diff options
Diffstat (limited to 'fs/ext4/inode.c')
-rw-r--r-- | fs/ext4/inode.c | 410 |
1 files changed, 176 insertions, 234 deletions
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c index 9297ad46c465..1a86282b9024 100644 --- a/fs/ext4/inode.c +++ b/fs/ext4/inode.c | |||
@@ -173,7 +173,7 @@ int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, | |||
173 | BUG_ON(EXT4_JOURNAL(inode) == NULL); | 173 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
174 | jbd_debug(2, "restarting handle %p\n", handle); | 174 | jbd_debug(2, "restarting handle %p\n", handle); |
175 | up_write(&EXT4_I(inode)->i_data_sem); | 175 | up_write(&EXT4_I(inode)->i_data_sem); |
176 | ret = ext4_journal_restart(handle, blocks_for_truncate(inode)); | 176 | ret = ext4_journal_restart(handle, nblocks); |
177 | down_write(&EXT4_I(inode)->i_data_sem); | 177 | down_write(&EXT4_I(inode)->i_data_sem); |
178 | ext4_discard_preallocations(inode); | 178 | ext4_discard_preallocations(inode); |
179 | 179 | ||
@@ -720,7 +720,7 @@ allocated: | |||
720 | return ret; | 720 | return ret; |
721 | failed_out: | 721 | failed_out: |
722 | for (i = 0; i < index; i++) | 722 | for (i = 0; i < index; i++) |
723 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); | 723 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
724 | return ret; | 724 | return ret; |
725 | } | 725 | } |
726 | 726 | ||
@@ -823,20 +823,20 @@ static int ext4_alloc_branch(handle_t *handle, struct inode *inode, | |||
823 | return err; | 823 | return err; |
824 | failed: | 824 | failed: |
825 | /* Allocation failed, free what we already allocated */ | 825 | /* Allocation failed, free what we already allocated */ |
826 | ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0); | 826 | ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0); |
827 | for (i = 1; i <= n ; i++) { | 827 | for (i = 1; i <= n ; i++) { |
828 | /* | 828 | /* |
829 | * branch[i].bh is newly allocated, so there is no | 829 | * branch[i].bh is newly allocated, so there is no |
830 | * need to revoke the block, which is why we don't | 830 | * need to revoke the block, which is why we don't |
831 | * need to set EXT4_FREE_BLOCKS_METADATA. | 831 | * need to set EXT4_FREE_BLOCKS_METADATA. |
832 | */ | 832 | */ |
833 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, | 833 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, |
834 | EXT4_FREE_BLOCKS_FORGET); | 834 | EXT4_FREE_BLOCKS_FORGET); |
835 | } | 835 | } |
836 | for (i = n+1; i < indirect_blks; i++) | 836 | for (i = n+1; i < indirect_blks; i++) |
837 | ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0); | 837 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0); |
838 | 838 | ||
839 | ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0); | 839 | ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0); |
840 | 840 | ||
841 | return err; | 841 | return err; |
842 | } | 842 | } |
@@ -924,7 +924,7 @@ err_out: | |||
924 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, | 924 | ext4_free_blocks(handle, inode, where[i].bh, 0, 1, |
925 | EXT4_FREE_BLOCKS_FORGET); | 925 | EXT4_FREE_BLOCKS_FORGET); |
926 | } | 926 | } |
927 | ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key), | 927 | ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key), |
928 | blks, 0); | 928 | blks, 0); |
929 | 929 | ||
930 | return err; | 930 | return err; |
@@ -973,6 +973,7 @@ static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode, | |||
973 | int count = 0; | 973 | int count = 0; |
974 | ext4_fsblk_t first_block = 0; | 974 | ext4_fsblk_t first_block = 0; |
975 | 975 | ||
976 | trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags); | ||
976 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); | 977 | J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))); |
977 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); | 978 | J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0); |
978 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, | 979 | depth = ext4_block_to_path(inode, map->m_lblk, offsets, |
@@ -1058,6 +1059,8 @@ cleanup: | |||
1058 | partial--; | 1059 | partial--; |
1059 | } | 1060 | } |
1060 | out: | 1061 | out: |
1062 | trace_ext4_ind_map_blocks_exit(inode, map->m_lblk, | ||
1063 | map->m_pblk, map->m_len, err); | ||
1061 | return err; | 1064 | return err; |
1062 | } | 1065 | } |
1063 | 1066 | ||
@@ -2060,7 +2063,7 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd, | |||
2060 | if (nr_pages == 0) | 2063 | if (nr_pages == 0) |
2061 | break; | 2064 | break; |
2062 | for (i = 0; i < nr_pages; i++) { | 2065 | for (i = 0; i < nr_pages; i++) { |
