diff options
| author | Mingming Cao <cmm@us.ibm.com> | 2009-09-28 15:49:08 -0400 |
|---|---|---|
| committer | Theodore Ts'o <tytso@mit.edu> | 2009-09-28 15:49:08 -0400 |
| commit | 0031462b5b392f90d17f1d75abb795883c44e969 (patch) | |
| tree | e8323861b8dede0f3ddbfc8324d650bf1f4fd74b | |
| parent | 9f0ccfd8e07d61b413e6536ffa02fbf60d2e20d8 (diff) | |
ext4: Split uninitialized extents for direct I/O
When writing into an unitialized extent via direct I/O, and the direct
I/O doesn't exactly cover the unitialized extent, split the extent
into uninitialized and initialized extents before submitting the I/O.
This avoids needing to deal with an ENOSPC error in the end_io
callback that gets used for direct I/O.
When the IO is complete, the written extent will be marked as initialized.
Singed-Off-By: Mingming Cao <cmm@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
| -rw-r--r-- | fs/ext4/ext4.h | 22 | ||||
| -rw-r--r-- | fs/ext4/ext4_extents.h | 7 | ||||
| -rw-r--r-- | fs/ext4/extents.c | 423 | ||||
| -rw-r--r-- | fs/ext4/inode.c | 3 | ||||
| -rw-r--r-- | fs/ext4/migrate.c | 2 | ||||
| -rw-r--r-- | fs/ext4/move_extent.c | 4 |
6 files changed, 419 insertions, 42 deletions
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h index a58438e18d0b..2b4293aac162 100644 --- a/fs/ext4/ext4.h +++ b/fs/ext4/ext4.h | |||
| @@ -128,6 +128,15 @@ struct mpage_da_data { | |||
| 128 | int retval; | 128 | int retval; |
| 129 | }; | 129 | }; |
| 130 | 130 | ||
| 131 | typedef struct ext4_io_end { | ||
| 132 | struct inode *inode; /* file being written to */ | ||
| 133 | unsigned int flag; /* sync IO or AIO */ | ||
| 134 | int error; /* I/O error code */ | ||
| 135 | ext4_lblk_t offset; /* offset in the file */ | ||
| 136 | size_t size; /* size of the extent */ | ||
| 137 | struct work_struct work; /* data work queue */ | ||
| 138 | } ext4_io_end_t; | ||
| 139 | |||
| 131 | /* | 140 | /* |
| 132 | * Special inodes numbers | 141 | * Special inodes numbers |
| 133 | */ | 142 | */ |
| @@ -347,7 +356,16 @@ struct ext4_new_group_data { | |||
| 347 | /* Call ext4_da_update_reserve_space() after successfully | 356 | /* Call ext4_da_update_reserve_space() after successfully |
| 348 | allocating the blocks */ | 357 | allocating the blocks */ |
| 349 | #define EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE 0x0008 | 358 | #define EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE 0x0008 |
| 350 | 359 | /* caller is from the direct IO path, request to creation of an | |
| 360 | unitialized extents if not allocated, split the uninitialized | ||
| 361 | extent if blocks has been preallocated already*/ | ||
| 362 | #define EXT4_GET_BLOCKS_DIO 0x0010 | ||
| 363 | #define EXT4_GET_BLOCKS_CONVERT 0x0020 | ||
| 364 | #define EXT4_GET_BLOCKS_DIO_CREATE_EXT (EXT4_GET_BLOCKS_DIO|\ | ||
| 365 | EXT4_GET_BLOCKS_CREATE_UNINIT_EXT) | ||
| 366 | /* Convert extent to initialized after direct IO complete */ | ||
| 367 | #define EXT4_GET_BLOCKS_DIO_CONVERT_EXT (EXT4_GET_BLOCKS_CONVERT|\ | ||
| 368 | EXT4_GET_BLOCKS_DIO_CREATE_EXT) | ||
| 351 | 369 | ||
| 352 | /* | 370 | /* |
| 353 | * ioctl commands | 371 | * ioctl commands |
| @@ -1700,6 +1718,8 @@ extern void ext4_ext_init(struct super_block *); | |||
| 1700 | extern void ext4_ext_release(struct super_block *); | 1718 | extern void ext4_ext_release(struct super_block *); |
| 1701 | extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset, | 1719 | extern long ext4_fallocate(struct inode *inode, int mode, loff_t offset, |
| 1702 | loff_t len); | 1720 | loff_t len); |
| 1721 | extern int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | ||
| 1722 | loff_t len); | ||
| 1703 | extern int ext4_get_blocks(handle_t *handle, struct inode *inode, | 1723 | extern int ext4_get_blocks(handle_t *handle, struct inode *inode, |
| 1704 | sector_t block, unsigned int max_blocks, | 1724 | sector_t block, unsigned int max_blocks, |
| 1705 | struct buffer_head *bh, int flags); | 1725 | struct buffer_head *bh, int flags); |
diff --git a/fs/ext4/ext4_extents.h b/fs/ext4/ext4_extents.h index 61652f1d15e6..2ca686454e87 100644 --- a/fs/ext4/ext4_extents.h +++ b/fs/ext4/ext4_extents.h | |||
