aboutsummaryrefslogtreecommitdiffstats
path: root/fs/ext4/inode.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/ext4/inode.c')
-rw-r--r--fs/ext4/inode.c309
1 files changed, 227 insertions, 82 deletions
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 98d3fe7057ef..a6444cee0c7e 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -72,12 +72,17 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
72 * "bh" may be NULL: a metadata block may have been freed from memory 72 * "bh" may be NULL: a metadata block may have been freed from memory
73 * but there may still be a record of it in the journal, and that record 73 * but there may still be a record of it in the journal, and that record
74 * still needs to be revoked. 74 * still needs to be revoked.
75 *
76 * If the handle isn't valid we're not journaling so there's nothing to do.
75 */ 77 */
76int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode, 78int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
77 struct buffer_head *bh, ext4_fsblk_t blocknr) 79 struct buffer_head *bh, ext4_fsblk_t blocknr)
78{ 80{
79 int err; 81 int err;
80 82
83 if (!ext4_handle_valid(handle))
84 return 0;
85
81 might_sleep(); 86 might_sleep();
82 87
83 BUFFER_TRACE(bh, "enter"); 88 BUFFER_TRACE(bh, "enter");
@@ -170,7 +175,9 @@ static handle_t *start_transaction(struct inode *inode)
170 */ 175 */
171static int try_to_extend_transaction(handle_t *handle, struct inode *inode) 176static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
172{ 177{
173 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS) 178 if (!ext4_handle_valid(handle))
179 return 0;
180 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
174 return 0; 181 return 0;
175 if (!ext4_journal_extend(handle, blocks_for_truncate(inode))) 182 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
176 return 0; 183 return 0;
@@ -184,6 +191,7 @@ static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
184 */ 191 */
185static int ext4_journal_test_restart(handle_t *handle, struct inode *inode) 192static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
186{ 193{
194 BUG_ON(EXT4_JOURNAL(inode) == NULL);
187 jbd_debug(2, "restarting handle %p\n", handle); 195 jbd_debug(2, "restarting handle %p\n", handle);
188 return ext4_journal_restart(handle, blocks_for_truncate(inode)); 196 return ext4_journal_restart(handle, blocks_for_truncate(inode));
189} 197}
@@ -216,7 +224,7 @@ void ext4_delete_inode(struct inode *inode)
216 } 224 }
217 225
218 if (IS_SYNC(inode)) 226 if (IS_SYNC(inode))
219 handle->h_sync = 1; 227 ext4_handle_sync(handle);
220 inode->i_size = 0; 228 inode->i_size = 0;
221 err = ext4_mark_inode_dirty(handle, inode); 229 err = ext4_mark_inode_dirty(handle, inode);
222 if (err) { 230 if (err) {
@@ -233,7 +241,7 @@ void ext4_delete_inode(struct inode *inode)
233 * enough credits left in the handle to remove the inode from 241 * enough credits left in the handle to remove the inode from
234 * the orphan list and set the dtime field. 242 * the orphan list and set the dtime field.
235 */ 243 */
236 if (handle->h_buffer_credits < 3) { 244 if (!ext4_handle_has_enough_credits(handle, 3)) {
237 err = ext4_journal_extend(handle, 3); 245 err = ext4_journal_extend(handle, 3);
238 if (err > 0) 246 if (err > 0)
239 err = ext4_journal_restart(handle, 3); 247 err = ext4_journal_restart(handle, 3);
@@ -506,10 +514,10 @@ static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
506 * return the total number of blocks to be allocate, including the 514 * return the total number of blocks to be allocate, including the
507 * direct and indirect blocks. 515 * direct and indirect blocks.