2063 | int commit_write = 0, redirty_page = 0; | 2066 | int commit_write = 0, skip_page = 0; |
2064 | struct page *page = pvec.pages[i]; | 2067 | struct page *page = pvec.pages[i]; |
2065 | 2068 | ||
2066 | index = page->index; | 2069 | index = page->index; |
@@ -2086,14 +2089,12 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd, | |||
2086 | * If the page does not have buffers (for | 2089 | * If the page does not have buffers (for |
2087 | * whatever reason), try to create them using | 2090 | * whatever reason), try to create them using |
2088 | * __block_write_begin. If this fails, | 2091 | * __block_write_begin. If this fails, |
2089 | * redirty the page and move on. | 2092 | * skip the page and move on. |
2090 | */ | 2093 | */ |
2091 | if (!page_has_buffers(page)) { | 2094 | if (!page_has_buffers(page)) { |
2092 | if (__block_write_begin(page, 0, len, | 2095 | if (__block_write_begin(page, 0, len, |
2093 | noalloc_get_block_write)) { | 2096 | noalloc_get_block_write)) { |
2094 | redirty_page: | 2097 | skip_page: |
2095 | redirty_page_for_writepage(mpd->wbc, | ||
2096 | page); | ||
2097 | unlock_page(page); | 2098 | unlock_page(page); |
2098 | continue; | 2099 | continue; |
2099 | } | 2100 | } |
@@ -2104,7 +2105,7 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd, | |||
2104 | block_start = 0; | 2105 | block_start = 0; |
2105 | do { | 2106 | do { |
2106 | if (!bh) | 2107 | if (!bh) |
2107 | goto redirty_page; | 2108 | goto skip_page; |
2108 | if (map && (cur_logical >= map->m_lblk) && | 2109 | if (map && (cur_logical >= map->m_lblk) && |
2109 | (cur_logical <= (map->m_lblk + | 2110 | (cur_logical <= (map->m_lblk + |
2110 | (map->m_len - 1)))) { | 2111 | (map->m_len - 1)))) { |
@@ -2120,22 +2121,23 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd, | |||
2120 | clear_buffer_unwritten(bh); | 2121 | clear_buffer_unwritten(bh); |
2121 | } | 2122 | } |
2122 | 2123 | ||
2123 | /* redirty page if block allocation undone */ | 2124 | /* skip page if block allocation undone */ |
2124 | if (buffer_delay(bh) || buffer_unwritten(bh)) | 2125 | if (buffer_delay(bh) || buffer_unwritten(bh)) |
2125 | redirty_page = 1; | 2126 | skip_page = 1; |
2126 | bh = bh->b_this_page; | 2127 | bh = bh->b_this_page; |
2127 | block_start += bh->b_size; | 2128 | block_start += bh->b_size; |
2128 | cur_logical++; | 2129 | cur_logical++; |
2129 | pblock++; | 2130 | pblock++; |
2130 | } while (bh != page_bufs); | 2131 | } while (bh != page_bufs); |
2131 | 2132 | ||
2132 | if (redirty_page) | 2133 | if (skip_page) |
2133 | goto redirty_page; | 2134 | goto skip_page; |
2134 | 2135 | ||
2135 | if (commit_write) | 2136 | if (commit_write) |
2136 | /* mark the buffer_heads as dirty & uptodate */ | 2137 | /* mark the buffer_heads as dirty & uptodate */ |
2137 | block_commit_write(page, 0, len); | 2138 | block_commit_write(page, 0, len); |
2138 | 2139 | ||
2140 | clear_page_dirty_for_io(page); | ||
2139 | /* | 2141 | /* |
2140 | * Delalloc doesn't support data journalling, | 2142 | * Delalloc doesn't support data journalling, |
2141 | * but eventually maybe we'll lift this | 2143 | * but eventually maybe we'll lift this |
@@ -2165,8 +2167,7 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd, | |||
2165 | return ret; | 2167 | return ret; |
2166 | } | 2168 | } |
2167 | 2169 | ||
2168 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, | 2170 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
2169 | sector_t logical, long blk_cnt) | ||
2170 | { | 2171 | { |
2171 | int nr_pages, i; | 2172 | int nr_pages, i; |
2172 | pgoff_t index, end; | 2173 | pgoff_t index, end; |
@@ -2174,9 +2175,8 @@ static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd, | |||
2174 | struct inode *inode = mpd->inode; | 2175 | struct inode *inode = mpd->inode; |
2175 | struct address_space *mapping = inode->i_mapping; | 2176 | struct address_space *mapping = inode->i_mapping; |
2176 | 2177 | ||
2177 | index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | 2178 | index = mpd->first_page; |
2178 | end = (logical + blk_cnt - 1) >> | 2179 | end = mpd->next_page - 1; |
2179 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | ||
2180 | while (index <= end) { | 2180 | while (index <= end) { |
2181 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | 2181 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
2182 | if (nr_pages == 0) | 2182 | if (nr_pages == 0) |
@@ -2279,9 +2279,8 @@ static void mpage_da_map_and_submit(struct mpage_da_data *mpd) | |||
2279 | err = blks; | 2279 | err = blks; |
2280 | /* | 2280 | /* |
2281 | * If get block returns EAGAIN or ENOSPC and there | 2281 | * If get block returns EAGAIN or ENOSPC and there |