| @@ -220,6 +220,11 @@ static inline int ext4_ext_get_actual_len(struct ext4_extent *ext) | |||
| 220 | (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN)); | 220 | (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN)); |
| 221 | } | 221 | } |
| 222 | 222 | ||
| 223 | static inline void ext4_ext_mark_initialized(struct ext4_extent *ext) | ||
| 224 | { | ||
| 225 | ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext)); | ||
| 226 | } | ||
| 227 | |||
| 223 | extern int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks); | 228 | extern int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks); |
| 224 | extern ext4_fsblk_t ext_pblock(struct ext4_extent *ex); | 229 | extern ext4_fsblk_t ext_pblock(struct ext4_extent *ex); |
| 225 | extern ext4_fsblk_t idx_pblock(struct ext4_extent_idx *); | 230 | extern ext4_fsblk_t idx_pblock(struct ext4_extent_idx *); |
| @@ -235,7 +240,7 @@ extern int ext4_ext_try_to_merge(struct inode *inode, | |||
| 235 | struct ext4_ext_path *path, | 240 | struct ext4_ext_path *path, |
| 236 | struct ext4_extent *); | 241 | struct ext4_extent *); |
| 237 | extern unsigned int ext4_ext_check_overlap(struct inode *, struct ext4_extent *, struct ext4_ext_path *); | 242 | extern unsigned int ext4_ext_check_overlap(struct inode *, struct ext4_extent *, struct ext4_ext_path *); |
| 238 | extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *); | 243 | extern int ext4_ext_insert_extent(handle_t *, struct inode *, struct ext4_ext_path *, struct ext4_extent *, int); |
| 239 | extern int ext4_ext_walk_space(struct inode *, ext4_lblk_t, ext4_lblk_t, | 244 | extern int ext4_ext_walk_space(struct inode *, ext4_lblk_t, ext4_lblk_t, |
| 240 | ext_prepare_callback, void *); | 245 | ext_prepare_callback, void *); |
| 241 | extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t, | 246 | extern struct ext4_ext_path *ext4_ext_find_extent(struct inode *, ext4_lblk_t, |
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c index 7a3832577923..a38e651c004e 100644 --- a/fs/ext4/extents.c +++ b/fs/ext4/extents.c | |||
| @@ -723,7 +723,7 @@ err: | |||
| 723 | * insert new index [@logical;@ptr] into the block at @curp; | 723 | * insert new index [@logical;@ptr] into the block at @curp; |
| 724 | * check where to insert: before @curp or after @curp | 724 | * check where to insert: before @curp or after @curp |
| 725 | */ | 725 | */ |
| 726 | static int ext4_ext_insert_index(handle_t *handle, struct inode *inode, | 726 | int ext4_ext_insert_index(handle_t *handle, struct inode *inode, |
| 727 | struct ext4_ext_path *curp, | 727 | struct ext4_ext_path *curp, |
| 728 | int logical, ext4_fsblk_t ptr) | 728 | int logical, ext4_fsblk_t ptr) |
| 729 | { | 729 | { |
| @@ -1586,7 +1586,7 @@ out: | |||
| 1586 | */ | 1586 | */ |
| 1587 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | 1587 | int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, |
| 1588 | struct ext4_ext_path *path, | 1588 | struct ext4_ext_path *path, |
| 1589 | struct ext4_extent *newext) | 1589 | struct ext4_extent *newext, int flag) |
| 1590 | { | 1590 | { |
| 1591 | struct ext4_extent_header *eh; | 1591 | struct ext4_extent_header *eh; |
| 1592 | struct ext4_extent *ex, *fex; | 1592 | struct ext4_extent *ex, *fex; |
| @@ -1602,7 +1602,8 @@ int ext4_ext_insert_extent(handle_t *handle, struct inode *inode, | |||
| 1602 | BUG_ON(path[depth].p_hdr == NULL); | 1602 | BUG_ON(path[depth].p_hdr == NULL); |
| 1603 | 1603 | ||
| 1604 | /* try to insert block into found extent and return */ | 1604 | /* try to insert block into found extent and return */ |
| 1605 | if (ex && ext4_can_extents_be_merged(inode, ex, newext)) { | 1605 | if (ex && (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT) |
| 1606 | && ext4_can_extents_be_merged(inode, ex, newext)) { | ||
| 1606 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", | 1607 | ext_debug("append [%d]%d block to %d:[%d]%d (from %llu)\n", |
| 1607 | ext4_ext_is_uninitialized(newext), | 1608 | ext4_ext_is_uninitialized(newext), |