508 */ 516 */
509static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks, 517static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
510 int blocks_to_boundary) 518 int blocks_to_boundary)
511{ 519{
512 unsigned long count = 0; 520 unsigned int count = 0;
513 521
514 /* 522 /*
515 * Simple case, [t,d]Indirect block(s) has not allocated yet 523 * Simple case, [t,d]Indirect block(s) has not allocated yet
@@ -547,6 +555,7 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
547 int indirect_blks, int blks, 555 int indirect_blks, int blks,
548 ext4_fsblk_t new_blocks[4], int *err) 556 ext4_fsblk_t new_blocks[4], int *err)
549{ 557{
558 struct ext4_allocation_request ar;
550 int target, i; 559 int target, i;
551 unsigned long count = 0, blk_allocated = 0; 560 unsigned long count = 0, blk_allocated = 0;
552 int index = 0; 561 int index = 0;
@@ -595,10 +604,17 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
595 if (!target) 604 if (!target)
596 goto allocated; 605 goto allocated;
597 /* Now allocate data blocks */ 606 /* Now allocate data blocks */
598 count = target; 607 memset(&ar, 0, sizeof(ar));
599 /* allocating blocks for data blocks */ 608 ar.inode = inode;
600 current_block = ext4_new_blocks(handle, inode, iblock, 609 ar.goal = goal;
601 goal, &count, err); 610 ar.len = target;
611 ar.logical = iblock;
612 if (S_ISREG(inode->i_mode))
613 /* enable in-core preallocation only for regular files */
614 ar.flags = EXT4_MB_HINT_DATA;
615
616 current_block = ext4_mb_new_blocks(handle, &ar, err);
617
602 if (*err && (target == blks)) { 618 if (*err && (target == blks)) {
603 /* 619 /*
604 * if the allocation failed and we didn't allocate 620 * if the allocation failed and we didn't allocate
@@ -614,7 +630,7 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
614 */ 630 */
615 new_blocks[index] = current_block; 631 new_blocks[index] = current_block;
616 } 632 }
617 blk_allocated += count; 633 blk_allocated += ar.len;
618 } 634 }
619allocated: 635allocated:
620 /* total number of blocks allocated for direct blocks */ 636 /* total number of blocks allocated for direct blocks */
@@ -709,8 +725,8 @@ static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
709 set_buffer_uptodate(bh); 725 set_buffer_uptodate(bh);
710 unlock_buffer(bh); 726 unlock_buffer(bh);
711 727
712 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); 728 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
713 err = ext4_journal_dirty_metadata(handle, bh); 729 err = ext4_handle_dirty_metadata(handle, inode, bh);
714 if (err) 730 if (err)
715 goto failed; 731 goto failed;
716 } 732 }
@@ -792,8 +808,8 @@ static int ext4_splice_branch(handle_t *handle, struct inode *inode,
792 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode. 808 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
793 */ 809 */
794 jbd_debug(5, "splicing indirect only\n"); 810 jbd_debug(5, "splicing indirect only\n");
795 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata"); 811 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
796 err = ext4_journal_dirty_metadata(handle, where->bh); 812 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
797 if (err) 813 if (err)
798 goto err_out; 814 goto err_out;
799 } else { 815 } else {
@@ -840,10 +856,10 @@ err_out:
840 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block 856 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
841 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem) 857 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
842 */ 858 */
843int ext4_get_blocks_handle(handle_t *handle, struct inode *inode, 859static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
844 ext4_lblk_t iblock, unsigned long maxblocks, 860 ext4_lblk_t iblock, unsigned int maxblocks,
845 struct buffer_head *bh_result, 861 struct buffer_head *bh_result,
846 int create, int extend_disksize) 862 int create, int extend_disksize)
847{ 863{
848 int err = -EIO; 864 int err = -EIO;
849 ext4_lblk_t offsets[4]; 865 ext4_lblk_t offsets[4];
@@ -1045,7 +1061,7 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
1045 * It returns the error in case of allocation failure. 1061 * It returns the error in case of allocation failure.
1046 */ 1062 */
1047int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block, 1063int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1048 unsigned long max_blocks, struct buffer_head *bh, 1064 unsigned int max_blocks, struct buffer_head *bh,
1049 int create, int extend_disksize, int flag) 1065 int create, int extend_disksize, int flag)
1050{ 1066{
1051 int retval; 1067 int retval;
@@ -1221,8 +1237,8 @@ struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1221 set_buffer_uptodate(bh); 1237 set_buffer_uptodate(bh);
1222 } 1238 }
1223 unlock_buffer(bh); 1239 unlock_buffer(bh);