2282 | * appears to be free blocks we will call | 2282 | * appears to be free blocks we will just let |
2283 | * ext4_writepage() for all of the pages which will | 2283 | * mpage_da_submit_io() unlock all of the pages. |
2284 | * just redirty the pages. | ||
2285 | */ | 2284 | */ |
2286 | if (err == -EAGAIN) | 2285 | if (err == -EAGAIN) |
2287 | goto submit_io; | 2286 | goto submit_io; |
@@ -2312,8 +2311,10 @@ static void mpage_da_map_and_submit(struct mpage_da_data *mpd) | |||
2312 | ext4_print_free_blocks(mpd->inode); | 2311 | ext4_print_free_blocks(mpd->inode); |
2313 | } | 2312 | } |
2314 | /* invalidate all the pages */ | 2313 | /* invalidate all the pages */ |
2315 | ext4_da_block_invalidatepages(mpd, next, | 2314 | ext4_da_block_invalidatepages(mpd); |
2316 | mpd->b_size >> mpd->inode->i_blkbits); | 2315 | |
2316 | /* Mark this page range as having been completed */ | ||
2317 | mpd->io_done = 1; | ||
2317 | return; | 2318 | return; |
2318 | } | 2319 | } |
2319 | BUG_ON(blks == 0); | 2320 | BUG_ON(blks == 0); |
@@ -2438,102 +2439,6 @@ static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) | |||
2438 | } | 2439 | } |
2439 | 2440 | ||
2440 | /* | 2441 | /* |
2441 | * __mpage_da_writepage - finds extent of pages and blocks | ||
2442 | * | ||
2443 | * @page: page to consider | ||
2444 | * @wbc: not used, we just follow rules | ||
2445 | * @data: context | ||
2446 | * | ||
2447 | * The function finds extents of pages and scan them for all blocks. | ||
2448 | */ | ||
2449 | static int __mpage_da_writepage(struct page *page, | ||
2450 | struct writeback_control *wbc, | ||
2451 | struct mpage_da_data *mpd) | ||
2452 | { | ||
2453 | struct inode *inode = mpd->inode; | ||
2454 | struct buffer_head *bh, *head; | ||
2455 | sector_t logical; | ||
2456 | |||
2457 | /* | ||
2458 | * Can we merge this page to current extent? | ||
2459 | */ | ||
2460 | if (mpd->next_page != page->index) { | ||
2461 | /* | ||
2462 | * Nope, we can't. So, we map non-allocated blocks | ||
2463 | * and start IO on them | ||
2464 | */ | ||
2465 | if (mpd->next_page != mpd->first_page) { | ||
2466 | mpage_da_map_and_submit(mpd); | ||
2467 | /* | ||
2468 | * skip rest of the page in the page_vec | ||
2469 | */ | ||
2470 | redirty_page_for_writepage(wbc, page); | ||
2471 | unlock_page(page); | ||
2472 | return MPAGE_DA_EXTENT_TAIL; | ||
2473 | } | ||
2474 | |||
2475 | /* | ||
2476 | * Start next extent of pages ... | ||
2477 | */ | ||
2478 | mpd->first_page = page->index; | ||
2479 | |||
2480 | /* | ||
2481 | * ... and blocks | ||
2482 | */ | ||
2483 | mpd->b_size = 0; | ||
2484 | mpd->b_state = 0; | ||
2485 | mpd->b_blocknr = 0; | ||
2486 | } | ||
2487 | |||
2488 | mpd->next_page = page->index + 1; | ||
2489 | logical = (sector_t) page->index << | ||
2490 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | ||
2491 | |||
2492 | if (!page_has_buffers(page)) { | ||
2493 | mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE, | ||
2494 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | ||
2495 | if (mpd->io_done) | ||
2496 | return MPAGE_DA_EXTENT_TAIL; | ||
2497 | } else { | ||
2498 | /* | ||
2499 | * Page with regular buffer heads, just add all dirty ones | ||
2500 | */ | ||
2501 | head = page_buffers(page); | ||
2502 | bh = head; | ||
2503 | do { | ||
2504 | BUG_ON(buffer_locked(bh)); | ||
2505 | /* | ||
2506 | * We need to try to allocate | ||
2507 | * unmapped blocks in the same page. | ||
2508 | * Otherwise we won't make progress | ||
2509 | * with the page in ext4_writepage | ||
2510 | */ | ||
2511 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | ||
2512 | mpage_add_bh_to_extent(mpd, logical, | ||
2513 | bh->b_size, | ||
2514 | bh->b_state); | ||
2515 | if (mpd->io_done) | ||
2516 | return MPAGE_DA_EXTENT_TAIL; | ||
2517 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { | ||
2518 | /* | ||
2519 | * mapped dirty buffer. We need to update | ||
2520 | * the b_state because we look at | ||
2521 | * b_state in mpage_da_map_blocks. We don't | ||
2522 | * update b_size because if we find an | ||
2523 | * unmapped buffer_head later we need to | ||
2524 | * use the b_state flag of that buffer_head. | ||
2525 | */ | ||
2526 | if (mpd->b_size == 0) | ||
2527 | mpd->b_state = bh->b_state & BH_FLAGS; | ||
2528 | } | ||
2529 | logical++; | ||
2530 | } while ((bh = bh->b_this_page) != head); | ||
2531 | } | ||
2532 | |||
2533 | return 0; | ||
2534 | } | ||
2535 | |||
2536 | /* | ||
2537 | * This is a special get_blocks_t callback which is used by | 2442 | * This is a special get_blocks_t callback which is used by |