| 1608 | ext4_ext_get_actual_len(newext), | 1609 | ext4_ext_get_actual_len(newext), |
| @@ -1722,7 +1723,8 @@ has_space: | |||
| 1722 | 1723 | ||
| 1723 | merge: | 1724 | merge: |
| 1724 | /* try to merge extents to the right */ | 1725 | /* try to merge extents to the right */ |
| 1725 | ext4_ext_try_to_merge(inode, path, nearex); | 1726 | if (flag != EXT4_GET_BLOCKS_DIO_CREATE_EXT) |
| 1727 | ext4_ext_try_to_merge(inode, path, nearex); | ||
| 1726 | 1728 | ||
| 1727 | /* try to merge extents to the left */ | 1729 | /* try to merge extents to the left */ |
| 1728 | 1730 | ||
| @@ -2490,7 +2492,6 @@ static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex) | |||
| 2490 | } | 2492 | } |
| 2491 | 2493 | ||
| 2492 | #define EXT4_EXT_ZERO_LEN 7 | 2494 | #define EXT4_EXT_ZERO_LEN 7 |
| 2493 | |||
| 2494 | /* | 2495 | /* |
| 2495 | * This function is called by ext4_ext_get_blocks() if someone tries to write | 2496 | * This function is called by ext4_ext_get_blocks() if someone tries to write |
| 2496 | * to an uninitialized extent. It may result in splitting the uninitialized | 2497 | * to an uninitialized extent. It may result in splitting the uninitialized |
| @@ -2583,7 +2584,8 @@ static int ext4_ext_convert_to_initialized(handle_t *handle, | |||
| 2583 | ex3->ee_block = cpu_to_le32(iblock); | 2584 | ex3->ee_block = cpu_to_le32(iblock); |
| 2584 | ext4_ext_store_pblock(ex3, newblock); | 2585 | ext4_ext_store_pblock(ex3, newblock); |
| 2585 | ex3->ee_len = cpu_to_le16(allocated); | 2586 | ex3->ee_len = cpu_to_le16(allocated); |
| 2586 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | 2587 | err = ext4_ext_insert_extent(handle, inode, path, |
| 2588 | ex3, 0); | ||
| 2587 | if (err == -ENOSPC) { | 2589 | if (err == -ENOSPC) { |
| 2588 | err = ext4_ext_zeroout(inode, &orig_ex); | 2590 | err = ext4_ext_zeroout(inode, &orig_ex); |
| 2589 | if (err) | 2591 | if (err) |
| @@ -2639,7 +2641,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle, | |||
| 2639 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | 2641 | ext4_ext_store_pblock(ex3, newblock + max_blocks); |
| 2640 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | 2642 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); |
| 2641 | ext4_ext_mark_uninitialized(ex3); | 2643 | ext4_ext_mark_uninitialized(ex3); |
| 2642 | err = ext4_ext_insert_extent(handle, inode, path, ex3); | 2644 | err = ext4_ext_insert_extent(handle, inode, path, ex3, 0); |
| 2643 | if (err == -ENOSPC) { | 2645 | if (err == -ENOSPC) { |
| 2644 | err = ext4_ext_zeroout(inode, &orig_ex); | 2646 | err = ext4_ext_zeroout(inode, &orig_ex); |
| 2645 | if (err) | 2647 | if (err) |
| @@ -2757,7 +2759,7 @@ static int ext4_ext_convert_to_initialized(handle_t *handle, | |||
| 2757 | err = ext4_ext_dirty(handle, inode, path + depth); | 2759 | err = ext4_ext_dirty(handle, inode, path + depth); |
| 2758 | goto out; | 2760 | goto out; |
| 2759 | insert: | 2761 | insert: |
| 2760 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | 2762 | err = ext4_ext_insert_extent(handle, inode, path, &newex, 0); |
| 2761 | if (err == -ENOSPC) { | 2763 | if (err == -ENOSPC) { |
| 2762 | err = ext4_ext_zeroout(inode, &orig_ex); | 2764 | err = ext4_ext_zeroout(inode, &orig_ex); |
| 2763 | if (err) | 2765 | if (err) |
| @@ -2785,6 +2787,320 @@ fix_extent_len: | |||
| 2785 | } | 2787 | } |
| 2786 | 2788 | ||
| 2787 | /* | 2789 | /* |
| 2790 | * This function is called by ext4_ext_get_blocks() from | ||
| 2791 | * ext4_get_blocks_dio_write() when DIO to write | ||
| 2792 | * to an uninitialized extent. | ||
| 2793 | * | ||
| 2794 | * Writing to an uninitized extent may result in splitting the uninitialized | ||
| 2795 | * extent into multiple /intialized unintialized extents (up to three) | ||
| 2796 | * There are three possibilities: | ||
| 2797 | * a> There is no split required: Entire extent should be uninitialized | ||
| 2798 | * b> Splits in two extents: Write is happening at either end of the extent | ||
| 2799 | * c> Splits in three extents: Somone is writing in middle of the extent | ||
