if (IS_ERR(handle)) {
4206 ext4_orphan_del(NULL, inode);
4207 goto err_out;
4208 }
4209 ext4_orphan_del(handle, inode);
4210 ext4_journal_stop(handle);
4211 goto err_out;
4212 }
4213 }
3176 } 4214 }
3177 4215
3178 rc = inode_setattr(inode, attr); 4216 rc = inode_setattr(inode, attr);
@@ -3193,6 +4231,32 @@ err_out:
3193 return error; 4231 return error;
3194} 4232}
3195 4233
4234int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4235 struct kstat *stat)
4236{
4237 struct inode *inode;
4238 unsigned long delalloc_blocks;
4239
4240 inode = dentry->d_inode;
4241 generic_fillattr(inode, stat);
4242
4243 /*
4244 * We can't update i_blocks if the block allocation is delayed
4245 * otherwise in the case of system crash before the real block
4246 * allocation is done, we will have i_blocks inconsistent with
4247 * on-disk file blocks.
4248 * We always keep i_blocks updated together with real
4249 * allocation. But to not confuse with user, stat
4250 * will return the blocks that include the delayed allocation
4251 * blocks for this file.
4252 */
4253 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4254 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4255 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4256
4257 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4258 return 0;
4259}
3196 4260
3197/* 4261/*
3198 * How many blocks doth make a writepage()? 4262 * How many blocks doth make a writepage()?
@@ -3506,3 +4570,64 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
3506 4570
3507 return err; 4571 return err;
3508} 4572}
4573
4574static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4575{
4576 return !buffer_mapped(bh);
4577}
4578
4579int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4580{
4581 loff_t size;
4582 unsigned long len;
4583 int ret = -EINVAL;
4584 struct file *file = vma->vm_file;
4585 struct inode *inode = file->f_path.dentry->d_inode;
4586 struct address_space *mapping = inode->i_mapping;
4587
4588 /*
4589 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4590 * get i_mutex because we are already holding mmap_sem.
4591 */
4592 down_read(&inode->i_alloc_sem);
4593 size = i_size_read(inode);
4594 if (page->mapping != mapping || size <= page_offset(page)
4595 || !PageUptodate(page)) {
4596 /* page got truncated from under us? */
4597 goto out_unlock;
4598 }
4599 ret = 0;
4600 if (PageMappedToDisk(page))
4601 goto out_unlock;
4602
4603 if (page->index == size >> PAGE_CACHE_SHIFT)
4604 len = size & ~PAGE_CACHE_MASK;
4605 else
4606 len = PAGE_CACHE_SIZE;
4607
4608 if (page_has_buffers(page)) {
4609 /* return if we have all the buffers mapped */
4610 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4611 ext4_bh_unmapped))
4612 goto out_unlock;
4613 }
4614 /*
4615 * OK, we need to fill the hole... Do write_begin write_end
4616 * to do block allocation/reservation.We are not holding
4617 * inode.i__mutex here. That allow * parallel write_begin,
4618 * write_end call. lock_page prevent this from happening
4619 * on the same page though
4620 */
4621 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4622 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4623 if (ret < 0)
4624 goto out_unlock;
4625 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4626 len, len, page, NULL);
4627 if (ret < 0)
4628 goto out_unlock;
4629 ret = 0;
4630out_unlock:
4631 up_read(&inode->i_alloc_sem);
4632 return ret;
4633}
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index c9900aade150..8d141a25bbee 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -381,22 +381,28 @@ static inline void mb_clear_bit_atomic(spinlock_t *lock, int bit, void *addr)
381 381
382static inline int mb_find_next_zero_bit(void *addr, int max, int start) 382static inline int mb_find_next_zero_bit(void *addr, int max, int start)
383{ 383{
384 int fix = 0; 384 int fix = 0, ret, tmpmax;
385 addr = mb_correct_addr_and_bit(&fix, addr); 385 addr = mb_correct_addr_and_bit(&fix, addr);
386 max += fix; 386 tmpmax = max + fix;
387 start += fix; 387 start += fix;
388 388
389 return ext4_find_next_zero_bit(addr, max, start) - fix; 389 ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
390 if (ret > max)
391 return max;
392 return ret;
390} 393}
391 394
392static inline int mb_find_next_bit(void *addr, int max, int start) 395static inline int mb_find_next_bit(void *addr, int max, int start)
393{ 396{
394 int fix = 0; 397 int fix = 0, ret, tmpmax;
395 addr = mb_correct_addr_and_bit(&fix, addr); 398 addr = mb_correct_addr_and_bit(&fix, addr);
396 max += fix; 399 tmpmax = max + fix;
397 start += fix; 400 start += fix;
398 401
399 return ext4_find_next_bit(addr, max, start) - fix; 402 ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
403 if (ret > max)
404 return max;
405 return ret;
400} 406}
401 407
402static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max) 408static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
@@ -803,6 +809,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore)
803 if (!buffer_uptodate(bh[i])) 809 if (!buffer_uptodate(bh[i]))
804 goto out; 810 goto out;
805 811
812 err = 0;
806 first_block = page->index * blocks_per_page; 813 first_block = page->index * blocks_per_page;
807 for (i = 0; i < blocks_per_page; i++) { 814 for (i = 0; i < blocks_per_page; i++) {
808 int group; 815 int group;
@@ -883,6 +890,7 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
883 int pnum; 890 int pnum;
884 int poff; 891 int poff;
885 struct page *page; 892 struct page *page;
893 int ret;
886 894
887 mb_debug("load group %lu\n", group); 895 mb_debug("load group %lu\n", group);
888 896
@@ -914,15 +922,21 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
914 if (page) { 922 if (page) {
915 BUG_ON(page->mapping != inode->i_mapping); 923 BUG_ON(page->mapping != inode->i_mapping);
916 if (!PageUptodate(page)) { 924 if (!PageUptodate(page)) {
917 ext4_mb_init_cache(page, NULL); 925 ret = ext4_mb_init_cache(page, NULL);
926 if (ret) {
927 unlock_page(page);
928 goto err;
929 }
918 mb_cmp_bitmaps(e4b, page_address(page) + 930 mb_cmp_bitmaps(e4b, page_address(page) +
919 (poff * sb->s_blocksize)); 931 (poff * sb->s_blocksize));
920 } 932 }
921 unlock_page(page); 933 unlock_page(page);
922 } 934 }
923 } 935 }
924 if (page == NULL || !PageUptodate(page)) 936 if (page == NULL || !PageUptodate(page)) {
937 ret = -EIO;
925 goto err; 938 goto err;
939 }
926 e4b->bd_bitmap_page = page; 940 e4b->bd_bitmap_page = page;
927 e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize); 941 e4b->bd_bitmap = page_address(page) + (poff * sb->s_blocksize);
928 mark_page_accessed(page); 942 mark_page_accessed(page);
@@ -938,14 +952,20 @@ ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
938 page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS); 952 page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
939 if (page) { 953 if (page) {
940 BUG_ON(page->mapping != inode->i_mapping); 954 BUG_ON(page->mapping != inode->i_mapping);
941 if (!PageUptodate(page)) 955 if (!PageUptodate(page)) {
942 ext4_mb_init_cache(page, e4b->bd_bitmap); 956 ret = ext4_mb_init_cache(page, e4b->bd_bitmap);
943 957 if (ret) {
958 unlock_page(page);
959 goto err;
960 }
961 }
944 unlock_page(page); 962 unlock_page(page);
945 } 963 }
946 } 964 }
947 if (page == NULL || !PageUptodate(page)) 965 if (page == NULL || !PageUptodate(page)) {
966 ret = -EIO;
948 goto err; 967 goto err;
968 }
949 e4b->bd_buddy_page = page; 969 e4b->bd_buddy_page = page;
950 e4b->bd_buddy = page_address(page) + (poff * sb->s_blocksize); 970 e4b->bd_buddy = page_address(page) + (poff * sb->s_blocksize);
951 mark_page_accessed(page); 971 mark_page_accessed(page);
@@ -962,7 +982,7 @@ err:
962 page_cache_release(e4b->bd_buddy_page); 982 page_cache_release(e4b->bd_buddy_page);
963 e4b->bd_buddy = NULL; 983 e4b->bd_buddy = NULL;
964 e4b->bd_bitmap = NULL; 984 e4b->bd_bitmap = NULL;
965 return -EIO; 985 return ret;
966} 986}
967 987
968static void ext4_mb_release_desc(struct ext4_buddy *e4b) 988static void ext4_mb_release_desc(struct ext4_buddy *e4b)
@@ -1031,7 +1051,7 @@ static void mb_set_bits(spinlock_t *lock, void *bm, int cur, int len)
1031 } 1051 }
1032} 1052}
1033 1053
1034static int mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b, 1054static void mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1035 int first, int count) 1055 int first, int count)
1036{ 1056{
1037 int block = 0; 1057 int block = 0;
@@ -1071,11 +1091,12 @@ static int mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1071 blocknr += block; 1091 blocknr += block;
1072 blocknr += 1092 blocknr +=
1073 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block); 1093 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block);
1074 1094 ext4_unlock_group(sb, e4b->bd_group);
1075 ext4_error(sb, __func__, "double-free of inode" 1095 ext4_error(sb, __func__, "double-free of inode"
1076 " %lu's block %llu(bit %u in group %lu)\n", 1096 " %lu's block %llu(bit %u in group %lu)\n",
1077 inode ? inode->i_ino : 0, blocknr, block, 1097 inode ? inode->i_ino : 0, blocknr, block,
1078 e4b->bd_group); 1098 e4b->bd_group);
1099 ext4_lock_group(sb, e4b->bd_group);
1079 } 1100 }
1080 mb_clear_bit(block, EXT4_MB_BITMAP(e4b)); 1101 mb_clear_bit(block, EXT4_MB_BITMAP(e4b));
1081 e4b->bd_info->bb_counters[order]++; 1102 e4b->bd_info->bb_counters[order]++;
@@ -1113,8 +1134,6 @@ static int mb_free_blocks(struct inode *inode, struct ext4_buddy *e4b,
1113 } while (1); 1134 } while (1);
1114 } 1135 }
1115 mb_check_buddy(e4b); 1136 mb_check_buddy(e4b);
1116
1117 return 0;
1118} 1137}
1119 1138
1120static int mb_find_extent(struct ext4_buddy *e4b, int order, int block, 1139static int mb_find_extent(struct ext4_buddy *e4b, int order, int block,
@@ -1730,10 +1749,6 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1730 ac->ac_g_ex.fe_start = sbi->s_mb_last_start; 1749 ac->ac_g_ex.fe_start = sbi->s_mb_last_start;
1731 spin_unlock(&sbi->s_md_lock); 1750 spin_unlock(&sbi->s_md_lock);
1732 } 1751 }
1733
1734 /* searching for the right group start from the goal value specified */
1735 group = ac->ac_g_ex.fe_group;
1736
1737 /* Let's just scan groups to find more-less suitable blocks */ 1752 /* Let's just scan groups to find more-less suitable blocks */
1738 cr = ac->ac_2order ? 0 : 1; 1753 cr = ac->ac_2order ? 0 : 1;
1739 /* 1754 /*
@@ -1743,6 +1758,12 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
1743repeat: 1758repeat:
1744 for (; cr < 4 && ac->ac_status == AC_STATUS_CONTINUE; cr++) { 1759 for (; cr < 4 && ac->ac_status == AC_STATUS_CONTINUE; cr++) {
1745 ac->ac_criteria = cr; 1760 ac->ac_criteria = cr;
1761 /*
1762 * searching for the right group start
1763 * from the goal value specified
1764 */
1765 group = ac->ac_g_ex.fe_group;
1766
1746 for (i = 0; i < EXT4_SB(sb)->s_groups_count; group++, i++) { 1767 for (i = 0; i < EXT4_SB(sb)->s_groups_count; group++, i++) {
1747 struct ext4_group_info *grp; 1768 struct ext4_group_info *grp;
1748 struct ext4_group_desc *desc; 1769 struct ext4_group_desc *desc;
@@ -1963,6 +1984,8 @@ static int ext4_mb_seq_history_open(struct inode *inode, struct file *file)
1963 int rc; 1984 int rc;
1964 int size; 1985 int size;
1965 1986
1987 if (unlikely(sbi->s_mb_history == NULL))
1988 return -ENOMEM;
1966 s = kmalloc(sizeof(*s), GFP_KERNEL); 1989 s = kmalloc(sizeof(*s), GFP_KERNEL);
1967 if (s == NULL) 1990 if (s == NULL)
1968 return -ENOMEM; 1991 return -ENOMEM;
@@ -2165,9 +2188,7 @@ static void ext4_mb_history_init(struct super_block *sb)
2165 sbi->s_mb_history_cur = 0; 2188 sbi->s_mb_history_cur = 0;
2166 spin_lock_init(&sbi->s_mb_history_lock); 2189 spin_lock_init(&sbi->s_mb_history_lock);
2167 i = sbi->s_mb_history_max * sizeof(struct ext4_mb_history); 2190 i = sbi->s_mb_history_max * sizeof(struct ext4_mb_history);
2168 sbi->s_mb_history = kmalloc(i, GFP_KERNEL); 2191 sbi->s_mb_history = kzalloc(i, GFP_KERNEL);
2169 if (likely(sbi->s_mb_history != NULL))
2170 memset(sbi->s_mb_history, 0, i);
2171 /* if we can't allocate history, then we simple won't use it */ 2192 /* if we can't allocate history, then we simple won't use it */
2172} 2193}
2173 2194
@@ -2215,21 +2236,192 @@ ext4_mb_store_history(struct ext4_allocation_context *ac)
2215#define ext4_mb_history_init(sb) 2236#define ext4_mb_history_init(sb)
2216#endif 2237#endif
2217 2238
2239
2240/* Create and initialize ext4_group_info data for the given group. */
2241int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
2242 struct ext4_group_desc *desc)
2243{
2244 int i, len;
2245 int metalen = 0;
2246 struct ext4_sb_info *sbi = EXT4_SB(sb);
2247 struct ext4_group_info **meta_group_info;
2248
2249 /*
2250 * First check if this group is the first of a reserved block.
2251 * If it's true, we have to allocate a new table of pointers
2252 * to ext4_group_info structures
2253 */
2254 if (group % EXT4_DESC_PER_BLOCK(sb) == 0) {
2255 metalen = sizeof(*meta_group_info) <<
2256 EXT4_DESC_PER_BLOCK_BITS(sb);
2257 meta_group_info = kmalloc(metalen, GFP_KERNEL);
2258 if (meta_group_info == NULL) {
2259 printk(KERN_ERR "EXT4-fs: can't allocate mem for a "
2260 "buddy group\n");
2261 goto exit_meta_group_info;
2262 }
2263 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)] =
2264 meta_group_info;
2265 }
2266
2267 /*
2268 * calculate needed size. if change bb_counters size,
2269 * don't forget about ext4_mb_generate_buddy()
2270 */
2271 len = offsetof(typeof(**meta_group_info),
2272 bb_counters[sb->s_blocksize_bits + 2]);
2273
2274 meta_group_info =
2275 sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)];
2276 i = group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2277
2278 meta_group_info[i] = kzalloc(len, GFP_KERNEL);
2279 if (meta_group_info[i] == NULL) {
2280 printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
2281 goto exit_group_info;
2282 }
2283 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT,
2284 &(meta_group_info[i]->bb_state));
2285
2286 /*
2287 * initialize bb_free to be able to skip
2288 * empty groups without initialization
2289 */
2290 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
2291 meta_group_info[i]->bb_free =
2292 ext4_free_blocks_after_init(sb, group, desc);
2293 } else {
2294 meta_group_info[i]->bb_free =
2295 le16_to_cpu(desc->bg_free_blocks_count);
2296 }
2297
2298 INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
2299
2300#ifdef DOUBLE_CHECK
2301 {
2302 struct buffer_head *bh;
2303 meta_group_info[i]->bb_bitmap =
2304 kmalloc(sb->s_blocksize, GFP_KERNEL);
2305 BUG_ON(meta_group_info[i]->bb_bitmap == NULL);
2306 bh = ext4_read_block_bitmap(sb, group);
2307 BUG_ON(bh == NULL);
2308 memcpy(meta_group_info[i]->bb_bitmap, bh->b_data,
2309 sb->s_blocksize);
2310 put_bh(bh);
2311 }
2312#endif
2313
2314 return 0;
2315
2316exit_group_info:
2317 /* If a meta_group_info table has been allocated, release it now */
2318 if (group % EXT4_DESC_PER_BLOCK(sb) == 0)
2319 kfree(sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)]);
2320exit_meta_group_info:
2321 return -ENOMEM;
2322} /* ext4_mb_add_groupinfo */
2323
2324/*
2325 * Add a group to the existing groups.