1224 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); 1240 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1225 err = ext4_journal_dirty_metadata(handle, bh); 1241 err = ext4_handle_dirty_metadata(handle, inode, bh);
1226 if (!fatal) 1242 if (!fatal)
1227 fatal = err; 1243 fatal = err;
1228 } else { 1244 } else {
@@ -1335,6 +1351,10 @@ static int ext4_write_begin(struct file *file, struct address_space *mapping,
1335 pgoff_t index; 1351 pgoff_t index;
1336 unsigned from, to; 1352 unsigned from, to;
1337 1353
1354 trace_mark(ext4_write_begin,
1355 "dev %s ino %lu pos %llu len %u flags %u",
1356 inode->i_sb->s_id, inode->i_ino,
1357 (unsigned long long) pos, len, flags);
1338 index = pos >> PAGE_CACHE_SHIFT; 1358 index = pos >> PAGE_CACHE_SHIFT;
1339 from = pos & (PAGE_CACHE_SIZE - 1); 1359 from = pos & (PAGE_CACHE_SIZE - 1);
1340 to = from + len; 1360 to = from + len;
@@ -1387,7 +1407,7 @@ static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1387 if (!buffer_mapped(bh) || buffer_freed(bh)) 1407 if (!buffer_mapped(bh) || buffer_freed(bh))
1388 return 0; 1408 return 0;
1389 set_buffer_uptodate(bh); 1409 set_buffer_uptodate(bh);
1390 return ext4_journal_dirty_metadata(handle, bh); 1410 return ext4_handle_dirty_metadata(handle, NULL, bh);
1391} 1411}
1392 1412
1393/* 1413/*
@@ -1406,6 +1426,10 @@ static int ext4_ordered_write_end(struct file *file,
1406 struct inode *inode = mapping->host; 1426 struct inode *inode = mapping->host;
1407 int ret = 0, ret2; 1427 int ret = 0, ret2;
1408 1428
1429 trace_mark(ext4_ordered_write_end,
1430 "dev %s ino %lu pos %llu len %u copied %u",
1431 inode->i_sb->s_id, inode->i_ino,
1432 (unsigned long long) pos, len, copied);
1409 ret = ext4_jbd2_file_inode(handle, inode); 1433 ret = ext4_jbd2_file_inode(handle, inode);
1410 1434
1411 if (ret == 0) { 1435 if (ret == 0) {
@@ -1444,6 +1468,10 @@ static int ext4_writeback_write_end(struct file *file,
1444 int ret = 0, ret2; 1468 int ret = 0, ret2;
1445 loff_t new_i_size; 1469 loff_t new_i_size;
1446 1470
1471 trace_mark(ext4_writeback_write_end,
1472 "dev %s ino %lu pos %llu len %u copied %u",
1473 inode->i_sb->s_id, inode->i_ino,
1474 (unsigned long long) pos, len, copied);
1447 new_i_size = pos + copied; 1475 new_i_size = pos + copied;
1448 if (new_i_size > EXT4_I(inode)->i_disksize) { 1476 if (new_i_size > EXT4_I(inode)->i_disksize) {
1449 ext4_update_i_disksize(inode, new_i_size); 1477 ext4_update_i_disksize(inode, new_i_size);
@@ -1479,6 +1507,10 @@ static int ext4_journalled_write_end(struct file *file,
1479 unsigned from, to; 1507 unsigned from, to;
1480 loff_t new_i_size; 1508 loff_t new_i_size;
1481 1509
1510 trace_mark(ext4_journalled_write_end,
1511 "dev %s ino %lu pos %llu len %u copied %u",
1512 inode->i_sb->s_id, inode->i_ino,
1513 (unsigned long long) pos, len, copied);
1482 from = pos & (PAGE_CACHE_SIZE - 1); 1514 from = pos & (PAGE_CACHE_SIZE - 1);
1483 to = from + len; 1515 to = from + len;
1484 1516
@@ -1625,7 +1657,7 @@ struct mpage_da_data {
1625 get_block_t *get_block; 1657 get_block_t *get_block;
1626 struct writeback_control *wbc; 1658 struct writeback_control *wbc;
1627 int io_done; 1659 int io_done;
1628 long pages_written; 1660 int pages_written;
1629 int retval; 1661 int retval;
1630}; 1662};
1631 1663
@@ -1645,35 +1677,39 @@ struct mpage_da_data {
1645 */ 1677 */
1646static int mpage_da_submit_io(struct mpage_da_data *mpd) 1678static int mpage_da_submit_io(struct mpage_da_data *mpd)
1647{ 1679{
1648 struct address_space *mapping = mpd->inode->i_mapping;
1649 int ret = 0, err, nr_pages, i;
1650 unsigned long index, end;
1651 struct pagevec pvec;
1652 long pages_skipped; 1680 long pages_skipped;
1681 struct pagevec pvec;
1682 unsigned long index, end;
1683 int ret = 0, err, nr_pages, i;
1684 struct inode *inode = mpd->inode;
1685 struct address_space *mapping = inode->i_mapping;
1653 1686
1654 BUG_ON(mpd->next_page <= mpd->first_page); 1687 BUG_ON(mpd->next_page <= mpd->first_page);
1655 pagevec_init(&pvec, 0); 1688 /*
1689 * We need to start from the first_page to the next_page - 1
1690 * to make sure we also write the mapped dirty buffer_heads.
1691 * If we look at mpd->lbh.b_blocknr we would only be looking
1692 * at the currently mapped buffer_heads.
1693 */
1656 index = mpd->first_page; 1694 index = mpd->first_page;
1657 end = mpd->next_page - 1; 1695 end = mpd->next_page - 1;
1658 1696
1697 pagevec_init(&pvec, 0);
1659 while (index <= end) { 1698 while (index <= end) {
1660 /* 1699 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1661 * We can use PAGECACHE_TAG_DIRTY lookup here because
1662 * even though we have cleared the dirty flag on the page
1663 * We still keep the page in the radix tree with tag
1664 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1665 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1666 * which is called via the below writepage callback.