2538 | * ext4_da_write_begin(). It will either return mapped block or | 2443 | * ext4_da_write_begin(). It will either return mapped block or |
2539 | * reserve space for a single block. | 2444 | * reserve space for a single block. |
@@ -2597,7 +2502,6 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |||
2597 | * for partial write. | 2502 | * for partial write. |
2598 | */ | 2503 | */ |
2599 | set_buffer_new(bh); | 2504 | set_buffer_new(bh); |
2600 | set_buffer_mapped(bh); | ||
2601 | } | 2505 | } |
2602 | return 0; | 2506 | return 0; |
2603 | } | 2507 | } |
@@ -2811,27 +2715,27 @@ static int ext4_da_writepages_trans_blocks(struct inode *inode) | |||
2811 | 2715 | ||
2812 | /* | 2716 | /* |
2813 | * write_cache_pages_da - walk the list of dirty pages of the given | 2717 | * write_cache_pages_da - walk the list of dirty pages of the given |
2814 | * address space and call the callback function (which usually writes | 2718 | * address space and accumulate pages that need writing, and call |
2815 | * the pages). | 2719 | * mpage_da_map_and_submit to map a single contiguous memory region |
2816 | * | 2720 | * and then write them. |
2817 | * This is a forked version of write_cache_pages(). Differences: | ||
2818 | * Range cyclic is ignored. | ||
2819 | * no_nrwrite_index_update is always presumed true | ||
2820 | */ | 2721 | */ |
2821 | static int write_cache_pages_da(struct address_space *mapping, | 2722 | static int write_cache_pages_da(struct address_space *mapping, |
2822 | struct writeback_control *wbc, | 2723 | struct writeback_control *wbc, |
2823 | struct mpage_da_data *mpd, | 2724 | struct mpage_da_data *mpd, |
2824 | pgoff_t *done_index) | 2725 | pgoff_t *done_index) |
2825 | { | 2726 | { |
2826 | int ret = 0; | 2727 | struct buffer_head *bh, *head; |
2827 | int done = 0; | 2728 | struct inode *inode = mapping->host; |
2828 | struct pagevec pvec; | 2729 | struct pagevec pvec; |
2829 | unsigned nr_pages; | 2730 | unsigned int nr_pages; |
2830 | pgoff_t index; | 2731 | sector_t logical; |
2831 | pgoff_t end; /* Inclusive */ | 2732 | pgoff_t index, end; |
2832 | long nr_to_write = wbc->nr_to_write; | 2733 | long nr_to_write = wbc->nr_to_write; |
2833 | int tag; | 2734 | int i, tag, ret = 0; |
2834 | 2735 | ||
2736 | memset(mpd, 0, sizeof(struct mpage_da_data)); | ||
2737 | mpd->wbc = wbc; | ||
2738 | mpd->inode = inode; | ||
2835 | pagevec_init(&pvec, 0); | 2739 | pagevec_init(&pvec, 0); |
2836 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | 2740 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
2837 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | 2741 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
@@ -2842,13 +2746,11 @@ static int write_cache_pages_da(struct address_space *mapping, | |||
2842 | tag = PAGECACHE_TAG_DIRTY; | 2746 | tag = PAGECACHE_TAG_DIRTY; |
2843 | 2747 | ||
2844 | *done_index = index; | 2748 | *done_index = index; |
2845 | while (!done && (index <= end)) { | 2749 | while (index <= end) { |
2846 | int i; | ||
2847 | |||
2848 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | 2750 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
2849 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | 2751 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2850 | if (nr_pages == 0) | 2752 | if (nr_pages == 0) |
2851 | break; | 2753 | return 0; |
2852 | 2754 | ||
2853 | for (i = 0; i < nr_pages; i++) { | 2755 | for (i = 0; i < nr_pages; i++) { |
2854 | struct page *page = pvec.pages[i]; | 2756 | struct page *page = pvec.pages[i]; |
@@ -2860,60 +2762,100 @@ static int write_cache_pages_da(struct address_space *mapping, | |||
2860 | * mapping. However, page->index will not change | 2762 | * mapping. However, page->index will not change |
2861 | * because we have a reference on the page. | 2763 | * because we have a reference on the page. |
2862 | */ | 2764 | */ |
2863 | if (page->index > end) { | 2765 | if (page->index > end) |
2864 | done = 1; | 2766 | goto out; |
2865 | break; | ||
2866 | } | ||
2867 | 2767 | ||
2868 | *done_index = page->index + 1; | 2768 | *done_index = page->index + 1; |
2869 | 2769 | ||
2770 | /* | ||
2771 | * If we can't merge this page, and we have | ||
2772 | * accumulated an contiguous region, write it | ||
2773 | */ | ||
2774 | if ((mpd->next_page != page->index) && | ||
2775 | (mpd->next_page != mpd->first_page)) { | ||
2776 | mpage_da_map_and_submit(mpd); | ||
2777 | goto ret_extent_tail; | ||
2778 | } | ||
2779 | |||
2870 | lock_page(page); | 2780 | lock_page(page); |
2871 | 2781 | ||
2872 | /* | 2782 | /* |