2326 * This function is used for online resize
2327 */
2328int ext4_mb_add_more_groupinfo(struct super_block *sb, ext4_group_t group,
2329 struct ext4_group_desc *desc)
2330{
2331 struct ext4_sb_info *sbi = EXT4_SB(sb);
2332 struct inode *inode = sbi->s_buddy_cache;
2333 int blocks_per_page;
2334 int block;
2335 int pnum;
2336 struct page *page;
2337 int err;
2338
2339 /* Add group based on group descriptor*/
2340 err = ext4_mb_add_groupinfo(sb, group, desc);
2341 if (err)
2342 return err;
2343
2344 /*
2345 * Cache pages containing dynamic mb_alloc datas (buddy and bitmap
2346 * datas) are set not up to date so that they will be re-initilaized
2347 * during the next call to ext4_mb_load_buddy
2348 */
2349
2350 /* Set buddy page as not up to date */
2351 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
2352 block = group * 2;
2353 pnum = block / blocks_per_page;
2354 page = find_get_page(inode->i_mapping, pnum);
2355 if (page != NULL) {
2356 ClearPageUptodate(page);
2357 page_cache_release(page);
2358 }
2359
2360 /* Set bitmap page as not up to date */
2361 block++;
2362 pnum = block / blocks_per_page;
2363 page = find_get_page(inode->i_mapping, pnum);
2364 if (page != NULL) {
2365 ClearPageUptodate(page);
2366 page_cache_release(page);
2367 }
2368
2369 return 0;
2370}
2371
2372/*
2373 * Update an existing group.
2374 * This function is used for online resize
2375 */
2376void ext4_mb_update_group_info(struct ext4_group_info *grp, ext4_grpblk_t add)
2377{
2378 grp->bb_free += add;
2379}
2380
2218static int ext4_mb_init_backend(struct super_block *sb) 2381static int ext4_mb_init_backend(struct super_block *sb)
2219{ 2382{
2220 ext4_group_t i; 2383 ext4_group_t i;
2221 int j, len, metalen; 2384 int metalen;
2222 struct ext4_sb_info *sbi = EXT4_SB(sb); 2385 struct ext4_sb_info *sbi = EXT4_SB(sb);
2223 int num_meta_group_infos = 2386 struct ext4_super_block *es = sbi->s_es;
2224 (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) >> 2387 int num_meta_group_infos;
2225 EXT4_DESC_PER_BLOCK_BITS(sb); 2388 int num_meta_group_infos_max;
2389 int array_size;
2226 struct ext4_group_info **meta_group_info; 2390 struct ext4_group_info **meta_group_info;
2391 struct ext4_group_desc *desc;
2392
2393 /* This is the number of blocks used by GDT */
2394 num_meta_group_infos = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) -
2395 1) >> EXT4_DESC_PER_BLOCK_BITS(sb);
2396
2397 /*
2398 * This is the total number of blocks used by GDT including
2399 * the number of reserved blocks for GDT.
2400 * The s_group_info array is allocated with this value
2401 * to allow a clean online resize without a complex
2402 * manipulation of pointer.
2403 * The drawback is the unused memory when no resize
2404 * occurs but it's very low in terms of pages
2405 * (see comments below)
2406 * Need to handle this properly when META_BG resizing is allowed
2407 */
2408 num_meta_group_infos_max = num_meta_group_infos +
2409 le16_to_cpu(es->s_reserved_gdt_blocks);
2227 2410
2411 /*
2412 * array_size is the size of s_group_info array. We round it
2413 * to the next power of two because this approximation is done
2414 * internally by kmalloc so we can have some more memory
2415 * for free here (e.g. may be used for META_BG resize).
2416 */
2417 array_size = 1;
2418 while (array_size < sizeof(*sbi->s_group_info) *
2419 num_meta_group_infos_max)
2420 array_size = array_size << 1;
2228 /* An 8TB filesystem with 64-bit pointers requires a 4096 byte 2421 /* An 8TB filesystem with 64-bit pointers requires a 4096 byte
2229 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem. 2422 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
2230 * So a two level scheme suffices for now. */ 2423 * So a two level scheme suffices for now. */
2231 sbi->s_group_info = kmalloc(sizeof(*sbi->s_group_info) * 2424 sbi->s_group_info = kmalloc(array_size, GFP_KERNEL);
2232 num_meta_group_infos, GFP_KERNEL);
2233 if (sbi->s_group_info == NULL) { 2425 if (sbi->s_group_info == NULL) {
2234 printk(KERN_ERR "EXT4-fs: can't allocate buddy meta group\n"); 2426 printk(KERN_ERR "EXT4-fs: can't allocate buddy meta group\n");
2235 return -ENOMEM; 2427 return -ENOMEM;
@@ -2256,63 +2448,15 @@ static int ext4_mb_init_backend(struct super_block *sb)
2256 sbi->s_group_info[i] = meta_group_info; 2448 sbi->s_group_info[i] = meta_group_info;
2257 } 2449 }
2258 2450
2259 /*
2260 * calculate needed size. if change bb_counters size,
2261 * don't forget about ext4_mb_generate_buddy()
2262 */
2263 len = sizeof(struct ext4_group_info);
2264 len += sizeof(unsigned short) * (sb->s_blocksize_bits + 2);
2265 for (i = 0; i < sbi->s_groups_count; i++) { 2451 for (i = 0; i < sbi->s_groups_count; i++) {
2266 struct ext4_group_desc *desc;
2267
2268 meta_group_info =
2269 sbi->s_group_info[i >> EXT4_DESC_PER_BLOCK_BITS(sb)];
2270 j = i & (EXT4_DESC_PER_BLOCK(sb) - 1);
2271
2272 meta_group_info[j] = kzalloc(len, GFP_KERNEL);
2273 if (meta_group_info[j] == NULL) {
2274 printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
2275 goto err_freebuddy;
2276 }
2277 desc = ext4_get_group_desc(sb, i, NULL); 2452 desc = ext4_get_group_desc(sb, i, NULL);
2278 if (desc == NULL) { 2453 if (desc == NULL) {
2279 printk(KERN_ERR 2454 printk(KERN_ERR
2280 "EXT4-fs: can't read descriptor %lu\n", i); 2455 "EXT4-fs: can't read descriptor %lu\n", i);
2281 i++;
2282 goto err_freebuddy; 2456 goto err_freebuddy;
2283 } 2457 }
2284 memset(meta_group_info[j], 0, len); 2458 if (ext4_mb_add_groupinfo(sb, i, desc) != 0)
2285 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, 2459 goto err_freebuddy;
2286 &(meta_group_info[j]->bb_state));
2287
2288 /*
2289 * initialize bb_free to be able to skip
2290 * empty groups without initialization
2291 */
2292 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
2293 meta_group_info[j]->bb_free =
2294 ext4_free_blocks_after_init(sb, i, desc);
2295 } else {
2296 meta_group_info[j]->bb_free =
2297 le16_to_cpu(desc->bg_free_blocks_count);
2298 }
2299
2300 INIT_LIST_HEAD(&meta_group_info[j]->bb_prealloc_list);
2301
2302#ifdef DOUBLE_CHECK
2303 {
2304 struct buffer_head *bh;
2305 meta_group_info[j]->bb_bitmap =
2306 kmalloc(sb->s_blocksize, GFP_KERNEL);
2307 BUG_ON(meta_group_info[j]->bb_bitmap == NULL);
2308 bh = read_block_bitmap(sb, i);
2309 BUG_ON(bh == NULL);
2310 memcpy(meta_group_info[j]->bb_bitmap, bh->b_data,
2311 sb->s_blocksize);
2312 put_bh(bh);
2313 }
2314#endif
2315
2316 } 2460 }
2317 2461
2318 return 0; 2462 return 0;
@@ -2336,6 +2480,7 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2336 unsigned i; 2480 unsigned i;
2337 unsigned offset; 2481 unsigned offset;
2338 unsigned max; 2482 unsigned max;
2483 int ret;
2339 2484
2340 if (!test_opt(sb, MBALLOC)) 2485 if (!test_opt(sb, MBALLOC))
2341 return 0; 2486 return 0;
@@ -2370,12 +2515,12 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
2370 } while (i <= sb->s_blocksize_bits + 1); 2515 } while (i <= sb->s_blocksize_bits + 1);
2371 2516
2372 /* init file for buddy data */ 2517 /* init file for buddy data */
2373 i = ext4_mb_init_backend(sb); 2518 ret = ext4_mb_init_backend(sb);
2374 if (i) { 2519 if (ret != 0) {
2375 clear_opt(sbi->s_mount_opt, MBALLOC); 2520 clear_opt(sbi->s_mount_opt, MBALLOC);
2376 kfree(sbi->s_mb_offsets); 2521 kfree(sbi->s_mb_offsets);
2377 kfree(sbi->s_mb_maxs); 2522 kfree(sbi->s_mb_maxs);
2378 return i; 2523 return ret;
2379 } 2524 }
2380 2525
2381 spin_lock_init(&sbi->s_md_lock); 2526 spin_lock_init(&sbi->s_md_lock);
@@ -2548,8 +2693,7 @@ ext4_mb_free_committed_blocks(struct super_block *sb)
2548 ext4_lock_group(sb, md->group); 2693 ext4_lock_group(sb, md->group);
2549 for (i = 0; i < md->num; i++) { 2694 for (i = 0; i < md->num; i++) {
2550 mb_debug(" %u", md->blocks[i]); 2695 mb_debug(" %u", md->blocks[i]);
2551 err = mb_free_blocks(NULL, &e4b, md->blocks[i], 1); 2696 mb_free_blocks(NULL, &e4b, md->blocks[i], 1);
2552 BUG_ON(err != 0);
2553 } 2697 }
2554 mb_debug("\n"); 2698 mb_debug("\n");
2555 ext4_unlock_group(sb, md->group); 2699 ext4_unlock_group(sb, md->group);
@@ -2575,25 +2719,24 @@ ext4_mb_free_committed_blocks(struct super_block *sb)
2575 2719
2576 2720
2577 2721
2578#define MB_PROC_VALUE_READ(name) \ 2722#define MB_PROC_FOPS(name) \
2579static int ext4_mb_read_##name(char *page, char **start, \ 2723static int ext4_mb_##name##_proc_show(struct seq_file *m, void *v) \
2580 off_t off, int count, int *eof, void *data) \
2581{ \ 2724{ \
2582 struct ext4_sb_info *sbi = data; \ 2725 struct ext4_sb_info *sbi = m->private; \
2583 int len; \ 2726 \
2584 *eof = 1; \ 2727 seq_printf(m, "%ld\n", sbi->s_mb_##name); \
2585 if (off != 0) \ 2728 return 0; \
2586 return 0; \ 2729} \
2587 len = sprintf(page, "%ld\n", sbi->s_mb_##name); \ 2730 \
2588 *start = page; \ 2731static int ext4_mb_##name##_proc_open(struct inode *inode, struct file *file)\
2589 return len; \ 2732{ \
2590} 2733 return single_open(file, ext4_mb_##name##_proc_show, PDE(inode)->data);\
2591 2734} \
2592#define MB_PROC_VALUE_WRITE(name) \ 2735 \
2593static int ext4_mb_write_##name(struct file *file, \ 2736static ssize_t ext4_mb_##name##_proc_write(struct file *file, \
2594 const char __user *buf, unsigned long cnt, void *data) \ 2737 const char __user *buf, size_t cnt, loff_t *ppos) \
2595{ \ 2738{ \
2596 struct ext4_sb_info *sbi = data; \ 2739 struct ext4_sb_info *sbi = PDE(file->f_path.dentry->d_inode)->data;\
2597 char str[32]; \ 2740 char str[32]; \
2598 long value; \ 2741 long value; \
2599 if (cnt >= sizeof(str)) \ 2742 if (cnt >= sizeof(str)) \
@@ -2605,31 +2748,32 @@ static int ext4_mb_write_##name(struct file *file, \
2605 return -ERANGE; \ 2748 return -ERANGE; \
2606 sbi->s_mb_##name = value; \ 2749 sbi->s_mb_##name = value; \
2607 return cnt; \ 2750 return cnt; \
2608} 2751} \
2752 \
2753static const struct file_operations ext4_mb_##name##_proc_fops = { \
2754 .owner = THIS_MODULE, \
2755 .open = ext4_mb_##name##_proc_open, \
2756 .read = seq_read, \
2757 .llseek = seq_lseek, \
2758 .release = single_release, \
2759 .write = ext4_mb_##name##_proc_write, \
2760};
2609 2761
2610MB_PROC_VALUE_READ(stats); 2762MB_PROC_FOPS(stats);
2611MB_PROC_VALUE_WRITE(stats); 2763MB_PROC_FOPS(max_to_scan);
2612MB_PROC_VALUE_READ(max_to_scan); 2764MB_PROC_FOPS(min_to_scan);
2613MB_PROC_VALUE_WRITE(max_to_scan); 2765MB_PROC_FOPS(order2_reqs);
2614MB_PROC_VALUE_READ(min_to_scan); 2766MB_PROC_FOPS(stream_request);
2615MB_PROC_VALUE_WRITE(min_to_scan); 2767MB_PROC_FOPS(group_prealloc);
2616MB_PROC_VALUE_READ(order2_reqs);
2617MB_PROC_VALUE_WRITE(order2_reqs);
2618MB_PROC_VALUE_READ(stream_request);
2619MB_PROC_VALUE_WRITE(stream_request);
2620MB_PROC_VALUE_READ(group_prealloc);
2621MB_PROC_VALUE_WRITE(group_prealloc);
2622 2768
2623#define MB_PROC_HANDLER(name, var) \ 2769#define MB_PROC_HANDLER(name, var) \
2624do { \ 2770do { \
2625 proc = create_proc_entry(name, mode, sbi->s_mb_proc); \ 2771 proc = proc_create_data(name, mode, sbi->s_mb_proc, \
2772 &ext4_mb_##var##_proc_fops, sbi); \
2626 if (proc == NULL) { \ 2773 if (proc == NULL) { \
2627 printk(KERN_ERR "EXT4-fs: can't to create %s\n", name); \ 2774 printk(KERN_ERR "EXT4-fs: can't to create %s\n", name); \
2628 goto err_out; \ 2775 goto err_out; \
2629 } \ 2776 } \
2630 proc->data = sbi; \
2631 proc->read_proc = ext4_mb_read_##var ; \
2632 proc->write_proc = ext4_mb_write_##var; \
2633} while (0) 2777} while (0)
2634 2778
2635static int ext4_mb_init_per_dev_proc(struct super_block *sb) 2779static int ext4_mb_init_per_dev_proc(struct super_block *sb)
@@ -2639,6 +2783,10 @@ static int ext4_mb_init_per_dev_proc(struct super_block *sb)
2639 struct proc_dir_entry *proc; 2783 struct proc_dir_entry *proc;
2640 char devname[64]; 2784 char devname[64];
2641 2785
2786 if (proc_root_ext4 == NULL) {
2787 sbi->s_mb_proc = NULL;
2788 return -EINVAL;
2789 }
2642 bdevname(sb->s_bdev, devname); 2790 bdevname(sb->s_bdev, devname);
2643 sbi->s_mb_proc = proc_mkdir(devname, proc_root_ext4); 2791 sbi->s_mb_proc = proc_mkdir(devname, proc_root_ext4);
2644 2792
@@ -2747,7 +2895,7 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2747 2895
2748 2896
2749 err = -EIO; 2897 err = -EIO;
2750 bitmap_bh = read_block_bitmap(sb, ac->ac_b_ex.fe_group); 2898 bitmap_bh = ext4_read_block_bitmap(sb, ac->ac_b_ex.fe_group);
2751 if (!bitmap_bh) 2899 if (!bitmap_bh)
2752 goto out_err; 2900 goto out_err;
2753 2901
@@ -2816,7 +2964,23 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
2816 le16_add_cpu(&gdp->bg_free_blocks_count, -ac->ac_b_ex.fe_len); 2964 le16_add_cpu(&gdp->bg_free_blocks_count, -ac->ac_b_ex.fe_len);
2817 gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp); 2965 gdp->bg_checksum = ext4_group_desc_csum(sbi, ac->ac_b_ex.fe_group, gdp);
2818 spin_unlock(sb_bgl_lock(sbi, ac->ac_b_ex.fe_group)); 2966 spin_unlock(sb_bgl_lock(sbi, ac->ac_b_ex.fe_group));
2819 percpu_counter_sub(&sbi->s_freeblocks_counter, ac->ac_b_ex.fe_len); 2967
2968 /*
2969 * free blocks account has already be reduced/reserved