1667 */
1668 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1669 PAGECACHE_TAG_DIRTY,
1670 min(end - index,
1671 (pgoff_t)PAGEVEC_SIZE-1) + 1);
1672 if (nr_pages == 0) 1700 if (nr_pages == 0)
1673 break; 1701 break;
1674 for (i = 0; i < nr_pages; i++) { 1702 for (i = 0; i < nr_pages; i++) {
1675 struct page *page = pvec.pages[i]; 1703 struct page *page = pvec.pages[i];
1676 1704
1705 index = page->index;
1706 if (index > end)
1707 break;
1708 index++;
1709
1710 BUG_ON(!PageLocked(page));
1711 BUG_ON(PageWriteback(page));
1712
1677 pages_skipped = mpd->wbc->pages_skipped; 1713 pages_skipped = mpd->wbc->pages_skipped;
1678 err = mapping->a_ops->writepage(page, mpd->wbc); 1714 err = mapping->a_ops->writepage(page, mpd->wbc);
1679 if (!err && (pages_skipped == mpd->wbc->pages_skipped)) 1715 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
@@ -1831,13 +1867,13 @@ static void ext4_print_free_blocks(struct inode *inode)
1831 ext4_count_free_blocks(inode->i_sb)); 1867 ext4_count_free_blocks(inode->i_sb));
1832 printk(KERN_EMERG "Free/Dirty block details\n"); 1868 printk(KERN_EMERG "Free/Dirty block details\n");
1833 printk(KERN_EMERG "free_blocks=%lld\n", 1869 printk(KERN_EMERG "free_blocks=%lld\n",
1834 percpu_counter_sum(&sbi->s_freeblocks_counter)); 1870 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1835 printk(KERN_EMERG "dirty_blocks=%lld\n", 1871 printk(KERN_EMERG "dirty_blocks=%lld\n",
1836 percpu_counter_sum(&sbi->s_dirtyblocks_counter)); 1872 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1837 printk(KERN_EMERG "Block reservation details\n"); 1873 printk(KERN_EMERG "Block reservation details\n");
1838 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n", 1874 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1839 EXT4_I(inode)->i_reserved_data_blocks); 1875 EXT4_I(inode)->i_reserved_data_blocks);
1840 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n", 1876 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1841 EXT4_I(inode)->i_reserved_meta_blocks); 1877 EXT4_I(inode)->i_reserved_meta_blocks);
1842 return; 1878 return;
1843} 1879}
@@ -2087,11 +2123,29 @@ static int __mpage_da_writepage(struct page *page,
2087 bh = head; 2123 bh = head;
2088 do { 2124 do {
2089 BUG_ON(buffer_locked(bh)); 2125 BUG_ON(buffer_locked(bh));
2126 /*
2127 * We need to try to allocate
2128 * unmapped blocks in the same page.
2129 * Otherwise we won't make progress
2130 * with the page in ext4_da_writepage
2131 */
2090 if (buffer_dirty(bh) && 2132 if (buffer_dirty(bh) &&
2091 (!buffer_mapped(bh) || buffer_delay(bh))) { 2133 (!buffer_mapped(bh) || buffer_delay(bh))) {
2092 mpage_add_bh_to_extent(mpd, logical, bh); 2134 mpage_add_bh_to_extent(mpd, logical, bh);
2093 if (mpd->io_done) 2135 if (mpd->io_done)
2094 return MPAGE_DA_EXTENT_TAIL; 2136 return MPAGE_DA_EXTENT_TAIL;
2137 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2138 /*
2139 * mapped dirty buffer. We need to update
2140 * the b_state because we look at
2141 * b_state in mpage_da_map_blocks. We don't
2142 * update b_size because if we find an
2143 * unmapped buffer_head later we need to
2144 * use the b_state flag of that buffer_head.
2145 */
2146 if (mpd->lbh.b_size == 0)
2147 mpd->lbh.b_state =
2148 bh->b_state & BH_FLAGS;
2095 } 2149 }
2096 logical++; 2150 logical++;
2097 } while ((bh = bh->b_this_page) != head); 2151 } while ((bh = bh->b_this_page) != head);
@@ -2269,10 +2323,13 @@ static int ext4_da_writepage(struct page *page,
2269{ 2323{
2270 int ret = 0; 2324 int ret = 0;
2271 loff_t size; 2325 loff_t size;
2272 unsigned long len; 2326 unsigned int len;
2273 struct buffer_head *page_bufs; 2327 struct buffer_head *page_bufs;
2274 struct inode *inode = page->mapping->host; 2328 struct inode *inode = page->mapping->host;
2275 2329
2330 trace_mark(ext4_da_writepage,
2331 "dev %s ino %lu page_index %lu",
2332 inode->i_sb->s_id, inode->i_ino, page->index);
2276 size = i_size_read(inode); 2333 size = i_size_read(inode);
2277 if (page->index == size >> PAGE_CACHE_SHIFT) 2334 if (page->index == size >> PAGE_CACHE_SHIFT)
2278 len = size & ~PAGE_CACHE_MASK; 2335 len = size & ~PAGE_CACHE_MASK;
@@ -2378,10 +2435,25 @@ static int ext4_da_writepages(struct address_space *mapping,
2378 struct mpage_da_data mpd; 2435 struct mpage_da_data mpd;
2379 struct inode *inode = mapping->host; 2436 struct inode *inode = mapping->host;
2380 int no_nrwrite_index_update; 2437 int no_nrwrite_index_update;
2381 long pages_written = 0, pages_skipped; 2438 int pages_written = 0;
2439 long pages_skipped;
2382 int needed_blocks, ret = 0, nr_to_writebump = 0; 2440 int needed_blocks, ret = 0, nr_to_writebump = 0;
2383 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); 2441 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2384 2442
2443 trace_mark(ext4_da_writepages,
2444 "dev %s ino %lu nr_t_write %ld "
2445 "pages_skipped %ld range_start %llu "
2446 "range_end %llu nonblocking %d "
2447 "for_kupdate %d for_reclaim %d "
2448 "for_writepages %d range_cyclic %d",
2449 inode->i_sb->s_id, inode->i_ino,
2450 wbc->nr_to_write, wbc->pages_skipped,
2451 (unsigned long long) wbc->range_start,
2452 (unsigned long long) wbc->range_end,
2453 wbc->nonblocking, wbc->for_kupdate,
2454 wbc->for_reclaim, wbc->for_writepages,
2455 wbc->range_cyclic);
2456
2385 /* 2457 /*
2386 * No pages to write? This is mainly a kludge to avoid starting 2458 * No pages to write? This is mainly a kludge to avoid starting
2387 * a transaction for special inodes like journal inode on last iput() 2459 * a transaction for special inodes like journal inode on last iput()
@@ -2389,6 +2461,20 @@ static int ext4_da_writepages(struct address_space *mapping,
2389 */ 2461 */
2390 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) 2462 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2391 return 0; 2463 return 0;
2464
2465 /*
2466 * If the filesystem has aborted, it is read-only, so return
2467 * right away instead of dumping stack traces later on that
2468 * will obscure the real source of the problem. We test
2469 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2470 * the latter could be true if the filesystem is mounted
2471 * read-only, and in that case, ext4_da_writepages should
2472 * *never* be called, so if that ever happens, we would want
2473 * the stack trace.