| 2800 | * | ||
| 2801 | * One of more index blocks maybe needed if the extent tree grow after | ||
| 2802 | * the unintialized extent split. To prevent ENOSPC occur at the IO | ||
| 2803 | * complete, we need to split the uninitialized extent before DIO submit | ||
| 2804 | * the IO. The uninitilized extent called at this time will be split | ||
| 2805 | * into three uninitialized extent(at most). After IO complete, the part | ||
| 2806 | * being filled will be convert to initialized by the end_io callback function | ||
| 2807 | * via ext4_convert_unwritten_extents(). | ||
| 2808 | */ | ||
| 2809 | static int ext4_split_unwritten_extents(handle_t *handle, | ||
| 2810 | struct inode *inode, | ||
| 2811 | struct ext4_ext_path *path, | ||
| 2812 | ext4_lblk_t iblock, | ||
| 2813 | unsigned int max_blocks, | ||
| 2814 | int flags) | ||
| 2815 | { | ||
| 2816 | struct ext4_extent *ex, newex, orig_ex; | ||
| 2817 | struct ext4_extent *ex1 = NULL; | ||
| 2818 | struct ext4_extent *ex2 = NULL; | ||
| 2819 | struct ext4_extent *ex3 = NULL; | ||
| 2820 | struct ext4_extent_header *eh; | ||
| 2821 | ext4_lblk_t ee_block; | ||
| 2822 | unsigned int allocated, ee_len, depth; | ||
| 2823 | ext4_fsblk_t newblock; | ||
| 2824 | int err = 0; | ||
| 2825 | int ret = 0; | ||
| 2826 | |||
| 2827 | ext_debug("ext4_split_unwritten_extents: inode %lu," | ||
| 2828 | "iblock %llu, max_blocks %u\n", inode->i_ino, | ||
| 2829 | (unsigned long long)iblock, max_blocks); | ||
| 2830 | depth = ext_depth(inode); | ||
| 2831 | eh = path[depth].p_hdr; | ||
| 2832 | ex = path[depth].p_ext; | ||
| 2833 | ee_block = le32_to_cpu(ex->ee_block); | ||
| 2834 | ee_len = ext4_ext_get_actual_len(ex); | ||
| 2835 | allocated = ee_len - (iblock - ee_block); | ||
| 2836 | newblock = iblock - ee_block + ext_pblock(ex); | ||
| 2837 | ex2 = ex; | ||
| 2838 | orig_ex.ee_block = ex->ee_block; | ||
| 2839 | orig_ex.ee_len = cpu_to_le16(ee_len); | ||
| 2840 | ext4_ext_store_pblock(&orig_ex, ext_pblock(ex)); | ||
| 2841 | |||
| 2842 | /* | ||
| 2843 | * if the entire unintialized extent length less than | ||
| 2844 | * the size of extent to write, there is no need to split | ||
| 2845 | * uninitialized extent | ||
| 2846 | */ | ||
| 2847 | if (allocated <= max_blocks) | ||
| 2848 | return ret; | ||
| 2849 | |||
| 2850 | err = ext4_ext_get_access(handle, inode, path + depth); | ||
| 2851 | if (err) | ||
| 2852 | goto out; | ||
| 2853 | /* ex1: ee_block to iblock - 1 : uninitialized */ | ||
| 2854 | if (iblock > ee_block) { | ||
| 2855 | ex1 = ex; | ||
| 2856 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | ||
| 2857 | ext4_ext_mark_uninitialized(ex1); | ||
| 2858 | ex2 = &newex; | ||
| 2859 | } | ||
| 2860 | /* | ||
| 2861 | * for sanity, update the length of the ex2 extent before | ||
| 2862 | * we insert ex3, if ex1 is NULL. This is to avoid temporary | ||
| 2863 | * overlap of blocks. | ||
| 2864 | */ | ||
| 2865 | if (!ex1 && allocated > max_blocks) | ||
| 2866 | ex2->ee_len = cpu_to_le16(max_blocks); | ||
| 2867 | /* ex3: to ee_block + ee_len : uninitialised */ | ||
| 2868 | if (allocated > max_blocks) { | ||
| 2869 | unsigned int newdepth; | ||
| 2870 | ex3 = &newex; | ||
| 2871 | ex3->ee_block = cpu_to_le32(iblock + max_blocks); | ||
| 2872 | ext4_ext_store_pblock(ex3, newblock + max_blocks); | ||
| 2873 | ex3->ee_len = cpu_to_le16(allocated - max_blocks); | ||
| 2874 | ext4_ext_mark_uninitialized(ex3); | ||
| 2875 | err = ext4_ext_insert_extent(handle, inode, path, ex3, flags); | ||
| 2876 | if (err == -ENOSPC) { | ||
| 2877 | err = ext4_ext_zeroout(inode, &orig_ex); | ||
| 2878 | if (err) | ||
| 2879 | goto fix_extent_len; | ||
| 2880 | /* update the extent length and mark as initialized */ | ||
| 2881 | ex->ee_block = orig_ex.ee_block; | ||
| 2882 | ex->ee_len = orig_ex.ee_len; | ||
| 2883 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | ||
| 2884 | ext4_ext_dirty(handle, inode, path + depth); | ||
| 2885 | /* zeroed the full extent */ | ||
| 2886 | /* blocks available from iblock */ | ||
| 2887 | return allocated; | ||
| 2888 | |||
| 2889 | } else if (err) | ||