2970 * at write_begin() time for delayed allocation
2971 * do not double accounting
2972 */
2973 if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
2974 percpu_counter_sub(&sbi->s_freeblocks_counter,
2975 ac->ac_b_ex.fe_len);
2976
2977 if (sbi->s_log_groups_per_flex) {
2978 ext4_group_t flex_group = ext4_flex_group(sbi,
2979 ac->ac_b_ex.fe_group);
2980 spin_lock(sb_bgl_lock(sbi, flex_group));
2981 sbi->s_flex_groups[flex_group].free_blocks -= ac->ac_b_ex.fe_len;
2982 spin_unlock(sb_bgl_lock(sbi, flex_group));
2983 }
2820 2984
2821 err = ext4_journal_dirty_metadata(handle, bitmap_bh); 2985 err = ext4_journal_dirty_metadata(handle, bitmap_bh);
2822 if (err) 2986 if (err)
@@ -3473,8 +3637,6 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
3473 if (bit >= end) 3637 if (bit >= end)
3474 break; 3638 break;
3475 next = mb_find_next_bit(bitmap_bh->b_data, end, bit); 3639 next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
3476 if (next > end)
3477 next = end;
3478 start = group * EXT4_BLOCKS_PER_GROUP(sb) + bit + 3640 start = group * EXT4_BLOCKS_PER_GROUP(sb) + bit +
3479 le32_to_cpu(sbi->s_es->s_first_data_block); 3641 le32_to_cpu(sbi->s_es->s_first_data_block);
3480 mb_debug(" free preallocated %u/%u in group %u\n", 3642 mb_debug(" free preallocated %u/%u in group %u\n",
@@ -3569,7 +3731,7 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
3569 if (list_empty(&grp->bb_prealloc_list)) 3731 if (list_empty(&grp->bb_prealloc_list))
3570 return 0; 3732 return 0;
3571 3733
3572 bitmap_bh = read_block_bitmap(sb, group); 3734 bitmap_bh = ext4_read_block_bitmap(sb, group);
3573 if (bitmap_bh == NULL) { 3735 if (bitmap_bh == NULL) {
3574 /* error handling here */ 3736 /* error handling here */
3575 ext4_mb_release_desc(&e4b); 3737 ext4_mb_release_desc(&e4b);
@@ -3743,7 +3905,7 @@ repeat:
3743 err = ext4_mb_load_buddy(sb, group, &e4b); 3905 err = ext4_mb_load_buddy(sb, group, &e4b);
3744 BUG_ON(err != 0); /* error handling here */ 3906 BUG_ON(err != 0); /* error handling here */
3745 3907
3746 bitmap_bh = read_block_bitmap(sb, group); 3908 bitmap_bh = ext4_read_block_bitmap(sb, group);
3747 if (bitmap_bh == NULL) { 3909 if (bitmap_bh == NULL) {
3748 /* error handling here */ 3910 /* error handling here */
3749 ext4_mb_release_desc(&e4b); 3911 ext4_mb_release_desc(&e4b);
@@ -4011,10 +4173,21 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
4011 sbi = EXT4_SB(sb); 4173 sbi = EXT4_SB(sb);
4012 4174
4013 if (!test_opt(sb, MBALLOC)) { 4175 if (!test_opt(sb, MBALLOC)) {
4014 block = ext4_new_blocks_old(handle, ar->inode, ar->goal, 4176 block = ext4_old_new_blocks(handle, ar->inode, ar->goal,
4015 &(ar->len), errp); 4177 &(ar->len), errp);
4016 return block; 4178 return block;
4017 } 4179 }
4180 if (!EXT4_I(ar->inode)->i_delalloc_reserved_flag) {
4181 /*
4182 * With delalloc we already reserved the blocks
4183 */
4184 ar->len = ext4_has_free_blocks(sbi, ar->len);
4185 }
4186
4187 if (ar->len == 0) {
4188 *errp = -ENOSPC;
4189 return 0;
4190 }
4018 4191
4019 while (ar->len && DQUOT_ALLOC_BLOCK(ar->inode, ar->len)) { 4192 while (ar->len && DQUOT_ALLOC_BLOCK(ar->inode, ar->len)) {
4020 ar->flags |= EXT4_MB_HINT_NOPREALLOC; 4193 ar->flags |= EXT4_MB_HINT_NOPREALLOC;
@@ -4026,10 +4199,14 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
4026 } 4199 }
4027 inquota = ar->len; 4200 inquota = ar->len;
4028 4201
4202 if (EXT4_I(ar->inode)->i_delalloc_reserved_flag)
4203 ar->flags |= EXT4_MB_DELALLOC_RESERVED;
4204
4029 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS); 4205 ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
4030 if (!ac) { 4206 if (!ac) {
4207 ar->len = 0;
4031 *errp = -ENOMEM; 4208 *errp = -ENOMEM;
4032 return 0; 4209 goto out1;
4033 } 4210 }
4034 4211
4035 ext4_mb_poll_new_transaction(sb, handle); 4212 ext4_mb_poll_new_transaction(sb, handle);
@@ -4037,12 +4214,11 @@ ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
4037 *errp = ext4_mb_initialize_context(ac, ar); 4214 *errp = ext4_mb_initialize_context(ac, ar);
4038 if (*errp) { 4215 if (*errp) {
4039 ar->len = 0; 4216 ar->len = 0;
4040 goto out; 4217 goto out2;
4041 } 4218 }
4042 4219
4043 ac->ac_op = EXT4_MB_HISTORY_PREALLOC; 4220 ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
4044 if (!ext4_mb_use_preallocated(ac)) { 4221 if (!ext4_mb_use_preallocated(ac)) {
4045
4046 ac->ac_op = EXT4_MB_HISTORY_ALLOC; 4222 ac->ac_op = EXT4_MB_HISTORY_ALLOC;
4047 ext4_mb_normalize_request(ac, ar); 4223 ext4_mb_normalize_request(ac, ar);
4048repeat: 4224repeat:
@@ -4085,11 +4261,12 @@ repeat:
4085 4261
4086 ext4_mb_release_context(ac); 4262 ext4_mb_release_context(ac);
4087 4263
4088out: 4264out2:
4265 kmem_cache_free(ext4_ac_cachep, ac);
4266out1:
4089 if (ar->len < inquota) 4267 if (ar->len < inquota)
4090 DQUOT_FREE_BLOCK(ar->inode, inquota - ar->len); 4268 DQUOT_FREE_BLOCK(ar->inode, inquota - ar->len);
4091 4269
4092 kmem_cache_free(ext4_ac_cachep, ac);
4093 return block; 4270 return block;
4094} 4271}
4095static void ext4_mb_poll_new_transaction(struct super_block *sb, 4272static void ext4_mb_poll_new_transaction(struct super_block *sb,
@@ -4242,7 +4419,7 @@ do_more:
4242 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb); 4419 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
4243 count -= overflow; 4420 count -= overflow;
4244 } 4421 }
4245 bitmap_bh = read_block_bitmap(sb, block_group); 4422 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
4246 if (!bitmap_bh) 4423 if (!bitmap_bh)
4247 goto error_return; 4424 goto error_return;
4248 gdp = ext4_get_group_desc(sb, block_group, &gd_bh); 4425 gdp = ext4_get_group_desc(sb, block_group, &gd_bh);
@@ -4309,10 +4486,9 @@ do_more:
4309 ext4_mb_free_metadata(handle, &e4b, block_group, bit, count); 4486 ext4_mb_free_metadata(handle, &e4b, block_group, bit, count);
4310 } else { 4487 } else {
4311 ext4_lock_group(sb, block_group); 4488 ext4_lock_group(sb, block_group);
4312 err = mb_free_blocks(inode, &e4b, bit, count); 4489 mb_free_blocks(inode, &e4b, bit, count);
4313 ext4_mb_return_to_preallocation(inode, &e4b, block, count); 4490 ext4_mb_return_to_preallocation(inode, &e4b, block, count);
4314 ext4_unlock_group(sb, block_group); 4491 ext4_unlock_group(sb, block_group);
4315 BUG_ON(err != 0);
4316 } 4492 }
4317 4493
4318 spin_lock(sb_bgl_lock(sbi, block_group)); 4494 spin_lock(sb_bgl_lock(sbi, block_group));
@@ -4321,6 +4497,13 @@ do_more:
4321 spin_unlock(sb_bgl_lock(sbi, block_group)); 4497 spin_unlock(sb_bgl_lock(sbi, block_group));
4322 percpu_counter_add(&sbi->s_freeblocks_counter, count); 4498 percpu_counter_add(&sbi->s_freeblocks_counter, count);
4323 4499
4500 if (sbi->s_log_groups_per_flex) {
4501 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
4502 spin_lock(sb_bgl_lock(sbi, flex_group));
4503 sbi->s_flex_groups[flex_group].free_blocks += count;
4504 spin_unlock(sb_bgl_lock(sbi, flex_group));
4505 }
4506
4324 ext4_mb_release_desc(&e4b); 4507 ext4_mb_release_desc(&e4b);
4325 4508
4326 *freed += count; 4509 *freed += count;
diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
index ab16beaa830d..387ad98350c3 100644
--- a/fs/ext4/namei.c
+++ b/fs/ext4/namei.c
@@ -183,6 +183,16 @@ static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
183 struct inode *inode); 183 struct inode *inode);
184 184
185/* 185/*
186 * p is at least 6 bytes before the end of page
187 */
188static inline struct ext4_dir_entry_2 *
189ext4_next_entry(struct ext4_dir_entry_2 *p)
190{
191 return (struct ext4_dir_entry_2 *)((char *)p +
192 ext4_rec_len_from_disk(p->rec_len));
193}
194
195/*
186 * Future: use high four bits of block for coalesce-on-delete flags 196 * Future: use high four bits of block for coalesce-on-delete flags
187 * Mask them off for now. 197 * Mask them off for now.
188 */ 198 */
@@ -231,13 +241,13 @@ static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
231{ 241{
232 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) - 242 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
233 EXT4_DIR_REC_LEN(2) - infosize; 243 EXT4_DIR_REC_LEN(2) - infosize;
234 return 0? 20: entry_space / sizeof(struct dx_entry); 244 return entry_space / sizeof(struct dx_entry);
235} 245}
236 246
237static inline unsigned dx_node_limit (struct inode *dir) 247static inline unsigned dx_node_limit (struct inode *dir)
238{ 248{
239 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0); 249 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
240 return 0? 22: entry_space / sizeof(struct dx_entry); 250 return entry_space / sizeof(struct dx_entry);
241} 251}
242 252
243/* 253/*
@@ -554,15 +564,6 @@ static int ext4_htree_next_block(struct inode *dir, __u32 hash,
554 564
555 565
556/* 566/*
557 * p is at least 6 bytes before the end of page
558 */
559static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
560{
561 return (struct ext4_dir_entry_2 *)((char *)p +
562 ext4_rec_len_from_disk(p->rec_len));
563}
564
565/*
566 * This function fills a red-black tree with information from a 567 * This function fills a red-black tree with information from a
567 * directory block. It returns the number directory entries loaded 568 * directory block. It returns the number directory entries loaded
568 * into the tree. If there is an error it is returned in err. 569 * into the tree. If there is an error it is returned in err.
@@ -993,19 +994,21 @@ static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
993 de = (struct ext4_dir_entry_2 *) bh->b_data; 994 de = (struct ext4_dir_entry_2 *) bh->b_data;
994 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize - 995 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
995 EXT4_DIR_REC_LEN(0)); 996 EXT4_DIR_REC_LEN(0));
996 for (; de < top; de = ext4_next_entry(de)) 997 for (; de < top; de = ext4_next_entry(de)) {
997 if (ext4_match (namelen, name, de)) { 998 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
998 if (!ext4_check_dir_entry("ext4_find_entry", 999 + ((char *) de - bh->b_data);
999 dir, de, bh, 1000
1000 (block<<EXT4_BLOCK_SIZE_BITS(sb)) 1001 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1001 +((char *)de - bh->b_data))) { 1002 brelse(bh);
1002 brelse (bh);
1003 *err = ERR_BAD_DX_DIR; 1003 *err = ERR_BAD_DX_DIR;
1004 goto errout; 1004 goto errout;
1005 } 1005 }
1006 *res_dir = de; 1006
1007 dx_release (frames); 1007 if (ext4_match(namelen, name, de)) {
1008 return bh; 1008 *res_dir = de;
1009 dx_release(frames);
1010 return bh;
1011 }
1009 } 1012 }
1010 brelse (bh); 1013 brelse (bh);
1011 /* Check to see if we should continue to search */ 1014 /* Check to see if we should continue to search */
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index 9ff7b1c04239..f000fbe2cd93 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -866,6 +866,15 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
866 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp); 866 gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
867 867
868 /* 868 /*
869 * We can allocate memory for mb_alloc based on the new group
870 * descriptor
871 */
872 if (test_opt(sb, MBALLOC)) {
873 err = ext4_mb_add_more_groupinfo(sb, input->group, gdp);
874 if (err)
875 goto exit_journal;
876 }
877 /*
869 * Make the new blocks and inodes valid next. We do this before 878 * Make the new blocks and inodes valid next. We do this before
870 * increasing the group count so that once the group is enabled, 879 * increasing the group count so that once the group is enabled,
871 * all of its blocks and inodes are already valid. 880 * all of its blocks and inodes are already valid.
@@ -957,6 +966,8 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
957 handle_t *handle; 966 handle_t *handle;
958 int err; 967 int err;
959 unsigned long freed_blocks; 968 unsigned long freed_blocks;
969 ext4_group_t group;
970 struct ext4_group_info *grp;
960 971
961 /* We don't need to worry about locking wrt other resizers just 972 /* We don't need to worry about locking wrt other resizers just
962 * yet: we're going to revalidate es->s_blocks_count after 973 * yet: we're going to revalidate es->s_blocks_count after
@@ -988,7 +999,7 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
988 } 999 }
989 1000
990 /* Handle the remaining blocks in the last group only. */ 1001 /* Handle the remaining blocks in the last group only. */
991 ext4_get_group_no_and_offset(sb, o_blocks_count, NULL, &last); 1002 ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
992 1003
993 if (last == 0) { 1004 if (last == 0) {
994 ext4_warning(sb, __func__, 1005 ext4_warning(sb, __func__,
@@ -1060,6 +1071,45 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1060 o_blocks_count + add); 1071 o_blocks_count + add);
1061 if ((err = ext4_journal_stop(handle))) 1072 if ((err = ext4_journal_stop(handle)))
1062 goto exit_put; 1073 goto exit_put;
1074
1075 /*
1076 * Mark mballoc pages as not up to date so that they will be updated
1077 * next time they are loaded by ext4_mb_load_buddy.