2873 | * Page truncated or invalidated. We can freely skip it | 2783 | * If the page is no longer dirty, or its |
2874 | * then, even for data integrity operations: the page | 2784 | * mapping no longer corresponds to inode we |
2875 | * has disappeared concurrently, so there could be no | 2785 | * are writing (which means it has been |
2876 | * real expectation of this data interity operation | 2786 | * truncated or invalidated), or the page is |
2877 | * even if there is now a new, dirty page at the same | 2787 | * already under writeback and we are not |
2878 | * pagecache address. | 2788 | * doing a data integrity writeback, skip the page |
2879 | */ | 2789 | */ |
2880 | if (unlikely(page->mapping != mapping)) { | 2790 | if (!PageDirty(page) || |
2881 | continue_unlock: | 2791 | (PageWriteback(page) && |
2792 | (wbc->sync_mode == WB_SYNC_NONE)) || | ||
2793 | unlikely(page->mapping != mapping)) { | ||
2882 | unlock_page(page); | 2794 | unlock_page(page); |
2883 | continue; | 2795 | continue; |
2884 | } | 2796 | } |
2885 | 2797 | ||
2886 | if (!PageDirty(page)) { | 2798 | if (PageWriteback(page)) |
2887 | /* someone wrote it for us */ | 2799 | wait_on_page_writeback(page); |
2888 | goto continue_unlock; | ||
2889 | } | ||
2890 | |||
2891 | if (PageWriteback(page)) { | ||
2892 | if (wbc->sync_mode != WB_SYNC_NONE) | ||
2893 | wait_on_page_writeback(page); | ||
2894 | else | ||
2895 | goto continue_unlock; | ||
2896 | } | ||
2897 | 2800 | ||
2898 | BUG_ON(PageWriteback(page)); | 2801 | BUG_ON(PageWriteback(page)); |
2899 | if (!clear_page_dirty_for_io(page)) | ||
2900 | goto continue_unlock; | ||
2901 | 2802 | ||
2902 | ret = __mpage_da_writepage(page, wbc, mpd); | 2803 | if (mpd->next_page != page->index) |
2903 | if (unlikely(ret)) { | 2804 | mpd->first_page = page->index; |
2904 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | 2805 | mpd->next_page = page->index + 1; |
2905 | unlock_page(page); | 2806 | logical = (sector_t) page->index << |
2906 | ret = 0; | 2807 | (PAGE_CACHE_SHIFT - inode->i_blkbits); |
2907 | } else { | 2808 | |
2908 | done = 1; | 2809 | if (!page_has_buffers(page)) { |
2909 | break; | 2810 | mpage_add_bh_to_extent(mpd, logical, |
2910 | } | 2811 | PAGE_CACHE_SIZE, |
2812 | (1 << BH_Dirty) | (1 << BH_Uptodate)); | ||
2813 | if (mpd->io_done) | ||
2814 | goto ret_extent_tail; | ||
2815 | } else { | ||
2816 | /* | ||
2817 | * Page with regular buffer heads, | ||
2818 | * just add all dirty ones | ||
2819 | */ | ||
2820 | head = page_buffers(page); | ||
2821 | bh = head; | ||
2822 | do { | ||
2823 | BUG_ON(buffer_locked(bh)); | ||
2824 | /* | ||
2825 | * We need to try to allocate | ||
2826 | * unmapped blocks in the same page. | ||
2827 | * Otherwise we won't make progress | ||
2828 | * with the page in ext4_writepage | ||
2829 | */ | ||
2830 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | ||
2831 | mpage_add_bh_to_extent(mpd, logical, | ||
2832 | bh->b_size, | ||
2833 | bh->b_state); | ||
2834 | if (mpd->io_done) | ||
2835 | goto ret_extent_tail; | ||
2836 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { | ||
2837 | /* | ||
2838 | * mapped dirty buffer. We need | ||
2839 | * to update the b_state | ||
2840 | * because we look at b_state | ||
2841 | * in mpage_da_map_blocks. We | ||
2842 | * don't update b_size because | ||
2843 | * if we find an unmapped | ||
2844 | * buffer_head later we need to | ||
2845 | * use the b_state flag of that | ||
2846 | * buffer_head. | ||
2847 | */ | ||
2848 | if (mpd->b_size == 0) | ||
2849 | mpd->b_state = bh->b_state & BH_FLAGS; | ||
2850 | } | ||
2851 | logical++; | ||
2852 | } while ((bh = bh->b_this_page) != head); | ||
2911 | } | 2853 | } |
2912 | 2854 | ||
2913 | if (nr_to_write > 0) { | 2855 | if (nr_to_write > 0) { |
2914 | nr_to_write--; | 2856 | nr_to_write--; |
2915 | if (nr_to_write == 0 && | 2857 | if (nr_to_write == 0 && |
2916 | wbc->sync_mode == WB_SYNC_NONE) { | 2858 | wbc->sync_mode == WB_SYNC_NONE) |
2917 | /* | 2859 | /* |
2918 | * We stop writing back only if we are | 2860 | * We stop writing back only if we are |
2919 | * not doing integrity sync. In case of | 2861 | * not doing integrity sync. In case of |
@@ -2924,14 +2866,18 @@ continue_unlock: | |||
2924 | * pages, but have not synced all of the | 2866 | * pages, but have not synced all of the |
2925 | * old dirty pages. | 2867 | * old dirty pages. |
2926 | */ | 2868 | */ |
2927 | done = 1; | 2869 | goto out; |
2928 | break; | ||
2929 | } | ||
2930 | } | 2870 | } |
2931 | } | 2871 | } |
2932 | pagevec_release(&pvec); | 2872 | pagevec_release(&pvec); |
2933 | cond_resched(); | 2873 | cond_resched(); |
2934 | } | 2874 | } |