| 2890 | goto fix_extent_len; | ||
| 2891 | /* | ||
| 2892 | * The depth, and hence eh & ex might change | ||
| 2893 | * as part of the insert above. | ||
| 2894 | */ | ||
| 2895 | newdepth = ext_depth(inode); | ||
| 2896 | /* | ||
| 2897 | * update the extent length after successful insert of the | ||
| 2898 | * split extent | ||
| 2899 | */ | ||
| 2900 | orig_ex.ee_len = cpu_to_le16(ee_len - | ||
| 2901 | ext4_ext_get_actual_len(ex3)); | ||
| 2902 | depth = newdepth; | ||
| 2903 | ext4_ext_drop_refs(path); | ||
| 2904 | path = ext4_ext_find_extent(inode, iblock, path); | ||
| 2905 | if (IS_ERR(path)) { | ||
| 2906 | err = PTR_ERR(path); | ||
| 2907 | goto out; | ||
| 2908 | } | ||
| 2909 | eh = path[depth].p_hdr; | ||
| 2910 | ex = path[depth].p_ext; | ||
| 2911 | if (ex2 != &newex) | ||
| 2912 | ex2 = ex; | ||
| 2913 | |||
| 2914 | err = ext4_ext_get_access(handle, inode, path + depth); | ||
| 2915 | if (err) | ||
| 2916 | goto out; | ||
| 2917 | |||
| 2918 | allocated = max_blocks; | ||
| 2919 | } | ||
| 2920 | /* | ||
| 2921 | * If there was a change of depth as part of the | ||
| 2922 | * insertion of ex3 above, we need to update the length | ||
| 2923 | * of the ex1 extent again here | ||
| 2924 | */ | ||
| 2925 | if (ex1 && ex1 != ex) { | ||
| 2926 | ex1 = ex; | ||
| 2927 | ex1->ee_len = cpu_to_le16(iblock - ee_block); | ||
| 2928 | ext4_ext_mark_uninitialized(ex1); | ||
| 2929 | ex2 = &newex; | ||
| 2930 | } | ||
| 2931 | /* | ||
| 2932 | * ex2: iblock to iblock + maxblocks-1 : to be direct IO written, | ||
| 2933 | * uninitialised still. | ||
| 2934 | */ | ||
| 2935 | ex2->ee_block = cpu_to_le32(iblock); | ||
| 2936 | ext4_ext_store_pblock(ex2, newblock); | ||
| 2937 | ex2->ee_len = cpu_to_le16(allocated); | ||
| 2938 | ext4_ext_mark_uninitialized(ex2); | ||
| 2939 | if (ex2 != ex) | ||
| 2940 | goto insert; | ||
| 2941 | /* Mark modified extent as dirty */ | ||
| 2942 | err = ext4_ext_dirty(handle, inode, path + depth); | ||
| 2943 | ext_debug("out here\n"); | ||
| 2944 | goto out; | ||
| 2945 | insert: | ||
| 2946 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); | ||
| 2947 | if (err == -ENOSPC) { | ||
| 2948 | err = ext4_ext_zeroout(inode, &orig_ex); | ||
| 2949 | if (err) | ||
| 2950 | goto fix_extent_len; | ||
| 2951 | /* update the extent length and mark as initialized */ | ||
| 2952 | ex->ee_block = orig_ex.ee_block; | ||
| 2953 | ex->ee_len = orig_ex.ee_len; | ||
| 2954 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | ||
| 2955 | ext4_ext_dirty(handle, inode, path + depth); | ||
| 2956 | /* zero out the first half */ | ||
| 2957 | return allocated; | ||
| 2958 | } else if (err) | ||
| 2959 | goto fix_extent_len; | ||
| 2960 | out: | ||
| 2961 | ext4_ext_show_leaf(inode, path); | ||
| 2962 | return err ? err : allocated; | ||
| 2963 | |||
| 2964 | fix_extent_len: | ||
| 2965 | ex->ee_block = orig_ex.ee_block; | ||
| 2966 | ex->ee_len = orig_ex.ee_len; | ||
| 2967 | ext4_ext_store_pblock(ex, ext_pblock(&orig_ex)); | ||
| 2968 | ext4_ext_mark_uninitialized(ex); | ||
| 2969 | ext4_ext_dirty(handle, inode, path + depth); | ||
| 2970 | return err; | ||
| 2971 | } | ||
| 2972 | static int ext4_convert_unwritten_extents_dio(handle_t *handle, | ||
| 2973 | struct inode *inode, | ||
| 2974 | struct ext4_ext_path *path) | ||
| 2975 | { | ||
| 2976 | struct ext4_extent *ex; | ||
| 2977 | struct ext4_extent_header *eh; | ||
| 2978 | int depth; | ||
| 2979 | int err = 0; | ||
| 2980 | int ret = 0; | ||
| 2981 | |||
| 2982 | depth = ext_depth(inode); | ||
| 2983 | eh = path[depth].p_hdr; | ||
| 2984 | ex = path[depth].p_ext; | ||
| 2985 | |||
| 2986 | err = ext4_ext_get_access(handle, inode, path + depth); | ||
| 2987 | if (err) | ||
| 2988 | goto out; | ||
| 2989 | /* first mark the extent as initialized */ | ||
| 2990 | ext4_ext_mark_initialized(ex); | ||
| 2991 | |||
| 2992 | /* | ||
| 2993 | * We have to see if it can be merged with the extent | ||
| 2994 | * on the left. | ||
| 2995 | */ | ||
| 2996 | if (ex > EXT_FIRST_EXTENT(eh)) { | ||
| 2997 | /* | ||
| 2998 | * To merge left, pass "ex - 1" to try_to_merge(), | ||