1078 */
1079 if (test_opt(sb, MBALLOC)) {
1080 struct ext4_sb_info *sbi = EXT4_SB(sb);
1081 struct inode *inode = sbi->s_buddy_cache;
1082 int blocks_per_page;
1083 int block;
1084 int pnum;
1085 struct page *page;
1086
1087 /* Set buddy page as not up to date */
1088 blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
1089 block = group * 2;
1090 pnum = block / blocks_per_page;
1091 page = find_get_page(inode->i_mapping, pnum);
1092 if (page != NULL) {
1093 ClearPageUptodate(page);
1094 page_cache_release(page);
1095 }
1096
1097 /* Set bitmap page as not up to date */
1098 block++;
1099 pnum = block / blocks_per_page;
1100 page = find_get_page(inode->i_mapping, pnum);
1101 if (page != NULL) {
1102 ClearPageUptodate(page);
1103 page_cache_release(page);
1104 }
1105
1106 /* Get the info on the last group */
1107 grp = ext4_get_group_info(sb, group);
1108
1109 /* Update free blocks in group info */
1110 ext4_mb_update_group_info(grp, add);
1111 }
1112
1063 if (test_opt(sb, DEBUG)) 1113 if (test_opt(sb, DEBUG))
1064 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n", 1114 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1065 ext4_blocks_count(es)); 1115 ext4_blocks_count(es));
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index 02bf24343979..1cb371dcd609 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -506,6 +506,7 @@ static void ext4_put_super (struct super_block * sb)
506 ext4_ext_release(sb); 506 ext4_ext_release(sb);
507 ext4_xattr_put_super(sb); 507 ext4_xattr_put_super(sb);
508 jbd2_journal_destroy(sbi->s_journal); 508 jbd2_journal_destroy(sbi->s_journal);
509 sbi->s_journal = NULL;
509 if (!(sb->s_flags & MS_RDONLY)) { 510 if (!(sb->s_flags & MS_RDONLY)) {
510 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER); 511 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
511 es->s_state = cpu_to_le16(sbi->s_mount_state); 512 es->s_state = cpu_to_le16(sbi->s_mount_state);
@@ -517,6 +518,7 @@ static void ext4_put_super (struct super_block * sb)
517 for (i = 0; i < sbi->s_gdb_count; i++) 518 for (i = 0; i < sbi->s_gdb_count; i++)
518 brelse(sbi->s_group_desc[i]); 519 brelse(sbi->s_group_desc[i]);
519 kfree(sbi->s_group_desc); 520 kfree(sbi->s_group_desc);
521 kfree(sbi->s_flex_groups);
520 percpu_counter_destroy(&sbi->s_freeblocks_counter); 522 percpu_counter_destroy(&sbi->s_freeblocks_counter);
521 percpu_counter_destroy(&sbi->s_freeinodes_counter); 523 percpu_counter_destroy(&sbi->s_freeinodes_counter);
522 percpu_counter_destroy(&sbi->s_dirs_counter); 524 percpu_counter_destroy(&sbi->s_dirs_counter);
@@ -571,6 +573,12 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
571 memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache)); 573 memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
572 INIT_LIST_HEAD(&ei->i_prealloc_list); 574 INIT_LIST_HEAD(&ei->i_prealloc_list);
573 spin_lock_init(&ei->i_prealloc_lock); 575 spin_lock_init(&ei->i_prealloc_lock);
576 jbd2_journal_init_jbd_inode(&ei->jinode, &ei->vfs_inode);
577 ei->i_reserved_data_blocks = 0;
578 ei->i_reserved_meta_blocks = 0;
579 ei->i_allocated_meta_blocks = 0;
580 ei->i_delalloc_reserved_flag = 0;
581 spin_lock_init(&(ei->i_block_reservation_lock));
574 return &ei->vfs_inode; 582 return &ei->vfs_inode;
575} 583}
576 584
@@ -635,6 +643,8 @@ static void ext4_clear_inode(struct inode *inode)
635 EXT4_I(inode)->i_block_alloc_info = NULL; 643 EXT4_I(inode)->i_block_alloc_info = NULL;
636 if (unlikely(rsv)) 644 if (unlikely(rsv))
637 kfree(rsv); 645 kfree(rsv);
646 jbd2_journal_release_jbd_inode(EXT4_SB(inode->i_sb)->s_journal,
647 &EXT4_I(inode)->jinode);
638} 648}
639 649
640static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb) 650static inline void ext4_show_quota_options(struct seq_file *seq, struct super_block *sb)
@@ -671,7 +681,6 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
671 unsigned long def_mount_opts; 681 unsigned long def_mount_opts;
672 struct super_block *sb = vfs->mnt_sb; 682 struct super_block *sb = vfs->mnt_sb;
673 struct ext4_sb_info *sbi = EXT4_SB(sb); 683 struct ext4_sb_info *sbi = EXT4_SB(sb);
674 journal_t *journal = sbi->s_journal;
675 struct ext4_super_block *es = sbi->s_es; 684 struct ext4_super_block *es = sbi->s_es;
676 685
677 def_mount_opts = le32_to_cpu(es->s_default_mount_opts); 686 def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
@@ -747,6 +756,9 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
747 seq_puts(seq, ",nomballoc"); 756 seq_puts(seq, ",nomballoc");
748 if (test_opt(sb, I_VERSION)) 757 if (test_opt(sb, I_VERSION))
749 seq_puts(seq, ",i_version"); 758 seq_puts(seq, ",i_version");
759 if (!test_opt(sb, DELALLOC))
760 seq_puts(seq, ",nodelalloc");
761
750 762
751 if (sbi->s_stripe) 763 if (sbi->s_stripe)
752 seq_printf(seq, ",stripe=%lu", sbi->s_stripe); 764 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
@@ -894,7 +906,7 @@ enum {
894 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota, 906 Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
895 Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota, 907 Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
896 Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version, 908 Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
897 Opt_mballoc, Opt_nomballoc, Opt_stripe, 909 Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc,
898}; 910};
899 911
900static match_table_t tokens = { 912static match_table_t tokens = {
@@ -953,6 +965,8 @@ static match_table_t tokens = {
953 {Opt_nomballoc, "nomballoc"}, 965 {Opt_nomballoc, "nomballoc"},
954 {Opt_stripe, "stripe=%u"}, 966 {Opt_stripe, "stripe=%u"},
955 {Opt_resize, "resize"}, 967 {Opt_resize, "resize"},
968 {Opt_delalloc, "delalloc"},
969 {Opt_nodelalloc, "nodelalloc"},
956 {Opt_err, NULL}, 970 {Opt_err, NULL},
957}; 971};
958 972
@@ -990,6 +1004,7 @@ static int parse_options (char *options, struct super_block *sb,
990 int qtype, qfmt; 1004 int qtype, qfmt;
991 char *qname; 1005 char *qname;
992#endif 1006#endif
1007 ext4_fsblk_t last_block;
993 1008
994 if (!options) 1009 if (!options)
995 return 1; 1010 return 1;
@@ -1309,15 +1324,39 @@ set_qf_format:
1309 clear_opt(sbi->s_mount_opt, NOBH); 1324 clear_opt(sbi->s_mount_opt, NOBH);
1310 break; 1325 break;
1311 case Opt_extents: 1326 case Opt_extents:
1327 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,
1328 EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1329 ext4_warning(sb, __func__,
1330 "extents feature not enabled "
1331 "on this filesystem, use tune2fs\n");
1332 return 0;
1333 }
1312 set_opt (sbi->s_mount_opt, EXTENTS); 1334 set_opt (sbi->s_mount_opt, EXTENTS);
1313 break; 1335 break;
1314 case Opt_noextents: 1336 case Opt_noextents:
1337 /*
1338 * When e2fsprogs support resizing an already existing
1339 * ext3 file system to greater than 2**32 we need to
1340 * add support to block allocator to handle growing
1341 * already existing block mapped inode so that blocks
1342 * allocated for them fall within 2**32
1343 */
1344 last_block = ext4_blocks_count(sbi->s_es) - 1;
1345 if (last_block > 0xffffffffULL) {
1346 printk(KERN_ERR "EXT4-fs: Filesystem too "
1347 "large to mount with "
1348 "-o noextents options\n");
1349 return 0;
1350 }
1315 clear_opt (sbi->s_mount_opt, EXTENTS); 1351 clear_opt (sbi->s_mount_opt, EXTENTS);
1316 break; 1352 break;
1317 case Opt_i_version: 1353 case Opt_i_version:
1318 set_opt(sbi->s_mount_opt, I_VERSION); 1354 set_opt(sbi->s_mount_opt, I_VERSION);
1319 sb->s_flags |= MS_I_VERSION; 1355 sb->s_flags |= MS_I_VERSION;
1320 break; 1356 break;
1357 case Opt_nodelalloc:
1358 clear_opt(sbi->s_mount_opt, DELALLOC);
1359 break;
1321 case Opt_mballoc: 1360 case Opt_mballoc:
1322 set_opt(sbi->s_mount_opt, MBALLOC); 1361 set_opt(sbi->s_mount_opt, MBALLOC);
1323 break; 1362 break;
@@ -1331,6 +1370,9 @@ set_qf_format:
1331 return 0; 1370 return 0;
1332 sbi->s_stripe = option; 1371 sbi->s_stripe = option;
1333 break; 1372 break;
1373 case Opt_delalloc:
1374 set_opt(sbi->s_mount_opt, DELALLOC);
1375 break;
1334 default: 1376 default:
1335 printk (KERN_ERR 1377 printk (KERN_ERR
1336 "EXT4-fs: Unrecognized mount option \"%s\" " 1378 "EXT4-fs: Unrecognized mount option \"%s\" "
@@ -1443,6 +1485,54 @@ static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1443 return res; 1485 return res;
1444} 1486}
1445 1487
1488static int ext4_fill_flex_info(struct super_block *sb)
1489{
1490 struct ext4_sb_info *sbi = EXT4_SB(sb);
1491 struct ext4_group_desc *gdp = NULL;
1492 struct buffer_head *bh;
1493 ext4_group_t flex_group_count;
1494 ext4_group_t flex_group;
1495 int groups_per_flex = 0;
1496 __u64 block_bitmap = 0;
1497 int i;
1498
1499 if (!sbi->s_es->s_log_groups_per_flex) {
1500 sbi->s_log_groups_per_flex = 0;
1501 return 1;
1502 }
1503
1504 sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1505 groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1506
1507 flex_group_count = (sbi->s_groups_count + groups_per_flex - 1) /
1508 groups_per_flex;
1509 sbi->s_flex_groups = kmalloc(flex_group_count *
1510 sizeof(struct flex_groups), GFP_KERNEL);
1511 if (sbi->s_flex_groups == NULL) {
1512 printk(KERN_ERR "EXT4-fs: not enough memory\n");
1513 goto failed;
1514 }
1515 memset(sbi->s_flex_groups, 0, flex_group_count *
1516 sizeof(struct flex_groups));
1517
1518 gdp = ext4_get_group_desc(sb, 1, &bh);
1519 block_bitmap = ext4_block_bitmap(sb, gdp) - 1;
1520
1521 for (i = 0; i < sbi->s_groups_count; i++) {
1522 gdp = ext4_get_group_desc(sb, i, &bh);
1523
1524 flex_group = ext4_flex_group(sbi, i);
1525 sbi->s_flex_groups[flex_group].free_inodes +=
1526 le16_to_cpu(gdp->bg_free_inodes_count);
1527 sbi->s_flex_groups[flex_group].free_blocks +=
1528 le16_to_cpu(gdp->bg_free_blocks_count);
1529 }
1530
1531 return 1;
1532failed:
1533 return 0;
1534}
1535
1446__le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group, 1536__le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1447 struct ext4_group_desc *gdp) 1537 struct ext4_group_desc *gdp)
1448{ 1538{
@@ -1810,8 +1900,8 @@ static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
1810} 1900}
1811 1901
1812static int ext4_fill_super (struct super_block *sb, void *data, int silent) 1902static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1813 __releases(kernel_sem) 1903 __releases(kernel_lock)
1814 __acquires(kernel_sem) 1904 __acquires(kernel_lock)
1815 1905
1816{ 1906{
1817 struct buffer_head * bh; 1907 struct buffer_head * bh;
@@ -1851,11 +1941,6 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1851 goto out_fail; 1941 goto out_fail;
1852 } 1942 }
1853 1943
1854 if (!sb_set_blocksize(sb, blocksize)) {
1855 printk(KERN_ERR "EXT4-fs: bad blocksize %d.\n", blocksize);
1856 goto out_fail;
1857 }
1858
1859 /* 1944 /*
1860 * The ext4 superblock will not be buffer aligned for other than 1kB 1945 * The ext4 superblock will not be buffer aligned for other than 1kB
1861 * block sizes. We need to calculate the offset from buffer start. 1946 * block sizes. We need to calculate the offset from buffer start.
@@ -1919,15 +2004,28 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
1919 2004
1920 /* 2005 /*
1921 * turn on extents feature by default in ext4 filesystem 2006 * turn on extents feature by default in ext4 filesystem
1922 * User -o noextents to turn it off 2007 * only if feature flag already set by mkfs or tune2fs.
2008 * Use -o noextents to turn it off
1923 */ 2009 */
1924 set_opt(sbi->s_mount_opt, EXTENTS); 2010 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
2011 set_opt(sbi->s_mount_opt, EXTENTS);
2012 else
2013 ext4_warning(sb, __func__,
2014 "extents feature not enabled on this filesystem, "
2015 "use tune2fs.\n");
1925 /* 2016 /*
1926 * turn on mballoc feature by default in ext4 filesystem 2017 * turn on mballoc code by default in ext4 filesystem
1927 * User -o nomballoc to turn it off 2018 * Use -o nomballoc to turn it off
1928 */ 2019 */
1929 set_opt(sbi->s_mount_opt, MBALLOC); 2020 set_opt(sbi->s_mount_opt, MBALLOC);
1930 2021
2022 /*
2023 * enable delayed allocation by default
2024 * Use -o nodelalloc to turn it off
2025 */
2026 set_opt(sbi->s_mount_opt, DELALLOC);
2027
2028
1931 if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum, 2029 if (!parse_options ((char *) data, sb, &journal_inum, &journal_devnum,
1932 NULL, 0)) 2030 NULL, 0))
1933 goto failed_mount; 2031 goto failed_mount;
@@ -2138,6 +2236,14 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
2138 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n"); 2236 printk(KERN_ERR "EXT4-fs: group descriptors corrupted!\n");
2139 goto failed_mount2; 2237 goto failed_mount2;
2140 } 2238 }
2239 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2240 if (!ext4_fill_flex_info(sb)) {
2241 printk(KERN_ERR
2242 "EXT4-fs: unable to initialize "
2243 "flex_bg meta info!\n");
2244 goto failed_mount2;
2245 }
2246
2141 sbi->s_gdb_count = db_count; 2247 sbi->s_gdb_count = db_count;
2142 get_random_bytes(&sbi->s_next_generation, sizeof(u32)); 2248 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
2143 spin_lock_init(&sbi->s_next_gen_lock); 2249 spin_lock_init(&sbi->s_next_gen_lock);
@@ -2358,6 +2464,13 @@ static int ext4_fill_super (struct super_block *sb, void *data, int silent)
2358 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered": 2464 test_opt(sb,DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA ? "ordered":
2359 "writeback"); 2465 "writeback");
2360 2466
2467 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
2468 printk(KERN_WARNING "EXT4-fs: Ignoring delalloc option - "
2469 "requested data journaling mode\n");
2470 clear_opt(sbi->s_mount_opt, DELALLOC);
2471 } else if (test_opt(sb, DELALLOC))
2472 printk(KERN_INFO "EXT4-fs: delayed allocation enabled\n");
2473
2361 ext4_ext_init(sb); 2474 ext4_ext_init(sb);
2362 ext4_mb_init(sb, needs_recovery); 2475 ext4_mb_init(sb, needs_recovery);
2363 2476
@@ -2372,6 +2485,7 @@ cantfind_ext4:
2372 2485
2373failed_mount4: 2486failed_mount4:
2374 jbd2_journal_destroy(sbi->s_journal); 2487 jbd2_journal_destroy(sbi->s_journal);
2488 sbi->s_journal = NULL;
2375failed_mount3: 2489failed_mount3:
2376 percpu_counter_destroy(&sbi->s_freeblocks_counter); 2490 percpu_counter_destroy(&sbi->s_freeblocks_counter);
2377 percpu_counter_destroy(&sbi->s_freeinodes_counter); 2491 percpu_counter_destroy(&sbi->s_freeinodes_counter);
@@ -3325,7 +3439,7 @@ static ssize_t ext4_quota_write(struct super_block *sb, int type,
3325 err = ext4_journal_dirty_metadata(handle, bh); 3439 err = ext4_journal_dirty_metadata(handle, bh);
3326 else { 3440 else {
3327 /* Always do at least ordered writes for quotas */ 3441 /* Always do at least ordered writes for quotas */
3328 err = ext4_journal_dirty_data(handle, bh); 3442 err = ext4_jbd2_file_inode(handle, inode);
3329 mark_buffer_dirty(bh); 3443 mark_buffer_dirty(bh);
3330 } 3444 }
3331 brelse(bh); 3445 brelse(bh);
diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c
index ff08633f398e..93c5fdcdad2e 100644
--- a/fs/ext4/xattr.c
+++ b/fs/ext4/xattr.c
@@ -810,7 +810,7 @@ inserted:
810 /* We need to allocate a new block */ 810 /* We need to allocate a new block */
811 ext4_fsblk_t goal = ext4_group_first_block_no(sb, 811 ext4_fsblk_t goal = ext4_group_first_block_no(sb,
812 EXT4_I(inode)->i_block_group); 812 EXT4_I(inode)->i_block_group);
813 ext4_fsblk_t block = ext4_new_block(handle, inode, 813 ext4_fsblk_t block = ext4_new_meta_block(handle, inode,
814 goal, &error); 814 goal, &error);
815 if (error) 815 if (error)
816 goto cleanup; 816 goto cleanup;
diff --git a/fs/ext4/xattr_trusted.c b/fs/ext4/xattr_trusted.c
index fff33382cadc..ac1a52cf2a37 100644
--- a/fs/ext4/xattr_trusted.c
+++ b/fs/ext4/xattr_trusted.c
@@ -13,13 +13,11 @@
13#include "ext4.h" 13#include "ext4.h"
14#include "xattr.h" 14#include "xattr.h"
15 15
16#define XATTR_TRUSTED_PREFIX "trusted."