2474 */
2475 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2476 return -EROFS;
2477
2392 /* 2478 /*
2393 * Make sure nr_to_write is >= sbi->s_mb_stream_request 2479 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2394 * This make sure small files blocks are allocated in 2480 * This make sure small files blocks are allocated in
@@ -2433,7 +2519,7 @@ static int ext4_da_writepages(struct address_space *mapping,
2433 handle = ext4_journal_start(inode, needed_blocks); 2519 handle = ext4_journal_start(inode, needed_blocks);
2434 if (IS_ERR(handle)) { 2520 if (IS_ERR(handle)) {
2435 ret = PTR_ERR(handle); 2521 ret = PTR_ERR(handle);
2436 printk(KERN_EMERG "%s: jbd2_start: " 2522 printk(KERN_CRIT "%s: jbd2_start: "
2437 "%ld pages, ino %lu; err %d\n", __func__, 2523 "%ld pages, ino %lu; err %d\n", __func__,
2438 wbc->nr_to_write, inode->i_ino, ret); 2524 wbc->nr_to_write, inode->i_ino, ret);
2439 dump_stack(); 2525 dump_stack();
@@ -2486,6 +2572,14 @@ out_writepages:
2486 if (!no_nrwrite_index_update) 2572 if (!no_nrwrite_index_update)
2487 wbc->no_nrwrite_index_update = 0; 2573 wbc->no_nrwrite_index_update = 0;
2488 wbc->nr_to_write -= nr_to_writebump; 2574 wbc->nr_to_write -= nr_to_writebump;
2575 trace_mark(ext4_da_writepage_result,
2576 "dev %s ino %lu ret %d pages_written %d "
2577 "pages_skipped %ld congestion %d "
2578 "more_io %d no_nrwrite_index_update %d",
2579 inode->i_sb->s_id, inode->i_ino, ret,
2580 pages_written, wbc->pages_skipped,
2581 wbc->encountered_congestion, wbc->more_io,
2582 wbc->no_nrwrite_index_update);
2489 return ret; 2583 return ret;
2490} 2584}
2491 2585
@@ -2537,6 +2631,11 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2537 len, flags, pagep, fsdata); 2631 len, flags, pagep, fsdata);
2538 } 2632 }
2539 *fsdata = (void *)0; 2633 *fsdata = (void *)0;
2634
2635 trace_mark(ext4_da_write_begin,
2636 "dev %s ino %lu pos %llu len %u flags %u",
2637 inode->i_sb->s_id, inode->i_ino,
2638 (unsigned long long) pos, len, flags);
2540retry: 2639retry:
2541 /* 2640 /*
2542 * With delayed allocation, we don't log the i_disksize update 2641 * With delayed allocation, we don't log the i_disksize update
@@ -2626,6 +2725,10 @@ static int ext4_da_write_end(struct file *file,
2626 } 2725 }
2627 } 2726 }
2628 2727
2728 trace_mark(ext4_da_write_end,
2729 "dev %s ino %lu pos %llu len %u copied %u",
2730 inode->i_sb->s_id, inode->i_ino,
2731 (unsigned long long) pos, len, copied);
2629 start = pos & (PAGE_CACHE_SIZE - 1); 2732 start = pos & (PAGE_CACHE_SIZE - 1);
2630 end = start + copied - 1; 2733 end = start + copied - 1;
2631 2734
@@ -2718,7 +2821,10 @@ static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2718 filemap_write_and_wait(mapping); 2821 filemap_write_and_wait(mapping);
2719 } 2822 }
2720 2823
2721 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) { 2824 BUG_ON(!EXT4_JOURNAL(inode) &&
2825 EXT4_I(inode)->i_state & EXT4_STATE_JDATA);
2826
2827 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2722 /* 2828 /*
2723 * This is a REALLY heavyweight approach, but the use of 2829 * This is a REALLY heavyweight approach, but the use of
2724 * bmap on dirty files is expected to be extremely rare: 2830 * bmap on dirty files is expected to be extremely rare:
@@ -2836,6 +2942,9 @@ static int ext4_normal_writepage(struct page *page,
2836 loff_t size = i_size_read(inode); 2942 loff_t size = i_size_read(inode);
2837 loff_t len; 2943 loff_t len;
2838 2944
2945 trace_mark(ext4_normal_writepage,
2946 "dev %s ino %lu page_index %lu",
2947 inode->i_sb->s_id, inode->i_ino, page->index);
2839 J_ASSERT(PageLocked(page)); 2948 J_ASSERT(PageLocked(page));