| 2999 | * since it merges towards right _only_. | ||
| 3000 | */ | ||
| 3001 | ret = ext4_ext_try_to_merge(inode, path, ex - 1); | ||
| 3002 | if (ret) { | ||
| 3003 | err = ext4_ext_correct_indexes(handle, inode, path); | ||
| 3004 | if (err) | ||
| 3005 | goto out; | ||
| 3006 | depth = ext_depth(inode); | ||
| 3007 | ex--; | ||
| 3008 | } | ||
| 3009 | } | ||
| 3010 | /* | ||
| 3011 | * Try to Merge towards right. | ||
| 3012 | */ | ||
| 3013 | ret = ext4_ext_try_to_merge(inode, path, ex); | ||
| 3014 | if (ret) { | ||
| 3015 | err = ext4_ext_correct_indexes(handle, inode, path); | ||
| 3016 | if (err) | ||
| 3017 | goto out; | ||
| 3018 | depth = ext_depth(inode); | ||
| 3019 | } | ||
| 3020 | /* Mark modified extent as dirty */ | ||
| 3021 | err = ext4_ext_dirty(handle, inode, path + depth); | ||
| 3022 | out: | ||
| 3023 | ext4_ext_show_leaf(inode, path); | ||
| 3024 | return err; | ||
| 3025 | } | ||
| 3026 | |||
| 3027 | static int | ||
| 3028 | ext4_ext_handle_uninitialized_extents(handle_t *handle, struct inode *inode, | ||
| 3029 | ext4_lblk_t iblock, unsigned int max_blocks, | ||
| 3030 | struct ext4_ext_path *path, int flags, | ||
| 3031 | unsigned int allocated, struct buffer_head *bh_result, | ||
| 3032 | ext4_fsblk_t newblock) | ||
| 3033 | { | ||
| 3034 | int ret = 0; | ||
| 3035 | int err = 0; | ||
| 3036 | |||
| 3037 | ext_debug("ext4_ext_handle_uninitialized_extents: inode %lu, logical" | ||
| 3038 | "block %llu, max_blocks %u, flags %d, allocated %u", | ||
| 3039 | inode->i_ino, (unsigned long long)iblock, max_blocks, | ||
| 3040 | flags, allocated); | ||
| 3041 | ext4_ext_show_leaf(inode, path); | ||
| 3042 | |||
| 3043 | /* DIO get_block() before submit the IO, split the extent */ | ||
| 3044 | if (flags == EXT4_GET_BLOCKS_DIO_CREATE_EXT) { | ||
| 3045 | ret = ext4_split_unwritten_extents(handle, | ||
| 3046 | inode, path, iblock, | ||
| 3047 | max_blocks, flags); | ||
| 3048 | goto out; | ||
| 3049 | } | ||
| 3050 | /* DIO end_io complete, convert the filled extent to written */ | ||
| 3051 | if (flags == EXT4_GET_BLOCKS_DIO_CONVERT_EXT) { | ||
| 3052 | ret = ext4_convert_unwritten_extents_dio(handle, inode, | ||
| 3053 | path); | ||
| 3054 | goto out2; | ||
| 3055 | } | ||
| 3056 | /* buffered IO case */ | ||
| 3057 | /* | ||
| 3058 | * repeat fallocate creation request | ||
| 3059 | * we already have an unwritten extent | ||
| 3060 | */ | ||
| 3061 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | ||
| 3062 | goto map_out; | ||
| 3063 | |||
| 3064 | /* buffered READ or buffered write_begin() lookup */ | ||
| 3065 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | ||
| 3066 | /* | ||
| 3067 | * We have blocks reserved already. We | ||
| 3068 | * return allocated blocks so that delalloc | ||
| 3069 | * won't do block reservation for us. But | ||
| 3070 | * the buffer head will be unmapped so that | ||
| 3071 | * a read from the block returns 0s. | ||
| 3072 | */ | ||
| 3073 | set_buffer_unwritten(bh_result); | ||
| 3074 | goto out1; | ||
| 3075 | } | ||
| 3076 | |||
| 3077 | /* buffered write, writepage time, convert*/ | ||
| 3078 | ret = ext4_ext_convert_to_initialized(handle, inode, | ||
| 3079 | path, iblock, | ||
| 3080 | max_blocks); | ||
| 3081 | out: | ||
| 3082 | if (ret <= 0) { | ||
| 3083 | err = ret; | ||
| 3084 | goto out2; | ||
| 3085 | } else | ||
| 3086 | allocated = ret; | ||
| 3087 | set_buffer_new(bh_result); | ||
| 3088 | map_out: | ||
| 3089 | set_buffer_mapped(bh_result); | ||
| 3090 | out1: | ||
| 3091 | if (allocated > max_blocks) | ||
| 3092 | allocated = max_blocks; | ||
| 3093 | ext4_ext_show_leaf(inode, path); | ||
| 3094 | bh_result->b_bdev = inode->i_sb->s_bdev; | ||
| 3095 | bh_result->b_blocknr = newblock; | ||
| 3096 | out2: | ||
| 3097 | if (path) { | ||
| 3098 | ext4_ext_drop_refs(path); | ||
| 3099 | kfree(path); | ||
| 3100 | } | ||
| 3101 | return err ? err : allocated; | ||
| 3102 | } | ||
| 3103 | /* | ||
| 2788 | * Block allocation/map/preallocation routine for extents based files | 3104 | * Block allocation/map/preallocation routine for extents based files |