17
18static size_t 16static size_t
19ext4_xattr_trusted_list(struct inode *inode, char *list, size_t list_size, 17ext4_xattr_trusted_list(struct inode *inode, char *list, size_t list_size,
20 const char *name, size_t name_len) 18 const char *name, size_t name_len)
21{ 19{
22 const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX)-1; 20 const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
23 const size_t total_len = prefix_len + name_len + 1; 21 const size_t total_len = prefix_len + name_len + 1;
24 22
25 if (!capable(CAP_SYS_ADMIN)) 23 if (!capable(CAP_SYS_ADMIN))
diff --git a/fs/ext4/xattr_user.c b/fs/ext4/xattr_user.c
index 67be723fcc4e..d91aa61b42aa 100644
--- a/fs/ext4/xattr_user.c
+++ b/fs/ext4/xattr_user.c
@@ -12,13 +12,11 @@
12#include "ext4.h" 12#include "ext4.h"
13#include "xattr.h" 13#include "xattr.h"
14 14
15#define XATTR_USER_PREFIX "user."
16
17static size_t 15static size_t
18ext4_xattr_user_list(struct inode *inode, char *list, size_t list_size, 16ext4_xattr_user_list(struct inode *inode, char *list, size_t list_size,
19 const char *name, size_t name_len) 17 const char *name, size_t name_len)
20{ 18{
21 const size_t prefix_len = sizeof(XATTR_USER_PREFIX)-1; 19 const size_t prefix_len = XATTR_USER_PREFIX_LEN;
22 const size_t total_len = prefix_len + name_len + 1; 20 const size_t total_len = prefix_len + name_len + 1;
23 21
24 if (!test_opt(inode->i_sb, XATTR_USER)) 22 if (!test_opt(inode->i_sb, XATTR_USER))
diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
index 6914598022ce..91389c8aee8a 100644
--- a/fs/jbd2/checkpoint.c
+++ b/fs/jbd2/checkpoint.c
@@ -688,7 +688,6 @@ void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transact
688 688
689 J_ASSERT(transaction->t_state == T_FINISHED); 689 J_ASSERT(transaction->t_state == T_FINISHED);
690 J_ASSERT(transaction->t_buffers == NULL); 690 J_ASSERT(transaction->t_buffers == NULL);
691 J_ASSERT(transaction->t_sync_datalist == NULL);
692 J_ASSERT(transaction->t_forget == NULL); 691 J_ASSERT(transaction->t_forget == NULL);
693 J_ASSERT(transaction->t_iobuf_list == NULL); 692 J_ASSERT(transaction->t_iobuf_list == NULL);
694 J_ASSERT(transaction->t_shadow_list == NULL); 693 J_ASSERT(transaction->t_shadow_list == NULL);
diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
index a2ed72f7ceee..f8b3be873226 100644
--- a/fs/jbd2/commit.c
+++ b/fs/jbd2/commit.c
@@ -22,6 +22,8 @@
22#include <linux/pagemap.h> 22#include <linux/pagemap.h>
23#include <linux/jiffies.h> 23#include <linux/jiffies.h>
24#include <linux/crc32.h> 24#include <linux/crc32.h>
25#include <linux/writeback.h>
26#include <linux/backing-dev.h>
25 27
26/* 28/*
27 * Default IO end handler for temporary BJ_IO buffer_heads. 29 * Default IO end handler for temporary BJ_IO buffer_heads.
@@ -37,8 +39,8 @@ static void journal_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
37} 39}
38 40
39/* 41/*
40 * When an ext3-ordered file is truncated, it is possible that many pages are 42 * When an ext4 file is truncated, it is possible that some pages are not
41 * not sucessfully freed, because they are attached to a committing transaction. 43 * successfully freed, because they are attached to a committing transaction.
42 * After the transaction commits, these pages are left on the LRU, with no 44 * After the transaction commits, these pages are left on the LRU, with no
43 * ->mapping, and with attached buffers. These pages are trivially reclaimable 45 * ->mapping, and with attached buffers. These pages are trivially reclaimable
44 * by the VM, but their apparent absence upsets the VM accounting, and it makes 46 * by the VM, but their apparent absence upsets the VM accounting, and it makes
@@ -80,21 +82,6 @@ nope:
80} 82}
81 83
82/* 84/*
83 * Try to acquire jbd_lock_bh_state() against the buffer, when j_list_lock is
84 * held. For ranking reasons we must trylock. If we lose, schedule away and
85 * return 0. j_list_lock is dropped in this case.
86 */
87static int inverted_lock(journal_t *journal, struct buffer_head *bh)
88{
89 if (!jbd_trylock_bh_state(bh)) {
90 spin_unlock(&journal->j_list_lock);
91 schedule();
92 return 0;
93 }
94 return 1;
95}
96
97/*
98 * Done it all: now submit the commit record. We should have 85 * Done it all: now submit the commit record. We should have
99 * cleaned up our previous buffers by now, so if we are in abort 86 * cleaned up our previous buffers by now, so if we are in abort
100 * mode we can now just skip the rest of the journal write 87 * mode we can now just skip the rest of the journal write
@@ -112,6 +99,7 @@ static int journal_submit_commit_record(journal_t *journal,
112 struct buffer_head *bh; 99 struct buffer_head *bh;
113 int ret; 100 int ret;
114 int barrier_done = 0; 101 int barrier_done = 0;
102 struct timespec now = current_kernel_time();
115 103
116 if (is_journal_aborted(journal)) 104 if (is_journal_aborted(journal))
117 return 0; 105 return 0;
@@ -126,6 +114,8 @@ static int journal_submit_commit_record(journal_t *journal,
126 tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); 114 tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
127 tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK); 115 tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
128 tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid); 116 tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
117 tmp->h_commit_sec = cpu_to_be64(now.tv_sec);
118 tmp->h_commit_nsec = cpu_to_be32(now.tv_nsec);
129 119
130 if (JBD2_HAS_COMPAT_FEATURE(journal, 120 if (JBD2_HAS_COMPAT_FEATURE(journal,
131 JBD2_FEATURE_COMPAT_CHECKSUM)) { 121 JBD2_FEATURE_COMPAT_CHECKSUM)) {
@@ -197,159 +187,104 @@ static int journal_wait_on_commit_record(struct buffer_head *bh)
197} 187}
198 188
199/* 189/*
200 * Wait for all submitted IO to complete. 190 * write the filemap data using writepage() address_space_operations.
191 * We don't do block allocation here even for delalloc. We don't
192 * use writepages() because with dealyed allocation we may be doing
193 * block allocation in writepages().
201 */ 194 */
202static int journal_wait_on_locked_list(journal_t *journal, 195static int journal_submit_inode_data_buffers(struct address_space *mapping)
203 transaction_t *commit_transaction)
204{ 196{
205 int ret = 0; 197 int ret;
206 struct journal_head *jh; 198 struct writeback_control wbc = {
207 199 .sync_mode = WB_SYNC_ALL,
208 while (commit_transaction->t_locked_list) { 200 .nr_to_write = mapping->nrpages * 2,
209 struct buffer_head *bh; 201 .range_start = 0,
210 202 .range_end = i_size_read(mapping->host),
211 jh = commit_transaction->t_locked_list->b_tprev; 203 .for_writepages = 1,
212 bh = jh2bh(jh); 204 };
213 get_bh(bh); 205
214 if (buffer_locked(bh)) { 206 ret = generic_writepages(mapping, &wbc);
215 spin_unlock(&journal->j_list_lock);
216 wait_on_buffer(bh);
217 if (unlikely(!buffer_uptodate(bh)))
218 ret = -EIO;
219 spin_lock(&journal->j_list_lock);
220 }
221 if (!inverted_lock(journal, bh)) {
222 put_bh(bh);
223 spin_lock(&journal->j_list_lock);
224 continue;
225 }
226 if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
227 __jbd2_journal_unfile_buffer(jh);
228 jbd_unlock_bh_state(bh);
229 jbd2_journal_remove_journal_head(bh);
230 put_bh(bh);
231 } else {
232 jbd_unlock_bh_state(bh);
233 }
234 put_bh(bh);
235 cond_resched_lock(&journal->j_list_lock);
236 }
237 return ret; 207 return ret;
238 } 208}
239 209
240static void journal_do_submit_data(struct buffer_head **wbuf, int bufs) 210/*
211 * Submit all the data buffers of inode associated with the transaction to
212 * disk.
213 *
214 * We are in a committing transaction. Therefore no new inode can be added to
215 * our inode list. We use JI_COMMIT_RUNNING flag to protect inode we currently
216 * operate on from being released while we write out pages.
217 */
218static int journal_submit_data_buffers(journal_t *journal,
219 transaction_t *commit_transaction)
241{ 220{
242 int i; 221 struct jbd2_inode *jinode;
222 int err, ret = 0;
223 struct address_space *mapping;
243 224
244 for (i = 0; i < bufs; i++) { 225 spin_lock(&journal->j_list_lock);
245 wbuf[i]->b_end_io = end_buffer_write_sync; 226 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
246 /* We use-up our safety reference in submit_bh() */ 227 mapping = jinode->i_vfs_inode->i_mapping;
247 submit_bh(WRITE, wbuf[i]); 228 jinode->i_flags |= JI_COMMIT_RUNNING;
229 spin_unlock(&journal->j_list_lock);
230 /*
231 * submit the inode data buffers. We use writepage
232 * instead of writepages. Because writepages can do
233 * block allocation with delalloc. We need to write
234 * only allocated blocks here.
235 */
236 err = journal_submit_inode_data_buffers(mapping);
237 if (!ret)
238 ret = err;
239 spin_lock(&journal->j_list_lock);
240 J_ASSERT(jinode->i_transaction == commit_transaction);
241 jinode->i_flags &= ~JI_COMMIT_RUNNING;
242 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
248 } 243 }
244 spin_unlock(&journal->j_list_lock);
245 return ret;
249} 246}
250 247
251/* 248/*
252 * Submit all the data buffers to disk 249 * Wait for data submitted for writeout, refile inodes to proper
250 * transaction if needed.
251 *
253 */ 252 */
254static void journal_submit_data_buffers(journal_t *journal, 253static int journal_finish_inode_data_buffers(journal_t *journal,
255 transaction_t *commit_transaction) 254 transaction_t *commit_transaction)
256{ 255{
257 struct journal_head *jh; 256 struct jbd2_inode *jinode, *next_i;
258 struct buffer_head *bh; 257 int err, ret = 0;
259 int locked;
260 int bufs = 0;
261 struct buffer_head **wbuf = journal->j_wbuf;
262 258
263 /* 259 /* For locking, see the comment in journal_submit_data_buffers() */
264 * Whenever we unlock the journal and sleep, things can get added
265 * onto ->t_sync_datalist, so we have to keep looping back to
266 * write_out_data until we *know* that the list is empty.
267 *
268 * Cleanup any flushed data buffers from the data list. Even in
269 * abort mode, we want to flush this out as soon as possible.
270 */
271write_out_data:
272 cond_resched();
273 spin_lock(&journal->j_list_lock); 260 spin_lock(&journal->j_list_lock);
261 list_for_each_entry(jinode, &commit_transaction->t_inode_list, i_list) {
262 jinode->i_flags |= JI_COMMIT_RUNNING;
263 spin_unlock(&journal->j_list_lock);
264 err = filemap_fdatawait(jinode->i_vfs_inode->i_mapping);
265 if (!ret)
266 ret = err;
267 spin_lock(&journal->j_list_lock);
268 jinode->i_flags &= ~JI_COMMIT_RUNNING;
269 wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
270 }
274 271
275 while (commit_transaction->t_sync_datalist) { 272 /* Now refile inode to proper lists */
276 jh = commit_transaction->t_sync_datalist; 273 list_for_each_entry_safe(jinode, next_i,
277 bh = jh2bh(jh); 274 &commit_transaction->t_inode_list, i_list) {
278 locked = 0; 275 list_del(&jinode->i_list);
279 276 if (jinode->i_next_transaction) {
280 /* Get reference just to make sure buffer does not disappear 277 jinode->i_transaction = jinode->i_next_transaction;
281 * when we are forced to drop various locks */ 278 jinode->i_next_transaction = NULL;
282 get_bh(bh); 279 list_add(&jinode->i_list,
283 /* If the buffer is dirty, we need to submit IO and hence 280 &jinode->i_transaction->t_inode_list);
284 * we need the buffer lock. We try to lock the buffer without
285 * blocking. If we fail, we need to drop j_list_lock and do
286 * blocking lock_buffer().
287 */
288 if (buffer_dirty(bh)) {
289 if (test_set_buffer_locked(bh)) {
290 BUFFER_TRACE(bh, "needs blocking lock");
291 spin_unlock(&journal->j_list_lock);
292 /* Write out all data to prevent deadlocks */
293 journal_do_submit_data(wbuf, bufs);
294 bufs = 0;
295 lock_buffer(bh);
296 spin_lock(&journal->j_list_lock);
297 }
298 locked = 1;
299 }
300 /* We have to get bh_state lock. Again out of order, sigh. */
301 if (!inverted_lock(journal, bh)) {
302 jbd_lock_bh_state(bh);
303 spin_lock(&journal->j_list_lock);
304 }
305 /* Someone already cleaned up the buffer? */
306 if (!buffer_jbd(bh)
307 || jh->b_transaction != commit_transaction
308 || jh->b_jlist != BJ_SyncData) {
309 jbd_unlock_bh_state(bh);
310 if (locked)
311 unlock_buffer(bh);
312 BUFFER_TRACE(bh, "already cleaned up");
313 put_bh(bh);
314 continue;
315 }
316 if (locked && test_clear_buffer_dirty(bh)) {
317 BUFFER_TRACE(bh, "needs writeout, adding to array");
318 wbuf[bufs++] = bh;
319 __jbd2_journal_file_buffer(jh, commit_transaction,
320 BJ_Locked);
321 jbd_unlock_bh_state(bh);
322 if (bufs == journal->j_wbufsize) {
323 spin_unlock(&journal->j_list_lock);
324 journal_do_submit_data(wbuf, bufs);
325 bufs = 0;
326 goto write_out_data;
327 }
328 } else if (!locked && buffer_locked(bh)) {
329 __jbd2_journal_file_buffer(jh, commit_transaction,
330 BJ_Locked);
331 jbd_unlock_bh_state(bh);
332 put_bh(bh);
333 } else { 281 } else {
334 BUFFER_TRACE(bh, "writeout complete: unfile"); 282 jinode->i_transaction = NULL;
335 __jbd2_journal_unfile_buffer(jh);
336 jbd_unlock_bh_state(bh);
337 if (locked)
338 unlock_buffer(bh);
339 jbd2_journal_remove_journal_head(bh);
340 /* Once for our safety reference, once for
341 * jbd2_journal_remove_journal_head() */
342 put_bh(bh);
343 put_bh(bh);
344 }
345
346 if (need_resched() || spin_needbreak(&journal->j_list_lock)) {
347 spin_unlock(&journal->j_list_lock);
348 goto write_out_data;
349 } 283 }
350 } 284 }
351 spin_unlock(&journal->j_list_lock); 285 spin_unlock(&journal->j_list_lock);
352 journal_do_submit_data(wbuf, bufs); 286
287 return ret;
353} 288}
354 289
355static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh) 290static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
@@ -524,21 +459,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
524 * Now start flushing things to disk, in the order they appear 459 * Now start flushing things to disk, in the order they appear
525 * on the transaction lists. Data blocks go first. 460 * on the transaction lists. Data blocks go first.