2840 if (page->index == size >> PAGE_CACHE_SHIFT) 2949 if (page->index == size >> PAGE_CACHE_SHIFT)
2841 len = size & ~PAGE_CACHE_MASK; 2950 len = size & ~PAGE_CACHE_MASK;
@@ -2921,6 +3030,9 @@ static int ext4_journalled_writepage(struct page *page,
2921 loff_t size = i_size_read(inode); 3030 loff_t size = i_size_read(inode);
2922 loff_t len; 3031 loff_t len;
2923 3032
3033 trace_mark(ext4_journalled_writepage,
3034 "dev %s ino %lu page_index %lu",
3035 inode->i_sb->s_id, inode->i_ino, page->index);
2924 J_ASSERT(PageLocked(page)); 3036 J_ASSERT(PageLocked(page));
2925 if (page->index == size >> PAGE_CACHE_SHIFT) 3037 if (page->index == size >> PAGE_CACHE_SHIFT)
2926 len = size & ~PAGE_CACHE_MASK; 3038 len = size & ~PAGE_CACHE_MASK;
@@ -2989,7 +3101,10 @@ static void ext4_invalidatepage(struct page *page, unsigned long offset)
2989 if (offset == 0) 3101 if (offset == 0)
2990 ClearPageChecked(page); 3102 ClearPageChecked(page);
2991 3103
2992 jbd2_journal_invalidatepage(journal, page, offset); 3104 if (journal)
3105 jbd2_journal_invalidatepage(journal, page, offset);
3106 else
3107 block_invalidatepage(page, offset);
2993} 3108}
2994 3109
2995static int ext4_releasepage(struct page *page, gfp_t wait) 3110static int ext4_releasepage(struct page *page, gfp_t wait)
@@ -2999,7 +3114,10 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
2999 WARN_ON(PageChecked(page)); 3114 WARN_ON(PageChecked(page));
3000 if (!page_has_buffers(page)) 3115 if (!page_has_buffers(page))
3001 return 0; 3116 return 0;
3002 return jbd2_journal_try_to_free_buffers(journal, page, wait); 3117 if (journal)
3118 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3119 else
3120 return try_to_free_buffers(page);
3003} 3121}
3004 3122
3005/* 3123/*
@@ -3271,7 +3389,7 @@ int ext4_block_truncate_page(handle_t *handle,
3271 3389
3272 err = 0; 3390 err = 0;
3273 if (ext4_should_journal_data(inode)) { 3391 if (ext4_should_journal_data(inode)) {
3274 err = ext4_journal_dirty_metadata(handle, bh); 3392 err = ext4_handle_dirty_metadata(handle, inode, bh);
3275 } else { 3393 } else {
3276 if (ext4_should_order_data(inode)) 3394 if (ext4_should_order_data(inode))
3277 err = ext4_jbd2_file_inode(handle, inode); 3395 err = ext4_jbd2_file_inode(handle, inode);
@@ -3395,8 +3513,8 @@ static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3395 __le32 *p; 3513 __le32 *p;
3396 if (try_to_extend_transaction(handle, inode)) { 3514 if (try_to_extend_transaction(handle, inode)) {
3397 if (bh) { 3515 if (bh) {
3398 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); 3516 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3399 ext4_journal_dirty_metadata(handle, bh); 3517 ext4_handle_dirty_metadata(handle, inode, bh);
3400 } 3518 }
3401 ext4_mark_inode_dirty(handle, inode); 3519 ext4_mark_inode_dirty(handle, inode);
3402 ext4_journal_test_restart(handle, inode); 3520 ext4_journal_test_restart(handle, inode);
@@ -3496,7 +3614,7 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
3496 count, block_to_free_p, p); 3614 count, block_to_free_p, p);
3497 3615
3498 if (this_bh) { 3616 if (this_bh) {
3499 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata"); 3617 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3500 3618
3501 /* 3619 /*
3502 * The buffer head should have an attached journal head at this 3620 * The buffer head should have an attached journal head at this
@@ -3505,7 +3623,7 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
3505 * the block was cleared. Check for this instead of OOPSing. 3623 * the block was cleared. Check for this instead of OOPSing.