| 2789 | * | 3105 | * |
| 2790 | * | 3106 | * |
| @@ -2889,33 +3205,10 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
| 2889 | EXT4_EXT_CACHE_EXTENT); | 3205 | EXT4_EXT_CACHE_EXTENT); |
| 2890 | goto out; | 3206 | goto out; |
| 2891 | } | 3207 | } |
| 2892 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) | 3208 | ret = ext4_ext_handle_uninitialized_extents(handle, |
| 2893 | goto out; | 3209 | inode, iblock, max_blocks, path, |
| 2894 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) { | 3210 | flags, allocated, bh_result, newblock); |
| 2895 | if (allocated > max_blocks) | 3211 | return ret; |
| 2896 | allocated = max_blocks; | ||
| 2897 | /* | ||
| 2898 | * We have blocks reserved already. We | ||
| 2899 | * return allocated blocks so that delalloc | ||
| 2900 | * won't do block reservation for us. But | ||
| 2901 | * the buffer head will be unmapped so that | ||
| 2902 | * a read from the block returns 0s. | ||
| 2903 | */ | ||
| 2904 | set_buffer_unwritten(bh_result); | ||
| 2905 | bh_result->b_bdev = inode->i_sb->s_bdev; | ||
| 2906 | bh_result->b_blocknr = newblock; | ||
| 2907 | goto out2; | ||
| 2908 | } | ||
| 2909 | |||
| 2910 | ret = ext4_ext_convert_to_initialized(handle, inode, | ||
| 2911 | path, iblock, | ||
| 2912 | max_blocks); | ||
| 2913 | if (ret <= 0) { | ||
| 2914 | err = ret; | ||
| 2915 | goto out2; | ||
| 2916 | } else | ||
| 2917 | allocated = ret; | ||
| 2918 | goto outnew; | ||
| 2919 | } | 3212 | } |
| 2920 | } | 3213 | } |
| 2921 | 3214 | ||
| @@ -2988,7 +3281,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
| 2988 | newex.ee_len = cpu_to_le16(ar.len); | 3281 | newex.ee_len = cpu_to_le16(ar.len); |
| 2989 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */ | 3282 | if (flags & EXT4_GET_BLOCKS_UNINIT_EXT) /* Mark uninitialized */ |
| 2990 | ext4_ext_mark_uninitialized(&newex); | 3283 | ext4_ext_mark_uninitialized(&newex); |
| 2991 | err = ext4_ext_insert_extent(handle, inode, path, &newex); | 3284 | err = ext4_ext_insert_extent(handle, inode, path, &newex, flags); |
| 2992 | if (err) { | 3285 | if (err) { |
| 2993 | /* free data blocks we just allocated */ | 3286 | /* free data blocks we just allocated */ |
| 2994 | /* not a good idea to call discard here directly, | 3287 | /* not a good idea to call discard here directly, |
| @@ -3002,7 +3295,6 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode, | |||
| 3002 | /* previous routine could use block we allocated */ | 3295 | /* previous routine could use block we allocated */ |
| 3003 | newblock = ext_pblock(&newex); | 3296 | newblock = ext_pblock(&newex); |
| 3004 | allocated = ext4_ext_get_actual_len(&newex); | 3297 | allocated = ext4_ext_get_actual_len(&newex); |
| 3005 | outnew: | ||
| 3006 | set_buffer_new(bh_result); | 3298 | set_buffer_new(bh_result); |
| 3007 | 3299 | ||
| 3008 | /* Cache only when it is _not_ an uninitialized extent */ | 3300 | /* Cache only when it is _not_ an uninitialized extent */ |
| @@ -3201,6 +3493,63 @@ retry: | |||
| 3201 | } | 3493 | } |
| 3202 | 3494 | ||
| 3203 | /* | 3495 | /* |
| 3496 | * This function convert a range of blocks to written extents | ||
| 3497 | * The caller of this function will pass the start offset and the size. | ||
| 3498 | * all unwritten extents within this range will be converted to | ||
| 3499 | * written extents. | ||
| 3500 | * | ||
| 3501 | * This function is called from the direct IO end io call back | ||
| 3502 | * function, to convert the fallocated extents after IO is completed. | ||
| 3503 | */ | ||
| 3504 | int ext4_convert_unwritten_extents(struct inode *inode, loff_t offset, | ||
| 3505 | loff_t len) | ||
| 3506 | { | ||
| 3507 | handle_t *handle; | ||
| 3508 | ext4_lblk_t block; | ||
| 3509 | unsigned int max_blocks; | ||
| 3510 | int ret = 0; | ||
| 3511 | int ret2 = 0; | ||
| 3512 | struct buffer_head map_bh; | ||
| 3513 | unsigned int credits, blkbits = inode->i_blkbits; | ||
| 3514 | |||
| 3515 | block = offset >> blkbits; | ||
| 3516 | /* | ||
| 3517 | * We can't just convert len to max_blocks because | ||