526 */ 461 */
527 err = 0; 462 err = journal_submit_data_buffers(journal, commit_transaction);
528 journal_submit_data_buffers(journal, commit_transaction);
529
530 /*
531 * Wait for all previously submitted IO to complete if commit
532 * record is to be written synchronously.
533 */
534 spin_lock(&journal->j_list_lock);
535 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
536 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
537 err = journal_wait_on_locked_list(journal,
538 commit_transaction);
539
540 spin_unlock(&journal->j_list_lock);
541
542 if (err) 463 if (err)
543 jbd2_journal_abort(journal, err); 464 jbd2_journal_abort(journal, err);
544 465
@@ -547,16 +468,6 @@ void jbd2_journal_commit_transaction(journal_t *journal)
547 jbd_debug(3, "JBD: commit phase 2\n"); 468 jbd_debug(3, "JBD: commit phase 2\n");
548 469
549 /* 470 /*
550 * If we found any dirty or locked buffers, then we should have
551 * looped back up to the write_out_data label. If there weren't
552 * any then journal_clean_data_list should have wiped the list
553 * clean by now, so check that it is in fact empty.
554 */
555 J_ASSERT (commit_transaction->t_sync_datalist == NULL);
556
557 jbd_debug (3, "JBD: commit phase 3\n");
558
559 /*
560 * Way to go: we have now written out all of the data for a 471 * Way to go: we have now written out all of the data for a
561 * transaction! Now comes the tricky part: we need to write out 472 * transaction! Now comes the tricky part: we need to write out
562 * metadata. Loop over the transaction's entire buffer list: 473 * metadata. Loop over the transaction's entire buffer list:
@@ -574,6 +485,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
574 J_ASSERT(commit_transaction->t_nr_buffers <= 485 J_ASSERT(commit_transaction->t_nr_buffers <=
575 commit_transaction->t_outstanding_credits); 486 commit_transaction->t_outstanding_credits);
576 487
488 err = 0;
577 descriptor = NULL; 489 descriptor = NULL;
578 bufs = 0; 490 bufs = 0;
579 while (commit_transaction->t_buffers) { 491 while (commit_transaction->t_buffers) {
@@ -748,15 +660,19 @@ start_journal_io:
748 &cbh, crc32_sum); 660 &cbh, crc32_sum);
749 if (err) 661 if (err)
750 __jbd2_journal_abort_hard(journal); 662 __jbd2_journal_abort_hard(journal);
751
752 spin_lock(&journal->j_list_lock);
753 err = journal_wait_on_locked_list(journal,
754 commit_transaction);
755 spin_unlock(&journal->j_list_lock);
756 if (err)
757 __jbd2_journal_abort_hard(journal);
758 } 663 }
759 664
665 /*
666 * This is the right place to wait for data buffers both for ASYNC
667 * and !ASYNC commit. If commit is ASYNC, we need to wait only after
668 * the commit block went to disk (which happens above). If commit is
669 * SYNC, we need to wait for data buffers before we start writing
670 * commit block, which happens below in such setting.
671 */
672 err = journal_finish_inode_data_buffers(journal, commit_transaction);
673 if (err)
674 jbd2_journal_abort(journal, err);
675
760 /* Lo and behold: we have just managed to send a transaction to 676 /* Lo and behold: we have just managed to send a transaction to
761 the log. Before we can commit it, wait for the IO so far to 677 the log. Before we can commit it, wait for the IO so far to
762 complete. Control buffers being written are on the 678 complete. Control buffers being written are on the
@@ -768,7 +684,7 @@ start_journal_io:
768 so we incur less scheduling load. 684 so we incur less scheduling load.
769 */ 685 */
770 686
771 jbd_debug(3, "JBD: commit phase 4\n"); 687 jbd_debug(3, "JBD: commit phase 3\n");
772 688
773 /* 689 /*
774 * akpm: these are BJ_IO, and j_list_lock is not needed. 690 * akpm: these are BJ_IO, and j_list_lock is not needed.
@@ -827,7 +743,7 @@ wait_for_iobuf:
827 743
828 J_ASSERT (commit_transaction->t_shadow_list == NULL); 744 J_ASSERT (commit_transaction->t_shadow_list == NULL);
829 745
830 jbd_debug(3, "JBD: commit phase 5\n"); 746 jbd_debug(3, "JBD: commit phase 4\n");
831 747
832 /* Here we wait for the revoke record and descriptor record buffers */ 748 /* Here we wait for the revoke record and descriptor record buffers */
833 wait_for_ctlbuf: 749 wait_for_ctlbuf:
@@ -854,7 +770,7 @@ wait_for_iobuf:
854 /* AKPM: bforget here */ 770 /* AKPM: bforget here */
855 } 771 }
856 772
857 jbd_debug(3, "JBD: commit phase 6\n"); 773 jbd_debug(3, "JBD: commit phase 5\n");
858 774
859 if (!JBD2_HAS_INCOMPAT_FEATURE(journal, 775 if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
860 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) { 776 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
@@ -874,9 +790,9 @@ wait_for_iobuf:
874 transaction can be removed from any checkpoint list it was on 790 transaction can be removed from any checkpoint list it was on
875 before. */ 791 before. */
876 792
877 jbd_debug(3, "JBD: commit phase 7\n"); 793 jbd_debug(3, "JBD: commit phase 6\n");
878 794
879 J_ASSERT(commit_transaction->t_sync_datalist == NULL); 795 J_ASSERT(list_empty(&commit_transaction->t_inode_list));
880 J_ASSERT(commit_transaction->t_buffers == NULL); 796 J_ASSERT(commit_transaction->t_buffers == NULL);
881 J_ASSERT(commit_transaction->t_checkpoint_list == NULL); 797 J_ASSERT(commit_transaction->t_checkpoint_list == NULL);
882 J_ASSERT(commit_transaction->t_iobuf_list == NULL); 798 J_ASSERT(commit_transaction->t_iobuf_list == NULL);
@@ -997,7 +913,7 @@ restart_loop:
997 913
998 /* Done with this transaction! */ 914 /* Done with this transaction! */
999 915
1000 jbd_debug(3, "JBD: commit phase 8\n"); 916 jbd_debug(3, "JBD: commit phase 7\n");
1001 917
1002 J_ASSERT(commit_transaction->t_state == T_COMMIT); 918 J_ASSERT(commit_transaction->t_state == T_COMMIT);
1003 919
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 2e24567c4a79..b26c6d9fe6ae 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -50,7 +50,6 @@ EXPORT_SYMBOL(jbd2_journal_unlock_updates);
50EXPORT_SYMBOL(jbd2_journal_get_write_access); 50EXPORT_SYMBOL(jbd2_journal_get_write_access);
51EXPORT_SYMBOL(jbd2_journal_get_create_access); 51EXPORT_SYMBOL(jbd2_journal_get_create_access);
52EXPORT_SYMBOL(jbd2_journal_get_undo_access); 52EXPORT_SYMBOL(jbd2_journal_get_undo_access);
53EXPORT_SYMBOL(jbd2_journal_dirty_data);
54EXPORT_SYMBOL(jbd2_journal_dirty_metadata); 53EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
55EXPORT_SYMBOL(jbd2_journal_release_buffer); 54EXPORT_SYMBOL(jbd2_journal_release_buffer);
56EXPORT_SYMBOL(jbd2_journal_forget); 55EXPORT_SYMBOL(jbd2_journal_forget);
@@ -82,6 +81,10 @@ EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
82EXPORT_SYMBOL(jbd2_journal_invalidatepage); 81EXPORT_SYMBOL(jbd2_journal_invalidatepage);
83EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); 82EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
84EXPORT_SYMBOL(jbd2_journal_force_commit); 83EXPORT_SYMBOL(jbd2_journal_force_commit);
84EXPORT_SYMBOL(jbd2_journal_file_inode);
85EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
86EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
87EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
85 88
86static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *); 89static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
87static void __journal_abort_soft (journal_t *journal, int errno); 90static void __journal_abort_soft (journal_t *journal, int errno);
@@ -2195,6 +2198,54 @@ void jbd2_journal_put_journal_head(struct journal_head *jh)
2195} 2198}
2196 2199
2197/* 2200/*
2201 * Initialize jbd inode head
2202 */
2203void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2204{
2205 jinode->i_transaction = NULL;
2206 jinode->i_next_transaction = NULL;
2207 jinode->i_vfs_inode = inode;
2208 jinode->i_flags = 0;
2209 INIT_LIST_HEAD(&jinode->i_list);
2210}
2211
2212/*
2213 * Function to be called before we start removing inode from memory (i.e.,
2214 * clear_inode() is a fine place to be called from). It removes inode from
2215 * transaction's lists.
2216 */
2217void jbd2_journal_release_jbd_inode(journal_t *journal,
2218 struct jbd2_inode *jinode)
2219{
2220 int writeout = 0;
2221
2222 if (!journal)
2223 return;
2224restart:
2225 spin_lock(&journal->j_list_lock);
2226 /* Is commit writing out inode - we have to wait */
2227 if (jinode->i_flags & JI_COMMIT_RUNNING) {
2228 wait_queue_head_t *wq;
2229 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2230 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2231 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2232 spin_unlock(&journal->j_list_lock);
2233 schedule();
2234 finish_wait(wq, &wait.wait);
2235 goto restart;
2236 }
2237
2238 /* Do we need to wait for data writeback? */
2239 if (journal->j_committing_transaction == jinode->i_transaction)
2240 writeout = 1;
2241 if (jinode->i_transaction) {
2242 list_del(&jinode->i_list);
2243 jinode->i_transaction = NULL;
2244 }
2245 spin_unlock(&journal->j_list_lock);
2246}
2247
2248/*
2198 * debugfs tunables 2249 * debugfs tunables
2199 */ 2250 */
2200#ifdef CONFIG_JBD2_DEBUG 2251#ifdef CONFIG_JBD2_DEBUG
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index d6e006e67804..4f7cadbb19fa 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -41,7 +41,6 @@ static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
41 * new transaction and we can't block without protecting against other 41 * new transaction and we can't block without protecting against other
42 * processes trying to touch the journal while it is in transition. 42 * processes trying to touch the journal while it is in transition.
43 * 43 *
44 * Called under j_state_lock
45 */ 44 */
46 45
47static transaction_t * 46static transaction_t *
@@ -52,6 +51,7 @@ jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
52 transaction->t_tid = journal->j_transaction_sequence++; 51 transaction->t_tid = journal->j_transaction_sequence++;
53 transaction->t_expires = jiffies + journal->j_commit_interval; 52 transaction->t_expires = jiffies + journal->j_commit_interval;
54 spin_lock_init(&transaction->t_handle_lock); 53 spin_lock_init(&transaction->t_handle_lock);
54 INIT_LIST_HEAD(&transaction->t_inode_list);
55 55
56 /* Set up the commit timer for the new transaction. */ 56 /* Set up the commit timer for the new transaction. */
57 journal->j_commit_timer.expires = round_jiffies(transaction->t_expires); 57 journal->j_commit_timer.expires = round_jiffies(transaction->t_expires);
@@ -943,183 +943,6 @@ out:
943} 943}
944 944
945/** 945/**
946 * int jbd2_journal_dirty_data() - mark a buffer as containing dirty data which
947 * needs to be flushed before we can commit the
948 * current transaction.
949 * @handle: transaction
950 * @bh: bufferhead to mark
951 *
952 * The buffer is placed on the transaction's data list and is marked as
953 * belonging to the transaction.
954 *
955 * Returns error number or 0 on success.
956 *
957 * jbd2_journal_dirty_data() can be called via page_launder->ext3_writepage
958 * by kswapd.
959 */
960int jbd2_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
961{
962 journal_t *journal = handle->h_transaction->t_journal;
963 int need_brelse = 0;
964 struct journal_head *jh;
965
966 if (is_handle_aborted(handle))
967 return 0;
968
969 jh = jbd2_journal_add_journal_head(bh);
970 JBUFFER_TRACE(jh, "entry");
971
972 /*
973 * The buffer could *already* be dirty. Writeout can start
974 * at any time.
975 */
976 jbd_debug(4, "jh: %p, tid:%d\n", jh, handle->h_transaction->t_tid);
977
978 /*
979 * What if the buffer is already part of a running transaction?
980 *
981 * There are two cases:
982 * 1) It is part of the current running transaction. Refile it,
983 * just in case we have allocated it as metadata, deallocated
984 * it, then reallocated it as data.
985 * 2) It is part of the previous, still-committing transaction.
986 * If all we want to do is to guarantee that the buffer will be
987 * written to disk before this new transaction commits, then
988 * being sure that the *previous* transaction has this same
989 * property is sufficient for us! Just leave it on its old
990 * transaction.
991 *
992 * In case (2), the buffer must not already exist as metadata
993 * --- that would violate write ordering (a transaction is free
994 * to write its data at any point, even before the previous
995 * committing transaction has committed). The caller must
996 * never, ever allow this to happen: there's nothing we can do
997 * about it in this layer.
998 */
999 jbd_lock_bh_state(bh);
1000 spin_lock(&journal->j_list_lock);
1001
1002 /* Now that we have bh_state locked, are we really still mapped? */
1003 if (!buffer_mapped(bh)) {
1004 JBUFFER_TRACE(jh, "unmapped buffer, bailing out");
1005 goto no_journal;
1006 }
1007
1008 if (jh->b_transaction) {
1009 JBUFFER_TRACE(jh, "has transaction");
1010 if (jh->b_transaction != handle->h_transaction) {
1011 JBUFFER_TRACE(jh, "belongs to older transaction");
1012 J_ASSERT_JH(jh, jh->b_transaction ==
1013 journal->j_committing_transaction);
1014
1015 /* @@@ IS THIS TRUE ? */
1016 /*
1017 * Not any more. Scenario: someone does a write()
1018 * in data=journal mode. The buffer's transaction has
1019 * moved into commit. Then someone does another
1020 * write() to the file. We do the frozen data copyout
1021 * and set b_next_transaction to point to j_running_t.
1022 * And while we're in that state, someone does a
1023 * writepage() in an attempt to pageout the same area
1024 * of the file via a shared mapping. At present that
1025 * calls jbd2_journal_dirty_data(), and we get right here.
1026 * It may be too late to journal the data. Simply
1027 * falling through to the next test will suffice: the
1028 * data will be dirty and wil be checkpointed. The
1029 * ordering comments in the next comment block still
1030 * apply.
1031 */
1032 //J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1033
1034 /*
1035 * If we're journalling data, and this buffer was
1036 * subject to a write(), it could be metadata, forget
1037 * or shadow against the committing transaction. Now,
1038 * someone has dirtied the same darn page via a mapping
1039 * and it is being writepage()'d.
1040 * We *could* just steal the page from commit, with some
1041 * fancy locking there. Instead, we just skip it -
1042 * don't tie the page's buffers to the new transaction
1043 * at all.
1044 * Implication: if we crash before the writepage() data
1045 * is written into the filesystem, recovery will replay
1046 * the write() data.
1047 */
1048 if (jh->b_jlist != BJ_None &&
1049 jh->b_jlist != BJ_SyncData &&
1050 jh->b_jlist != BJ_Locked) {
1051 JBUFFER_TRACE(jh, "Not stealing");
1052 goto no_journal;
1053 }
1054
1055 /*
1056 * This buffer may be undergoing writeout in commit. We
1057 * can't return from here and let the caller dirty it
1058 * again because that can cause the write-out loop in
1059 * commit to never terminate.
1060 */
1061 if (buffer_dirty(bh)) {
1062 get_bh(bh);
1063 spin_unlock(&journal->j_list_lock);
1064 jbd_unlock_bh_state(bh);
1065 need_brelse = 1;
1066 sync_dirty_buffer(bh);
1067 jbd_lock_bh_state(bh);
1068 spin_lock(&journal->j_list_lock);
1069 /* Since we dropped the lock... */
1070 if (!buffer_mapped(bh)) {
1071 JBUFFER_TRACE(jh, "buffer got unmapped");
1072 goto no_journal;
1073 }
1074 /* The buffer may become locked again at any
1075 time if it is redirtied */
1076 }
1077
1078 /* journal_clean_data_list() may have got there first */
1079 if (jh->b_transaction != NULL) {
1080 JBUFFER_TRACE(jh, "unfile from commit");
1081 __jbd2_journal_temp_unlink_buffer(jh);
1082 /* It still points to the committing
1083 * transaction; move it to this one so
1084 * that the refile assert checks are
1085 * happy. */
1086 jh->b_transaction = handle->h_transaction;
1087 }
1088 /* The buffer will be refiled below */
1089
1090 }
1091 /*
1092 * Special case --- the buffer might actually have been
1093 * allocated and then immediately deallocated in the previous,
1094 * committing transaction, so might still be left on that
1095 * transaction's metadata lists.