3506 */ 3624 */
3507 if (bh2jh(this_bh)) 3625 if (bh2jh(this_bh))
3508 ext4_journal_dirty_metadata(handle, this_bh); 3626 ext4_handle_dirty_metadata(handle, inode, this_bh);
3509 else 3627 else
3510 ext4_error(inode->i_sb, __func__, 3628 ext4_error(inode->i_sb, __func__,
3511 "circular indirect block detected, " 3629 "circular indirect block detected, "
@@ -3535,7 +3653,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
3535 ext4_fsblk_t nr; 3653 ext4_fsblk_t nr;
3536 __le32 *p; 3654 __le32 *p;
3537 3655
3538 if (is_handle_aborted(handle)) 3656 if (ext4_handle_is_aborted(handle))
3539 return; 3657 return;
3540 3658
3541 if (depth--) { 3659 if (depth--) {
@@ -3605,7 +3723,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
3605 * will merely complain about releasing a free block, 3723 * will merely complain about releasing a free block,
3606 * rather than leaking blocks. 3724 * rather than leaking blocks.
3607 */ 3725 */
3608 if (is_handle_aborted(handle)) 3726 if (ext4_handle_is_aborted(handle))
3609 return; 3727 return;
3610 if (try_to_extend_transaction(handle, inode)) { 3728 if (try_to_extend_transaction(handle, inode)) {
3611 ext4_mark_inode_dirty(handle, inode); 3729 ext4_mark_inode_dirty(handle, inode);
@@ -3624,9 +3742,10 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
3624 parent_bh)){ 3742 parent_bh)){
3625 *p = 0; 3743 *p = 0;
3626 BUFFER_TRACE(parent_bh, 3744 BUFFER_TRACE(parent_bh,
3627 "call ext4_journal_dirty_metadata"); 3745 "call ext4_handle_dirty_metadata");
3628 ext4_journal_dirty_metadata(handle, 3746 ext4_handle_dirty_metadata(handle,
3629 parent_bh); 3747 inode,
3748 parent_bh);
3630 } 3749 }
3631 } 3750 }
3632 } 3751 }
@@ -3814,7 +3933,7 @@ do_indirects:
3814 * synchronous 3933 * synchronous
3815 */ 3934 */
3816 if (IS_SYNC(inode)) 3935 if (IS_SYNC(inode))
3817 handle->h_sync = 1; 3936 ext4_handle_sync(handle);
3818out_stop: 3937out_stop:
3819 /* 3938 /*
3820 * If this was a simple ftruncate(), and the file will remain alive 3939 * If this was a simple ftruncate(), and the file will remain alive
@@ -3844,7 +3963,7 @@ static int __ext4_get_inode_loc(struct inode *inode,
3844 ext4_fsblk_t block; 3963 ext4_fsblk_t block;
3845 int inodes_per_block, inode_offset; 3964 int inodes_per_block, inode_offset;
3846 3965
3847 iloc->bh = 0; 3966 iloc->bh = NULL;
3848 if (!ext4_valid_inum(sb, inode->i_ino)) 3967 if (!ext4_valid_inum(sb, inode->i_ino))
3849 return -EIO; 3968 return -EIO;
3850 3969
@@ -3951,7 +4070,7 @@ make_io:
3951 num = EXT4_INODES_PER_GROUP(sb); 4070 num = EXT4_INODES_PER_GROUP(sb);
3952 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, 4071 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3953 EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) 4072 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3954 num -= le16_to_cpu(gdp->bg_itable_unused); 4073 num -= ext4_itable_unused_count(sb, gdp);
3955 table += num / inodes_per_block; 4074 table += num / inodes_per_block;
3956 if (end > table) 4075 if (end > table)
3957 end = table; 4076 end = table;
@@ -4313,8 +4432,8 @@ static int ext4_do_update_inode(handle_t *handle,
4313 EXT4_SET_RO_COMPAT_FEATURE(sb, 4432 EXT4_SET_RO_COMPAT_FEATURE(sb,
4314 EXT4_FEATURE_RO_COMPAT_LARGE_FILE); 4433 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4315 sb->s_dirt = 1; 4434 sb->s_dirt = 1;
4316 handle->h_sync = 1; 4435 ext4_handle_sync(handle);
4317 err = ext4_journal_dirty_metadata(handle, 4436 err = ext4_handle_dirty_metadata(handle, inode,
4318 EXT4_SB(sb)->s_sbh); 4437 EXT4_SB(sb)->s_sbh);
4319 } 4438 }
4320 } 4439 }
@@ -4341,9 +4460,8 @@ static int ext4_do_update_inode(handle_t *handle,
4341 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); 4460 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4342 } 4461 }
4343 4462
4344 4463 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4345 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata"); 4464 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4346 rc = ext4_journal_dirty_metadata(handle, bh);
4347 if (!err) 4465 if (!err)
4348 err = rc; 4466 err = rc;
4349 ei->i_state &= ~EXT4_STATE_NEW; 4467 ei->i_state &= ~EXT4_STATE_NEW;