2875 | return 0; | ||
2876 | ret_extent_tail: | ||
2877 | ret = MPAGE_DA_EXTENT_TAIL; | ||
2878 | out: | ||
2879 | pagevec_release(&pvec); | ||
2880 | cond_resched(); | ||
2935 | return ret; | 2881 | return ret; |
2936 | } | 2882 | } |
2937 | 2883 | ||
@@ -2945,7 +2891,6 @@ static int ext4_da_writepages(struct address_space *mapping, | |||
2945 | struct mpage_da_data mpd; | 2891 | struct mpage_da_data mpd; |
2946 | struct inode *inode = mapping->host; | 2892 | struct inode *inode = mapping->host; |
2947 | int pages_written = 0; | 2893 | int pages_written = 0; |
2948 | long pages_skipped; | ||
2949 | unsigned int max_pages; | 2894 | unsigned int max_pages; |
2950 | int range_cyclic, cycled = 1, io_done = 0; | 2895 | int range_cyclic, cycled = 1, io_done = 0; |
2951 | int needed_blocks, ret = 0; | 2896 | int needed_blocks, ret = 0; |
@@ -3028,11 +2973,6 @@ static int ext4_da_writepages(struct address_space *mapping, | |||
3028 | wbc->nr_to_write = desired_nr_to_write; | 2973 | wbc->nr_to_write = desired_nr_to_write; |
3029 | } | 2974 | } |
3030 | 2975 | ||
3031 | mpd.wbc = wbc; | ||
3032 | mpd.inode = mapping->host; | ||
3033 | |||
3034 | pages_skipped = wbc->pages_skipped; | ||
3035 | |||
3036 | retry: | 2976 | retry: |
3037 | if (wbc->sync_mode == WB_SYNC_ALL) | 2977 | if (wbc->sync_mode == WB_SYNC_ALL) |
3038 | tag_pages_for_writeback(mapping, index, end); | 2978 | tag_pages_for_writeback(mapping, index, end); |
@@ -3059,22 +2999,10 @@ retry: | |||
3059 | } | 2999 | } |
3060 | 3000 | ||
3061 | /* | 3001 | /* |
3062 | * Now call __mpage_da_writepage to find the next | 3002 | * Now call write_cache_pages_da() to find the next |
3063 | * contiguous region of logical blocks that need | 3003 | * contiguous region of logical blocks that need |
3064 | * blocks to be allocated by ext4. We don't actually | 3004 | * blocks to be allocated by ext4 and submit them. |
3065 | * submit the blocks for I/O here, even though | ||
3066 | * write_cache_pages thinks it will, and will set the | ||
3067 | * pages as clean for write before calling | ||
3068 | * __mpage_da_writepage(). | ||
3069 | */ | 3005 | */ |
3070 | mpd.b_size = 0; | ||
3071 | mpd.b_state = 0; | ||
3072 | mpd.b_blocknr = 0; | ||
3073 | mpd.first_page = 0; | ||
3074 | mpd.next_page = 0; | ||
3075 | mpd.io_done = 0; | ||
3076 | mpd.pages_written = 0; | ||
3077 | mpd.retval = 0; | ||
3078 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); | 3006 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
3079 | /* | 3007 | /* |
3080 | * If we have a contiguous extent of pages and we | 3008 | * If we have a contiguous extent of pages and we |
@@ -3096,7 +3024,6 @@ retry: | |||
3096 | * and try again | 3024 | * and try again |
3097 | */ | 3025 | */ |
3098 | jbd2_journal_force_commit_nested(sbi->s_journal); | 3026 | jbd2_journal_force_commit_nested(sbi->s_journal); |
3099 | wbc->pages_skipped = pages_skipped; | ||
3100 | ret = 0; | 3027 | ret = 0; |
3101 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | 3028 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { |
3102 | /* | 3029 | /* |
@@ -3104,7 +3031,6 @@ retry: | |||
3104 | * rest of the pages | 3031 | * rest of the pages |
3105 | */ | 3032 | */ |
3106 | pages_written += mpd.pages_written; | 3033 | pages_written += mpd.pages_written; |
3107 | wbc->pages_skipped = pages_skipped; | ||
3108 | ret = 0; | 3034 | ret = 0; |
3109 | io_done = 1; | 3035 | io_done = 1; |
3110 | } else if (wbc->nr_to_write) | 3036 | } else if (wbc->nr_to_write) |
@@ -3122,11 +3048,6 @@ retry: | |||
3122 | wbc->range_end = mapping->writeback_index - 1; | 3048 | wbc->range_end = mapping->writeback_index - 1; |
3123 | goto retry; | 3049 | goto retry; |
3124 | } | 3050 | } |
3125 | if (pages_skipped != wbc->pages_skipped) | ||
3126 | ext4_msg(inode->i_sb, KERN_CRIT, | ||
3127 | "This should not happen leaving %s " | ||
3128 | "with nr_to_write = %ld ret = %d", | ||
3129 | __func__, wbc->nr_to_write, ret); | ||
3130 | 3051 | ||
3131 | /* Update index */ | 3052 | /* Update index */ |
3132 | wbc->range_cyclic = range_cyclic; | 3053 | wbc->range_cyclic = range_cyclic; |
@@ -3460,6 +3381,7 @@ static sector_t ext4_bmap(struct address_space *mapping, sector_t block) | |||
3460 | 3381 | ||
3461 | static int ext4_readpage(struct file *file, struct page *page) | 3382 | static int ext4_readpage(struct file *file, struct page *page) |
3462 | { | 3383 | { |
3384 | trace_ext4_readpage(page); | ||
3463 | return mpage_readpage(page, ext4_get_block); | 3385 | return mpage_readpage(page, ext4_get_block); |
3464 | } | 3386 | } |
3465 | 3387 | ||