1096 */
1097 if (jh->b_jlist != BJ_SyncData && jh->b_jlist != BJ_Locked) {
1098 JBUFFER_TRACE(jh, "not on correct data list: unfile");
1099 J_ASSERT_JH(jh, jh->b_jlist != BJ_Shadow);
1100 __jbd2_journal_temp_unlink_buffer(jh);
1101 jh->b_transaction = handle->h_transaction;
1102 JBUFFER_TRACE(jh, "file as data");
1103 __jbd2_journal_file_buffer(jh, handle->h_transaction,
1104 BJ_SyncData);
1105 }
1106 } else {
1107 JBUFFER_TRACE(jh, "not on a transaction");
1108 __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_SyncData);
1109 }
1110no_journal:
1111 spin_unlock(&journal->j_list_lock);
1112 jbd_unlock_bh_state(bh);
1113 if (need_brelse) {
1114 BUFFER_TRACE(bh, "brelse");
1115 __brelse(bh);
1116 }
1117 JBUFFER_TRACE(jh, "exit");
1118 jbd2_journal_put_journal_head(jh);
1119 return 0;
1120}
1121
1122/**
1123 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata 946 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1124 * @handle: transaction to add buffer to. 947 * @handle: transaction to add buffer to.
1125 * @bh: buffer to mark 948 * @bh: buffer to mark
@@ -1541,10 +1364,10 @@ __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1541 * Remove a buffer from the appropriate transaction list. 1364 * Remove a buffer from the appropriate transaction list.
1542 * 1365 *
1543 * Note that this function can *change* the value of 1366 * Note that this function can *change* the value of
1544 * bh->b_transaction->t_sync_datalist, t_buffers, t_forget, 1367 * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list,
1545 * t_iobuf_list, t_shadow_list, t_log_list or t_reserved_list. If the caller 1368 * t_log_list or t_reserved_list. If the caller is holding onto a copy of one
1546 * is holding onto a copy of one of thee pointers, it could go bad. 1369 * of these pointers, it could go bad. Generally the caller needs to re-read
1547 * Generally the caller needs to re-read the pointer from the transaction_t. 1370 * the pointer from the transaction_t.
1548 * 1371 *
1549 * Called under j_list_lock. The journal may not be locked. 1372 * Called under j_list_lock. The journal may not be locked.
1550 */ 1373 */
@@ -1566,9 +1389,6 @@ void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1566 switch (jh->b_jlist) { 1389 switch (jh->b_jlist) {
1567 case BJ_None: 1390 case BJ_None:
1568 return; 1391 return;
1569 case BJ_SyncData:
1570 list = &transaction->t_sync_datalist;
1571 break;
1572 case BJ_Metadata: 1392 case BJ_Metadata:
1573 transaction->t_nr_buffers--; 1393 transaction->t_nr_buffers--;
1574 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); 1394 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
@@ -1589,9 +1409,6 @@ void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1589 case BJ_Reserved: 1409 case BJ_Reserved:
1590 list = &transaction->t_reserved_list; 1410 list = &transaction->t_reserved_list;
1591 break; 1411 break;
1592 case BJ_Locked:
1593 list = &transaction->t_locked_list;
1594 break;
1595 } 1412 }
1596 1413
1597 __blist_del_buffer(list, jh); 1414 __blist_del_buffer(list, jh);
@@ -1634,15 +1451,7 @@ __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1634 goto out; 1451 goto out;
1635 1452
1636 spin_lock(&journal->j_list_lock); 1453 spin_lock(&journal->j_list_lock);
1637 if (jh->b_transaction != NULL && jh->b_cp_transaction == NULL) { 1454 if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
1638 if (jh->b_jlist == BJ_SyncData || jh->b_jlist == BJ_Locked) {
1639 /* A written-back ordered data buffer */
1640 JBUFFER_TRACE(jh, "release data");
1641 __jbd2_journal_unfile_buffer(jh);
1642 jbd2_journal_remove_journal_head(bh);
1643 __brelse(bh);
1644 }
1645 } else if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
1646 /* written-back checkpointed metadata buffer */ 1455 /* written-back checkpointed metadata buffer */
1647 if (jh->b_jlist == BJ_None) { 1456 if (jh->b_jlist == BJ_None) {
1648 JBUFFER_TRACE(jh, "remove from checkpoint list"); 1457 JBUFFER_TRACE(jh, "remove from checkpoint list");
@@ -1656,12 +1465,43 @@ out:
1656 return; 1465 return;
1657} 1466}
1658 1467
1468/*
1469 * jbd2_journal_try_to_free_buffers() could race with
1470 * jbd2_journal_commit_transaction(). The later might still hold the
1471 * reference count to the buffers when inspecting them on
1472 * t_syncdata_list or t_locked_list.
1473 *
1474 * jbd2_journal_try_to_free_buffers() will call this function to
1475 * wait for the current transaction to finish syncing data buffers, before
1476 * try to free that buffer.
1477 *
1478 * Called with journal->j_state_lock hold.
1479 */
1480static void jbd2_journal_wait_for_transaction_sync_data(journal_t *journal)
1481{
1482 transaction_t *transaction;
1483 tid_t tid;
1484
1485 spin_lock(&journal->j_state_lock);
1486 transaction = journal->j_committing_transaction;
1487
1488 if (!transaction) {
1489 spin_unlock(&journal->j_state_lock);
1490 return;
1491 }
1492
1493 tid = transaction->t_tid;
1494 spin_unlock(&journal->j_state_lock);
1495 jbd2_log_wait_commit(journal, tid);
1496}
1659 1497
1660/** 1498/**
1661 * int jbd2_journal_try_to_free_buffers() - try to free page buffers. 1499 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
1662 * @journal: journal for operation 1500 * @journal: journal for operation
1663 * @page: to try and free 1501 * @page: to try and free
1664 * @unused_gfp_mask: unused 1502 * @gfp_mask: we use the mask to detect how hard should we try to release
1503 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1504 * release the buffers.
1665 * 1505 *
1666 * 1506 *
1667 * For all the buffers on this page, 1507 * For all the buffers on this page,
@@ -1690,9 +1530,11 @@ out:
1690 * journal_try_to_free_buffer() is changing its state. But that 1530 * journal_try_to_free_buffer() is changing its state. But that
1691 * cannot happen because we never reallocate freed data as metadata 1531 * cannot happen because we never reallocate freed data as metadata
1692 * while the data is part of a transaction. Yes? 1532 * while the data is part of a transaction. Yes?
1533 *
1534 * Return 0 on failure, 1 on success
1693 */ 1535 */
1694int jbd2_journal_try_to_free_buffers(journal_t *journal, 1536int jbd2_journal_try_to_free_buffers(journal_t *journal,
1695 struct page *page, gfp_t unused_gfp_mask) 1537 struct page *page, gfp_t gfp_mask)
1696{ 1538{
1697 struct buffer_head *head; 1539 struct buffer_head *head;
1698 struct buffer_head *bh; 1540 struct buffer_head *bh;
@@ -1708,7 +1550,8 @@ int jbd2_journal_try_to_free_buffers(journal_t *journal,
1708 /* 1550 /*
1709 * We take our own ref against the journal_head here to avoid 1551 * We take our own ref against the journal_head here to avoid
1710 * having to add tons of locking around each instance of 1552 * having to add tons of locking around each instance of
1711 * jbd2_journal_remove_journal_head() and jbd2_journal_put_journal_head(). 1553 * jbd2_journal_remove_journal_head() and
1554 * jbd2_journal_put_journal_head().
1712 */ 1555 */
1713 jh = jbd2_journal_grab_journal_head(bh); 1556 jh = jbd2_journal_grab_journal_head(bh);
1714 if (!jh) 1557 if (!jh)
@@ -1721,7 +1564,28 @@ int jbd2_journal_try_to_free_buffers(journal_t *journal,
1721 if (buffer_jbd(bh)) 1564 if (buffer_jbd(bh))
1722 goto busy; 1565 goto busy;
1723 } while ((bh = bh->b_this_page) != head); 1566 } while ((bh = bh->b_this_page) != head);
1567
1724 ret = try_to_free_buffers(page); 1568 ret = try_to_free_buffers(page);
1569
1570 /*
1571 * There are a number of places where jbd2_journal_try_to_free_buffers()
1572 * could race with jbd2_journal_commit_transaction(), the later still
1573 * holds the reference to the buffers to free while processing them.
1574 * try_to_free_buffers() failed to free those buffers. Some of the
1575 * caller of releasepage() request page buffers to be dropped, otherwise
1576 * treat the fail-to-free as errors (such as generic_file_direct_IO())
1577 *
1578 * So, if the caller of try_to_release_page() wants the synchronous
1579 * behaviour(i.e make sure buffers are dropped upon return),
1580 * let's wait for the current transaction to finish flush of
1581 * dirty data buffers, then try to free those buffers again,
1582 * with the journal locked.
1583 */
1584 if (ret == 0 && (gfp_mask & __GFP_WAIT) && (gfp_mask & __GFP_FS)) {
1585 jbd2_journal_wait_for_transaction_sync_data(journal);
1586 ret = try_to_free_buffers(page);
1587 }
1588
1725busy: 1589busy:
1726 return ret; 1590 return ret;
1727} 1591}
@@ -1823,6 +1687,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
1823 if (!buffer_jbd(bh)) 1687 if (!buffer_jbd(bh))
1824 goto zap_buffer_unlocked; 1688 goto zap_buffer_unlocked;
1825 1689
1690 /* OK, we have data buffer in journaled mode */
1826 spin_lock(&journal->j_state_lock); 1691 spin_lock(&journal->j_state_lock);
1827 jbd_lock_bh_state(bh); 1692 jbd_lock_bh_state(bh);
1828 spin_lock(&journal->j_list_lock); 1693 spin_lock(&journal->j_list_lock);
@@ -1886,15 +1751,6 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
1886 } 1751 }
1887 } else if (transaction == journal->j_committing_transaction) { 1752 } else if (transaction == journal->j_committing_transaction) {
1888 JBUFFER_TRACE(jh, "on committing transaction"); 1753 JBUFFER_TRACE(jh, "on committing transaction");
1889 if (jh->b_jlist == BJ_Locked) {
1890 /*
1891 * The buffer is on the committing transaction's locked
1892 * list. We have the buffer locked, so I/O has
1893 * completed. So we can nail the buffer now.
1894 */
1895 may_free = __dispose_buffer(jh, transaction);
1896 goto zap_buffer;
1897 }
1898 /* 1754 /*
1899 * If it is committing, we simply cannot touch it. We 1755 * If it is committing, we simply cannot touch it. We
1900 * can remove it's next_transaction pointer from the 1756 * can remove it's next_transaction pointer from the
@@ -2027,9 +1883,6 @@ void __jbd2_journal_file_buffer(struct journal_head *jh,
2027 J_ASSERT_JH(jh, !jh->b_committed_data); 1883 J_ASSERT_JH(jh, !jh->b_committed_data);
2028 J_ASSERT_JH(jh, !jh->b_frozen_data); 1884 J_ASSERT_JH(jh, !jh->b_frozen_data);
2029 return; 1885 return;
2030 case BJ_SyncData:
2031 list = &transaction->t_sync_datalist;
2032 break;
2033 case BJ_Metadata: 1886 case BJ_Metadata:
2034 transaction->t_nr_buffers++; 1887 transaction->t_nr_buffers++;
2035 list = &transaction->t_buffers; 1888 list = &transaction->t_buffers;
@@ -2049,9 +1902,6 @@ void __jbd2_journal_file_buffer(struct journal_head *jh,
2049 case BJ_Reserved: 1902 case BJ_Reserved:
2050 list = &transaction->t_reserved_list; 1903 list = &transaction->t_reserved_list;
2051 break; 1904 break;
2052 case BJ_Locked:
2053 list = &transaction->t_locked_list;
2054 break;
2055 } 1905 }
2056 1906
2057 __blist_add_buffer(list, jh); 1907 __blist_add_buffer(list, jh);
@@ -2141,3 +1991,88 @@ void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2141 spin_unlock(&journal->j_list_lock); 1991 spin_unlock(&journal->j_list_lock);
2142 __brelse(bh); 1992 __brelse(bh);
2143} 1993}
1994
1995/*
1996 * File inode in the inode list of the handle's transaction
1997 */
1998int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
1999{
2000 transaction_t *transaction = handle->h_transaction;
2001 journal_t *journal = transaction->t_journal;
2002
2003 if (is_handle_aborted(handle))
2004 return -EIO;
2005
2006 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2007 transaction->t_tid);
2008
2009 /*
2010 * First check whether inode isn't already on the transaction's
2011 * lists without taking the lock. Note that this check is safe
2012 * without the lock as we cannot race with somebody removing inode
2013 * from the transaction. The reason is that we remove inode from the
2014 * transaction only in journal_release_jbd_inode() and when we commit
2015 * the transaction. We are guarded from the first case by holding
2016 * a reference to the inode. We are safe against the second case
2017 * because if jinode->i_transaction == transaction, commit code
2018 * cannot touch the transaction because we hold reference to it,
2019 * and if jinode->i_next_transaction == transaction, commit code
2020 * will only file the inode where we want it.
2021 */
2022 if (jinode->i_transaction == transaction ||
2023 jinode->i_next_transaction == transaction)
2024 return 0;
2025
2026 spin_lock(&journal->j_list_lock);
2027
2028 if (jinode->i_transaction == transaction ||
2029 jinode->i_next_transaction == transaction)
2030 goto done;
2031
2032 /* On some different transaction's list - should be
2033 * the committing one */
2034 if (jinode->i_transaction) {
2035 J_ASSERT(jinode->i_next_transaction == NULL);
2036 J_ASSERT(jinode->i_transaction ==
2037 journal->j_committing_transaction);
2038 jinode->i_next_transaction = transaction;
2039 goto done;
2040 }
2041 /* Not on any transaction list... */
2042 J_ASSERT(!jinode->i_next_transaction);
2043 jinode->i_transaction = transaction;
2044 list_add(&jinode->i_list, &transaction->t_inode_list);
2045done:
2046 spin_unlock(&journal->j_list_lock);
2047
2048 return 0;
2049}
2050
2051/*
2052 * This function must be called when inode is journaled in ordered mode
2053 * before truncation happens. It starts writeout of truncated part in
2054 * case it is in the committing transaction so that we stand to ordered
2055 * mode consistency guarantees.
2056 */
2057int jbd2_journal_begin_ordered_truncate(struct jbd2_inode *inode,
2058 loff_t new_size)
2059{
2060 journal_t *journal;
2061 transaction_t *commit_trans;
2062 int ret = 0;
2063
2064 if (!inode->i_transaction && !inode->i_next_transaction)
2065 goto out;
2066 journal = inode->i_transaction->t_journal;
2067 spin_lock(&journal->j_state_lock);
2068 commit_trans = journal->j_committing_transaction;
2069 spin_unlock(&journal->j_state_lock);
2070 if (inode->i_transaction == commit_trans) {
2071 ret = filemap_fdatawrite_range(inode->i_vfs_inode->i_mapping,
2072 new_size, LLONG_MAX);
2073 if (ret)
2074 jbd2_journal_abort(journal, ret);
2075 }
2076out:
2077 return ret;
2078}
diff --git a/fs/mpage.c b/fs/mpage.c
index 235e4d3873a8..dbcc7af76a15 100644
--- a/fs/mpage.c
+++ b/fs/mpage.c
@@ -82,7 +82,7 @@ static void mpage_end_io_write(struct bio *bio, int err)
82 bio_put(bio); 82 bio_put(bio);
83} 83}
84 84
85static struct bio *mpage_bio_submit(int rw, struct bio *bio) 85struct bio *mpage_bio_submit(int rw, struct bio *bio)
86{ 86{
87 bio->bi_end_io = mpage_end_io_read; 87 bio->bi_end_io = mpage_end_io_read;
88 if (rw == WRITE) 88 if (rw == WRITE)
@@ -90,6 +90,7 @@ static struct bio *mpage_bio_submit(int rw, struct bio *bio)
90 submit_bio(rw, bio); 90 submit_bio(rw, bio);
91 return NULL; 91 return NULL;
92} 92}
93EXPORT_SYMBOL(mpage_bio_submit);
93 94
94static struct bio * 95static struct bio *
95mpage_alloc(struct block_device *bdev, 96mpage_alloc(struct block_device *bdev,
@@ -435,15 +436,9 @@ EXPORT_SYMBOL(mpage_readpage);
435 * written, so it can intelligently allocate a suitably-sized BIO. For now, 436 * written, so it can intelligently allocate a suitably-sized BIO. For now,
436 * just allocate full-size (16-page) BIOs. 437 * just allocate full-size (16-page) BIOs.