| 3518 | * If blocksize = 4096 offset = 3072 and len = 2048 | ||
| 3519 | */ | ||
| 3520 | max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits) | ||
| 3521 | - block; | ||
| 3522 | /* | ||
| 3523 | * credits to insert 1 extent into extent tree | ||
| 3524 | */ | ||
| 3525 | credits = ext4_chunk_trans_blocks(inode, max_blocks); | ||
| 3526 | while (ret >= 0 && ret < max_blocks) { | ||
| 3527 | block = block + ret; | ||
| 3528 | max_blocks = max_blocks - ret; | ||
| 3529 | handle = ext4_journal_start(inode, credits); | ||
| 3530 | if (IS_ERR(handle)) { | ||
| 3531 | ret = PTR_ERR(handle); | ||
| 3532 | break; | ||
| 3533 | } | ||
| 3534 | map_bh.b_state = 0; | ||
| 3535 | ret = ext4_get_blocks(handle, inode, block, | ||
| 3536 | max_blocks, &map_bh, | ||
| 3537 | EXT4_GET_BLOCKS_DIO_CONVERT_EXT); | ||
| 3538 | if (ret <= 0) { | ||
| 3539 | WARN_ON(ret <= 0); | ||
| 3540 | printk(KERN_ERR "%s: ext4_ext_get_blocks " | ||
| 3541 | "returned error inode#%lu, block=%u, " | ||
| 3542 | "max_blocks=%u", __func__, | ||
| 3543 | inode->i_ino, block, max_blocks); | ||
| 3544 | } | ||
| 3545 | ext4_mark_inode_dirty(handle, inode); | ||
| 3546 | ret2 = ext4_journal_stop(handle); | ||
| 3547 | if (ret <= 0 || ret2 ) | ||
| 3548 | break; | ||
| 3549 | } | ||
| 3550 | return ret > 0 ? ret2 : ret; | ||
| 3551 | } | ||
| 3552 | /* | ||
| 3204 | * Callback function called for each extent to gather FIEMAP information. | 3553 | * Callback function called for each extent to gather FIEMAP information. |
| 3205 | */ | 3554 | */ |
| 3206 | static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path, | 3555 | static int ext4_ext_fiemap_cb(struct inode *inode, struct ext4_ext_path *path, |
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c index 219067ce09d9..da4f2ecb5447 100644 --- a/fs/ext4/inode.c +++ b/fs/ext4/inode.c | |||
| @@ -1233,6 +1233,9 @@ int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block, | |||
| 1233 | clear_buffer_mapped(bh); | 1233 | clear_buffer_mapped(bh); |
| 1234 | clear_buffer_unwritten(bh); | 1234 | clear_buffer_unwritten(bh); |
| 1235 | 1235 | ||
| 1236 | ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u," | ||
| 1237 | "logical block %lu\n", inode->i_ino, flags, max_blocks, | ||
| 1238 | (unsigned long)block); | ||
| 1236 | /* | 1239 | /* |
| 1237 | * Try to see if we can get the block without requesting a new | 1240 | * Try to see if we can get the block without requesting a new |
| 1238 | * file system block. | 1241 | * file system block. |
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c index bf519f239ae6..a93d5b80f3e2 100644 --- a/fs/ext4/migrate.c +++ b/fs/ext4/migrate.c | |||
| @@ -75,7 +75,7 @@ static int finish_range(handle_t *handle, struct inode *inode, | |||
| 75 | goto err_out; | 75 | goto err_out; |
| 76 | } | 76 | } |
| 77 | } | 77 | } |
| 78 | retval = ext4_ext_insert_extent(handle, inode, path, &newext); | 78 | retval = ext4_ext_insert_extent(handle, inode, path, &newext, 0); |
| 79 | err_out: | 79 | err_out: |
| 80 | if (path) { | 80 | if (path) { |
| 81 | ext4_ext_drop_refs(path); | 81 | ext4_ext_drop_refs(path); |
diff --git a/fs/ext4/move_extent.c b/fs/ext4/move_extent.c index c07a2915e40b..5332fd4c4028 100644 --- a/fs/ext4/move_extent.c +++ b/fs/ext4/move_extent.c | |||
| @@ -322,7 +322,7 @@ mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode, | |||
| 322 | goto out; | 322 | goto out; |
| 323 | 323 | ||
| 324 | if (ext4_ext_insert_extent(handle, orig_inode, | 324 | if (ext4_ext_insert_extent(handle, orig_inode, |
| 325 | orig_path, new_ext)) | 325 | orig_path, new_ext, 0)) |
| 326 | goto out; | 326 | goto out; |
| 327 | } | 327 | } |
| 328 | 328 | ||
| @@ -333,7 +333,7 @@ mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode, | |||
| 333 | goto out; | 333 | goto out; |
| 334 | 334 | ||
| 335 | if (ext4_ext_insert_extent(handle, orig_inode, | 335 | if (ext4_ext_insert_extent(handle, orig_inode, |
| 336 | orig_path, end_ext)) | 336 | orig_path, end_ext, 0)) |
| 337 | goto out; | 337 | goto out; |
| 338 | } | 338 | } |
| 339 | out: | 339 | out: |