@@ -3494,6 +3416,8 @@ static void ext4_invalidatepage(struct page *page, unsigned long offset) | |||
3494 | { | 3416 | { |
3495 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | 3417 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
3496 | 3418 | ||
3419 | trace_ext4_invalidatepage(page, offset); | ||
3420 | |||
3497 | /* | 3421 | /* |
3498 | * free any io_end structure allocated for buffers to be discarded | 3422 | * free any io_end structure allocated for buffers to be discarded |
3499 | */ | 3423 | */ |
@@ -3515,6 +3439,8 @@ static int ext4_releasepage(struct page *page, gfp_t wait) | |||
3515 | { | 3439 | { |
3516 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | 3440 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
3517 | 3441 | ||
3442 | trace_ext4_releasepage(page); | ||
3443 | |||
3518 | WARN_ON(PageChecked(page)); | 3444 | WARN_ON(PageChecked(page)); |
3519 | if (!page_has_buffers(page)) | 3445 | if (!page_has_buffers(page)) |
3520 | return 0; | 3446 | return 0; |
@@ -3873,11 +3799,16 @@ static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |||
3873 | { | 3799 | { |
3874 | struct file *file = iocb->ki_filp; | 3800 | struct file *file = iocb->ki_filp; |
3875 | struct inode *inode = file->f_mapping->host; | 3801 | struct inode *inode = file->f_mapping->host; |
3802 | ssize_t ret; | ||
3876 | 3803 | ||
3804 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); | ||
3877 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | 3805 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3878 | return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); | 3806 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3879 | 3807 | else | |
3880 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | 3808 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3809 | trace_ext4_direct_IO_exit(inode, offset, | ||
3810 | iov_length(iov, nr_segs), rw, ret); | ||
3811 | return ret; | ||
3881 | } | 3812 | } |
3882 | 3813 | ||
3883 | /* | 3814 | /* |
@@ -4173,6 +4104,9 @@ no_top: | |||
4173 | * | 4104 | * |
4174 | * We release `count' blocks on disk, but (last - first) may be greater | 4105 | * We release `count' blocks on disk, but (last - first) may be greater |
4175 | * than `count' because there can be holes in there. | 4106 | * than `count' because there can be holes in there. |
4107 | * | ||
4108 | * Return 0 on success, 1 on invalid block range | ||
4109 | * and < 0 on fatal error. | ||
4176 | */ | 4110 | */ |
4177 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, | 4111 | static int ext4_clear_blocks(handle_t *handle, struct inode *inode, |
4178 | struct buffer_head *bh, | 4112 | struct buffer_head *bh, |
@@ -4199,33 +4133,32 @@ static int ext4_clear_blocks(handle_t *handle, struct inode *inode, | |||
4199 | if (bh) { | 4133 | if (bh) { |
4200 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | 4134 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
4201 | err = ext4_handle_dirty_metadata(handle, inode, bh); | 4135 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
4202 | if (unlikely(err)) { | 4136 | if (unlikely(err)) |
4203 | ext4_std_error(inode->i_sb, err); | 4137 | goto out_err; |
4204 | return 1; | ||
4205 | } | ||
4206 | } | 4138 | } |
4207 | err = ext4_mark_inode_dirty(handle, inode); | 4139 | err = ext4_mark_inode_dirty(handle, inode); |
4208 | if (unlikely(err)) { | 4140 | if (unlikely(err)) |
4209 | ext4_std_error(inode->i_sb, err); | 4141 | goto out_err; |
4210 | return 1; | ||
4211 | } | ||
4212 | err = ext4_truncate_restart_trans(handle, inode, | 4142 | err = ext4_truncate_restart_trans(handle, inode, |
4213 | blocks_for_truncate(inode)); | 4143 | blocks_for_truncate(inode)); |
4214 | if (unlikely(err)) { | 4144 | if (unlikely(err)) |
4215 | ext4_std_error(inode->i_sb, err); | 4145 | goto out_err; |
4216 | return 1; | ||
4217 | } | ||
4218 | if (bh) { | 4146 | if (bh) { |
4219 | BUFFER_TRACE(bh, "retaking write access"); | 4147 | BUFFER_TRACE(bh, "retaking write access"); |
4220 | ext4_journal_get_write_access(handle, bh); | 4148 | err = ext4_journal_get_write_access(handle, bh); |
4149 | if (unlikely(err)) | ||
4150 | goto out_err; | ||
4221 | } | 4151 | } |
4222 | } | 4152 | } |
4223 | 4153 | ||
4224 | for (p = first; p < last; p++) | 4154 | for (p = first; p < last; p++) |
4225 | *p = 0; | 4155 | *p = 0; |
4226 | 4156 | ||
4227 | ext4_free_blocks(handle, inode, 0, block_to_free, count, flags); | 4157 | ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags); |
4228 | return 0; | 4158 | return 0; |
4159 | out_err: | ||
4160 | ext4_std_error(inode->i_sb, err); | ||
4161 | return err; | ||
4229 | } | 4162 | } |
4230 | 4163 | ||