@@ -4406,6 +4524,25 @@ int ext4_write_inode(struct inode *inode, int wait)
4406 return ext4_force_commit(inode->i_sb); 4524 return ext4_force_commit(inode->i_sb);
4407} 4525}
4408 4526
4527int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4528{
4529 int err = 0;
4530
4531 mark_buffer_dirty(bh);
4532 if (inode && inode_needs_sync(inode)) {
4533 sync_dirty_buffer(bh);
4534 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4535 ext4_error(inode->i_sb, __func__,
4536 "IO error syncing inode, "
4537 "inode=%lu, block=%llu",
4538 inode->i_ino,
4539 (unsigned long long)bh->b_blocknr);
4540 err = -EIO;
4541 }
4542 }
4543 return err;
4544}
4545
4409/* 4546/*
4410 * ext4_setattr() 4547 * ext4_setattr()
4411 * 4548 *
@@ -4710,16 +4847,15 @@ int
4710ext4_reserve_inode_write(handle_t *handle, struct inode *inode, 4847ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4711 struct ext4_iloc *iloc) 4848 struct ext4_iloc *iloc)
4712{ 4849{
4713 int err = 0; 4850 int err;
4714 if (handle) { 4851
4715 err = ext4_get_inode_loc(inode, iloc); 4852 err = ext4_get_inode_loc(inode, iloc);
4716 if (!err) { 4853 if (!err) {
4717 BUFFER_TRACE(iloc->bh, "get_write_access"); 4854 BUFFER_TRACE(iloc->bh, "get_write_access");
4718 err = ext4_journal_get_write_access(handle, iloc->bh); 4855 err = ext4_journal_get_write_access(handle, iloc->bh);
4719 if (err) { 4856 if (err) {
4720 brelse(iloc->bh); 4857 brelse(iloc->bh);
4721 iloc->bh = NULL; 4858 iloc->bh = NULL;
4722 }
4723 } 4859 }
4724 } 4860 }
4725 ext4_std_error(inode->i_sb, err); 4861 ext4_std_error(inode->i_sb, err);
@@ -4791,7 +4927,8 @@ int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4791 4927
4792 might_sleep(); 4928 might_sleep();
4793 err = ext4_reserve_inode_write(handle, inode, &iloc); 4929 err = ext4_reserve_inode_write(handle, inode, &iloc);
4794 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && 4930 if (ext4_handle_valid(handle) &&
4931 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4795 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) { 4932 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4796 /* 4933 /*
4797 * We need extra buffer credits since we may write into EA block 4934 * We need extra buffer credits since we may write into EA block
@@ -4843,6 +4980,11 @@ void ext4_dirty_inode(struct inode *inode)
4843 handle_t *current_handle = ext4_journal_current_handle(); 4980 handle_t *current_handle = ext4_journal_current_handle();
4844 handle_t *handle; 4981 handle_t *handle;
4845 4982
4983 if (!ext4_handle_valid(current_handle)) {
4984 ext4_mark_inode_dirty(current_handle, inode);
4985 return;
4986 }
4987
4846 handle = ext4_journal_start(inode, 2); 4988 handle = ext4_journal_start(inode, 2);
4847 if (IS_ERR(handle)) 4989 if (IS_ERR(handle))
4848 goto out; 4990 goto out;
@@ -4880,8 +5022,9 @@ static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4880 BUFFER_TRACE(iloc.bh, "get_write_access"); 5022 BUFFER_TRACE(iloc.bh, "get_write_access");
4881 err = jbd2_journal_get_write_access(handle, iloc.bh); 5023 err = jbd2_journal_get_write_access(handle, iloc.bh);
4882 if (!err) 5024 if (!err)
4883 err = ext4_journal_dirty_metadata(handle, 5025 err = ext4_handle_dirty_metadata(handle,
4884 iloc.bh); 5026 inode,
5027 iloc.bh);
4885 brelse(iloc.bh); 5028 brelse(iloc.bh);
4886 } 5029 }
4887 } 5030 }
@@ -4907,6 +5050,8 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
4907 */ 5050 */
4908 5051
4909 journal = EXT4_JOURNAL(inode); 5052 journal = EXT4_JOURNAL(inode);
5053 if (!journal)
5054 return 0;
4910 if (is_journal_aborted(journal)) 5055 if (is_journal_aborted(journal))
4911 return -EROFS; 5056 return -EROFS;
4912 5057
@@ -4936,7 +5081,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
4936 return PTR_ERR(handle); 5081 return PTR_ERR(handle);
4937 5082
4938 err = ext4_mark_inode_dirty(handle, inode); 5083 err = ext4_mark_inode_dirty(handle, inode);
4939 handle->h_sync = 1; 5084 ext4_handle_sync(handle);
4940 ext4_journal_stop(handle); 5085 ext4_journal_stop(handle);
4941 ext4_std_error(inode->i_sb, err); 5086 ext4_std_error(inode->i_sb, err);
4942 5087