437 */ 438 */
438struct mpage_data {
439 struct bio *bio;
440 sector_t last_block_in_bio;
441 get_block_t *get_block;
442 unsigned use_writepage;
443};
444 439
445static int __mpage_writepage(struct page *page, struct writeback_control *wbc, 440int __mpage_writepage(struct page *page, struct writeback_control *wbc,
446 void *data) 441 void *data)
447{ 442{
448 struct mpage_data *mpd = data; 443 struct mpage_data *mpd = data;
449 struct bio *bio = mpd->bio; 444 struct bio *bio = mpd->bio;
@@ -651,6 +646,7 @@ out:
651 mpd->bio = bio; 646 mpd->bio = bio;
652 return ret; 647 return ret;
653} 648}
649EXPORT_SYMBOL(__mpage_writepage);
654 650
655/** 651/**
656 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them 652 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
diff --git a/include/linux/fs.h b/include/linux/fs.h
index faac13e2cc5c..52e510a0aec2 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1740,6 +1740,8 @@ extern int wait_on_page_writeback_range(struct address_space *mapping,
1740 pgoff_t start, pgoff_t end); 1740 pgoff_t start, pgoff_t end);
1741extern int __filemap_fdatawrite_range(struct address_space *mapping, 1741extern int __filemap_fdatawrite_range(struct address_space *mapping,
1742 loff_t start, loff_t end, int sync_mode); 1742 loff_t start, loff_t end, int sync_mode);
1743extern int filemap_fdatawrite_range(struct address_space *mapping,
1744 loff_t start, loff_t end);
1743 1745
1744extern long do_fsync(struct file *file, int datasync); 1746extern long do_fsync(struct file *file, int datasync);
1745extern void sync_supers(void); 1747extern void sync_supers(void);
diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
index d147f0f90360..3dd209007098 100644
--- a/include/linux/jbd2.h
+++ b/include/linux/jbd2.h
@@ -168,6 +168,8 @@ struct commit_header {
168 unsigned char h_chksum_size; 168 unsigned char h_chksum_size;
169 unsigned char h_padding[2]; 169 unsigned char h_padding[2];
170 __be32 h_chksum[JBD2_CHECKSUM_BYTES]; 170 __be32 h_chksum[JBD2_CHECKSUM_BYTES];
171 __be64 h_commit_sec;
172 __be32 h_commit_nsec;
171}; 173};
172 174
173/* 175/*
@@ -379,6 +381,38 @@ static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
379 bit_spin_unlock(BH_JournalHead, &bh->b_state); 381 bit_spin_unlock(BH_JournalHead, &bh->b_state);
380} 382}
381 383
384/* Flags in jbd_inode->i_flags */
385#define __JI_COMMIT_RUNNING 0
386/* Commit of the inode data in progress. We use this flag to protect us from
387 * concurrent deletion of inode. We cannot use reference to inode for this
388 * since we cannot afford doing last iput() on behalf of kjournald
389 */
390#define JI_COMMIT_RUNNING (1 << __JI_COMMIT_RUNNING)
391
392/**
393 * struct jbd_inode is the structure linking inodes in ordered mode
394 * present in a transaction so that we can sync them during commit.
395 */
396struct jbd2_inode {
397 /* Which transaction does this inode belong to? Either the running
398 * transaction or the committing one. [j_list_lock] */
399 transaction_t *i_transaction;
400
401 /* Pointer to the running transaction modifying inode's data in case
402 * there is already a committing transaction touching it. [j_list_lock] */
403 transaction_t *i_next_transaction;
404
405 /* List of inodes in the i_transaction [j_list_lock] */
406 struct list_head i_list;
407
408 /* VFS inode this inode belongs to [constant during the lifetime
409 * of the structure] */
410 struct inode *i_vfs_inode;
411
412 /* Flags of inode [j_list_lock] */
413 unsigned int i_flags;
414};
415
382struct jbd2_revoke_table_s; 416struct jbd2_revoke_table_s;
383 417
384/** 418/**
@@ -509,24 +543,12 @@ struct transaction_s
509 struct journal_head *t_reserved_list; 543 struct journal_head *t_reserved_list;
510 544
511 /* 545 /*
512 * Doubly-linked circular list of all buffers under writeout during
513 * commit [j_list_lock]
514 */
515 struct journal_head *t_locked_list;
516
517 /*
518 * Doubly-linked circular list of all metadata buffers owned by this 546 * Doubly-linked circular list of all metadata buffers owned by this
519 * transaction [j_list_lock] 547 * transaction [j_list_lock]
520 */ 548 */
521 struct journal_head *t_buffers; 549 struct journal_head *t_buffers;
522 550
523 /* 551 /*
524 * Doubly-linked circular list of all data buffers still to be
525 * flushed before this transaction can be committed [j_list_lock]
526 */
527 struct journal_head *t_sync_datalist;
528
529 /*
530 * Doubly-linked circular list of all forget buffers (superseded 552 * Doubly-linked circular list of all forget buffers (superseded
531 * buffers which we can un-checkpoint once this transaction commits) 553 * buffers which we can un-checkpoint once this transaction commits)
532 * [j_list_lock] 554 * [j_list_lock]
@@ -565,6 +587,12 @@ struct transaction_s
565 struct journal_head *t_log_list; 587 struct journal_head *t_log_list;
566 588
567 /* 589 /*
590 * List of inodes whose data we've modified in data=ordered mode.
591 * [j_list_lock]
592 */
593 struct list_head t_inode_list;
594
595 /*
568 * Protects info related to handles 596 * Protects info related to handles
569 */ 597 */
570 spinlock_t t_handle_lock; 598 spinlock_t t_handle_lock;
@@ -1004,7 +1032,6 @@ extern int jbd2_journal_extend (handle_t *, int nblocks);
1004extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *); 1032extern int jbd2_journal_get_write_access(handle_t *, struct buffer_head *);
1005extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *); 1033extern int jbd2_journal_get_create_access (handle_t *, struct buffer_head *);
1006extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *); 1034extern int jbd2_journal_get_undo_access(handle_t *, struct buffer_head *);
1007extern int jbd2_journal_dirty_data (handle_t *, struct buffer_head *);
1008extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *); 1035extern int jbd2_journal_dirty_metadata (handle_t *, struct buffer_head *);
1009extern void jbd2_journal_release_buffer (handle_t *, struct buffer_head *); 1036extern void jbd2_journal_release_buffer (handle_t *, struct buffer_head *);
1010extern int jbd2_journal_forget (handle_t *, struct buffer_head *); 1037extern int jbd2_journal_forget (handle_t *, struct buffer_head *);
@@ -1044,6 +1071,10 @@ extern void jbd2_journal_ack_err (journal_t *);
1044extern int jbd2_journal_clear_err (journal_t *); 1071extern int jbd2_journal_clear_err (journal_t *);
1045extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *); 1072extern int jbd2_journal_bmap(journal_t *, unsigned long, unsigned long long *);
1046extern int jbd2_journal_force_commit(journal_t *); 1073extern int jbd2_journal_force_commit(journal_t *);
1074extern int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *inode);
1075extern int jbd2_journal_begin_ordered_truncate(struct jbd2_inode *inode, loff_t new_size);
1076extern void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode);
1077extern void jbd2_journal_release_jbd_inode(journal_t *journal, struct jbd2_inode *jinode);
1047 1078
1048/* 1079/*
1049 * journal_head management 1080 * journal_head management
@@ -1179,15 +1210,13 @@ static inline int jbd_space_needed(journal_t *journal)
1179 1210
1180/* journaling buffer types */ 1211/* journaling buffer types */
1181#define BJ_None 0 /* Not journaled */ 1212#define BJ_None 0 /* Not journaled */
1182#define BJ_SyncData 1 /* Normal data: flush before commit */ 1213#define BJ_Metadata 1 /* Normal journaled metadata */
1183#define BJ_Metadata 2 /* Normal journaled metadata */ 1214#define BJ_Forget 2 /* Buffer superseded by this transaction */
1184#define BJ_Forget 3 /* Buffer superseded by this transaction */ 1215#define BJ_IO 3 /* Buffer is for temporary IO use */
1185#define BJ_IO 4 /* Buffer is for temporary IO use */ 1216#define BJ_Shadow 4 /* Buffer contents being shadowed to the log */
1186#define BJ_Shadow 5 /* Buffer contents being shadowed to the log */ 1217#define BJ_LogCtl 5 /* Buffer contains log descriptors */
1187#define BJ_LogCtl 6 /* Buffer contains log descriptors */ 1218#define BJ_Reserved 6 /* Buffer is reserved for access by journal */
1188#define BJ_Reserved 7 /* Buffer is reserved for access by journal */ 1219#define BJ_Types 7
1189#define BJ_Locked 8 /* Locked for I/O during commit */
1190#define BJ_Types 9
1191 1220
1192extern int jbd_blocks_per_page(struct inode *inode); 1221extern int jbd_blocks_per_page(struct inode *inode);
1193 1222
diff --git a/include/linux/mpage.h b/include/linux/mpage.h
index 068a0c9946af..5c42821da2d1 100644
--- a/include/linux/mpage.h
+++ b/include/linux/mpage.h
@@ -11,11 +11,21 @@
11 */ 11 */
12#ifdef CONFIG_BLOCK 12#ifdef CONFIG_BLOCK
13 13
14struct mpage_data {
15 struct bio *bio;
16 sector_t last_block_in_bio;
17 get_block_t *get_block;
18 unsigned use_writepage;
19};
20
14struct writeback_control; 21struct writeback_control;
15 22
23struct bio *mpage_bio_submit(int rw, struct bio *bio);
16int mpage_readpages(struct address_space *mapping, struct list_head *pages, 24int mpage_readpages(struct address_space *mapping, struct list_head *pages,
17 unsigned nr_pages, get_block_t get_block); 25 unsigned nr_pages, get_block_t get_block);
18int mpage_readpage(struct page *page, get_block_t get_block); 26int mpage_readpage(struct page *page, get_block_t get_block);
27int __mpage_writepage(struct page *page, struct writeback_control *wbc,
28 void *data);
19int mpage_writepages(struct address_space *mapping, 29int mpage_writepages(struct address_space *mapping,
20 struct writeback_control *wbc, get_block_t get_block); 30 struct writeback_control *wbc, get_block_t get_block);
21int mpage_writepage(struct page *page, get_block_t *get_block, 31int mpage_writepage(struct page *page, get_block_t *get_block,
diff --git a/include/linux/percpu_counter.h b/include/linux/percpu_counter.h
index 9007ccdfc112..208388835357 100644
--- a/include/linux/percpu_counter.h
+++ b/include/linux/percpu_counter.h
@@ -35,7 +35,7 @@ int percpu_counter_init_irq(struct percpu_counter *fbc, s64 amount);
35void percpu_counter_destroy(struct percpu_counter *fbc); 35void percpu_counter_destroy(struct percpu_counter *fbc);
36void percpu_counter_set(struct percpu_counter *fbc, s64 amount); 36void percpu_counter_set(struct percpu_counter *fbc, s64 amount);
37void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch); 37void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch);
38s64 __percpu_counter_sum(struct percpu_counter *fbc); 38s64 __percpu_counter_sum(struct percpu_counter *fbc, int set);
39 39
40static inline void percpu_counter_add(struct percpu_counter *fbc, s64 amount) 40static inline void percpu_counter_add(struct percpu_counter *fbc, s64 amount)
41{ 41{
@@ -44,13 +44,19 @@ static inline void percpu_counter_add(struct percpu_counter *fbc, s64 amount)
44 44
45static inline s64 percpu_counter_sum_positive(struct percpu_counter *fbc) 45static inline s64 percpu_counter_sum_positive(struct percpu_counter *fbc)
46{ 46{
47 s64 ret = __percpu_counter_sum(fbc); 47 s64 ret = __percpu_counter_sum(fbc, 0);
48 return ret < 0 ? 0 : ret; 48 return ret < 0 ? 0 : ret;
49} 49}
50 50
51static inline s64 percpu_counter_sum_and_set(struct percpu_counter *fbc)
52{
53 return __percpu_counter_sum(fbc, 1);
54}
55
56
51static inline s64 percpu_counter_sum(struct percpu_counter *fbc) 57static inline s64 percpu_counter_sum(struct percpu_counter *fbc)
52{ 58{
53 return __percpu_counter_sum(fbc); 59 return __percpu_counter_sum(fbc, 0);
54} 60}
55 61
56static inline s64 percpu_counter_read(struct percpu_counter *fbc) 62static inline s64 percpu_counter_read(struct percpu_counter *fbc)
diff --git a/include/linux/writeback.h b/include/linux/writeback.h
index bd91987c065f..12b15c561a1f 100644
--- a/include/linux/writeback.h
+++ b/include/linux/writeback.h
@@ -63,6 +63,7 @@ struct writeback_control {
63 unsigned for_writepages:1; /* This is a writepages() call */ 63 unsigned for_writepages:1; /* This is a writepages() call */
64 unsigned range_cyclic:1; /* range_start is cyclic */ 64 unsigned range_cyclic:1; /* range_start is cyclic */
65 unsigned more_io:1; /* more io to be dispatched */ 65 unsigned more_io:1; /* more io to be dispatched */
66 unsigned range_cont:1;
66}; 67};
67 68
68/* 69/*
diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c
index 119174494cb5..4a8ba4bf5f6f 100644
--- a/lib/percpu_counter.c
+++ b/lib/percpu_counter.c
@@ -52,7 +52,7 @@ EXPORT_SYMBOL(__percpu_counter_add);
52 * Add up all the per-cpu counts, return the result. This is a more accurate 52 * Add up all the per-cpu counts, return the result. This is a more accurate
53 * but much slower version of percpu_counter_read_positive() 53 * but much slower version of percpu_counter_read_positive()
54 */ 54 */
55s64 __percpu_counter_sum(struct percpu_counter *fbc) 55s64 __percpu_counter_sum(struct percpu_counter *fbc, int set)
56{ 56{
57 s64 ret; 57 s64 ret;
58 int cpu; 58 int cpu;
@@ -62,7 +62,12 @@ s64 __percpu_counter_sum(struct percpu_counter *fbc)
62 for_each_online_cpu(cpu) { 62 for_each_online_cpu(cpu) {
63 s32 *pcount = per_cpu_ptr(fbc->counters, cpu); 63 s32 *pcount = per_cpu_ptr(fbc->counters, cpu);
64 ret += *pcount; 64 ret += *pcount;
65 if (set)
66 *pcount = 0;
65 } 67 }
68 if (set)
69 fbc->count = ret;
70
66 spin_unlock(&fbc->lock); 71 spin_unlock(&fbc->lock);
67 return ret; 72 return ret;
68} 73}
diff --git a/mm/filemap.c b/mm/filemap.c
index 1e6a7d34874f..65d9d9e2b755 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -236,11 +236,12 @@ int filemap_fdatawrite(struct address_space *mapping)
236} 236}
237EXPORT_SYMBOL(filemap_fdatawrite); 237EXPORT_SYMBOL(filemap_fdatawrite);
238 238
239static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 239int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
240 loff_t end) 240 loff_t end)
241{ 241{
242 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 242 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
243} 243}
244EXPORT_SYMBOL(filemap_fdatawrite_range);
244 245
245/** 246/**
246 * filemap_flush - mostly a non-blocking flush 247 * filemap_flush - mostly a non-blocking flush
diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index b38f700825fc..94c6d8988ab3 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -960,6 +960,9 @@ retry:
960 } 960 }
961 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) 961 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
962 mapping->writeback_index = index; 962 mapping->writeback_index = index;
963
964 if (wbc->range_cont)
965 wbc->range_start = index << PAGE_CACHE_SHIFT;
963 return ret; 966 return ret;
964} 967}
965EXPORT_SYMBOL(write_cache_pages); 968EXPORT_SYMBOL(write_cache_pages);