4231 | /** | 4164 | /** |
@@ -4259,7 +4192,7 @@ static void ext4_free_data(handle_t *handle, struct inode *inode, | |||
4259 | ext4_fsblk_t nr; /* Current block # */ | 4192 | ext4_fsblk_t nr; /* Current block # */ |
4260 | __le32 *p; /* Pointer into inode/ind | 4193 | __le32 *p; /* Pointer into inode/ind |
4261 | for current block */ | 4194 | for current block */ |
4262 | int err; | 4195 | int err = 0; |
4263 | 4196 | ||
4264 | if (this_bh) { /* For indirect block */ | 4197 | if (this_bh) { /* For indirect block */ |
4265 | BUFFER_TRACE(this_bh, "get_write_access"); | 4198 | BUFFER_TRACE(this_bh, "get_write_access"); |
@@ -4281,9 +4214,10 @@ static void ext4_free_data(handle_t *handle, struct inode *inode, | |||
4281 | } else if (nr == block_to_free + count) { | 4214 | } else if (nr == block_to_free + count) { |
4282 | count++; | 4215 | count++; |
4283 | } else { | 4216 | } else { |
4284 | if (ext4_clear_blocks(handle, inode, this_bh, | 4217 | err = ext4_clear_blocks(handle, inode, this_bh, |
4285 | block_to_free, count, | 4218 | block_to_free, count, |
4286 | block_to_free_p, p)) | 4219 | block_to_free_p, p); |
4220 | if (err) | ||
4287 | break; | 4221 | break; |
4288 | block_to_free = nr; | 4222 | block_to_free = nr; |
4289 | block_to_free_p = p; | 4223 | block_to_free_p = p; |
@@ -4292,9 +4226,12 @@ static void ext4_free_data(handle_t *handle, struct inode *inode, | |||
4292 | } | 4226 | } |
4293 | } | 4227 | } |
4294 | 4228 | ||
4295 | if (count > 0) | 4229 | if (!err && count > 0) |
4296 | ext4_clear_blocks(handle, inode, this_bh, block_to_free, | 4230 | err = ext4_clear_blocks(handle, inode, this_bh, block_to_free, |
4297 | count, block_to_free_p, p); | 4231 | count, block_to_free_p, p); |
4232 | if (err < 0) | ||
4233 | /* fatal error */ | ||
4234 | return; | ||
4298 | 4235 | ||
4299 | if (this_bh) { | 4236 | if (this_bh) { |
4300 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); | 4237 | BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata"); |
@@ -4412,7 +4349,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode, | |||
4412 | * transaction where the data blocks are | 4349 | * transaction where the data blocks are |
4413 | * actually freed. | 4350 | * actually freed. |
4414 | */ | 4351 | */ |
4415 | ext4_free_blocks(handle, inode, 0, nr, 1, | 4352 | ext4_free_blocks(handle, inode, NULL, nr, 1, |
4416 | EXT4_FREE_BLOCKS_METADATA| | 4353 | EXT4_FREE_BLOCKS_METADATA| |
4417 | EXT4_FREE_BLOCKS_FORGET); | 4354 | EXT4_FREE_BLOCKS_FORGET); |
4418 | 4355 | ||
@@ -4496,6 +4433,8 @@ void ext4_truncate(struct inode *inode) | |||
4496 | ext4_lblk_t last_block; | 4433 | ext4_lblk_t last_block; |
4497 | unsigned blocksize = inode->i_sb->s_blocksize; | 4434 | unsigned blocksize = inode->i_sb->s_blocksize; |
4498 | 4435 | ||
4436 | trace_ext4_truncate_enter(inode); | ||
4437 | |||
4499 | if (!ext4_can_truncate(inode)) | 4438 | if (!ext4_can_truncate(inode)) |
4500 | return; | 4439 | return; |
4501 | 4440 | ||
@@ -4506,6 +4445,7 @@ void ext4_truncate(struct inode *inode) | |||
4506 | 4445 | ||
4507 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | 4446 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
4508 | ext4_ext_truncate(inode); | 4447 | ext4_ext_truncate(inode); |
4448 | trace_ext4_truncate_exit(inode); | ||
4509 | return; | 4449 | return; |
4510 | } | 4450 | } |
4511 | 4451 | ||
@@ -4635,6 +4575,7 @@ out_stop: | |||
4635 | ext4_orphan_del(handle, inode); | 4575 | ext4_orphan_del(handle, inode); |
4636 | 4576 | ||
4637 | ext4_journal_stop(handle); | 4577 | ext4_journal_stop(handle); |
4578 | trace_ext4_truncate_exit(inode); | ||
4638 | } | 4579 | } |
4639 | 4580 | ||
4640 | /* | 4581 | /* |
@@ -4766,6 +4707,7 @@ make_io: | |||
4766 | * has in-inode xattrs, or we don't have this inode in memory. | 4707 | * has in-inode xattrs, or we don't have this inode in memory. |
4767 | * Read the block from disk. | 4708 | * Read the block from disk. |
4768 | */ | 4709 | */ |
4710 | trace_ext4_load_inode(inode); | ||
4769 | get_bh(bh); | 4711 | get_bh(bh); |
4770 | bh->b_end_io = end_buffer_read_sync; | 4712 | bh->b_end_io = end_buffer_read_sync; |
4771 | submit_bh(READ_META, bh); | 4713 | submit_bh(READ_META, bh); |
@@ -4871,7 +4813,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino) | |||
4871 | return inode; | 4813 | return inode; |
4872 | 4814 | ||
4873 | ei = EXT4_I(inode); | 4815 | ei = EXT4_I(inode); |
4874 | iloc.bh = 0; | 4816 | iloc.bh = NULL; |
4875 | 4817 | ||
4876 | ret = __ext4_get_inode_loc(inode, &iloc, 0); | 4818 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4877 | if (ret < 0) | 4819 | if (ret < 0) |