f2fs: speed up handling holes in fiemap
authorChao Yu <chao2.yu@samsung.com>
Tue, 26 Jan 2016 07:42:58 +0000 (15:42 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Tue, 23 Feb 2016 00:07:23 +0000 (16:07 -0800)
This patch makes f2fs_map_blocks supporting returning next potential
page offset which skips hole region in indirect tree of inode, and
use it to speed up fiemap in handling big hole case.

Test method:
xfs_io -f /mnt/f2fs/file  -c "pwrite 1099511627776 4096"
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"

Before:
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"
/mnt/f2fs/file:
 EXT: FILE-OFFSET              BLOCK-RANGE      TOTAL FLAGS
   0: [0..2147483647]:         hole             2147483648
   1: [2147483648..2147483655]: 81920..81927         8   0x1

real    3m3.518s
user    0m0.000s
sys     3m3.456s

After:
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"
/mnt/f2fs/file:
 EXT: FILE-OFFSET              BLOCK-RANGE      TOTAL FLAGS
   0: [0..2147483647]:         hole             2147483648
   1: [2147483648..2147483655]: 81920..81927         8   0x1

real    0m0.008s
user    0m0.000s
sys     0m0.008s

Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/data.c
fs/f2fs/f2fs.h
fs/f2fs/file.c

index bb60e6afbb722ba09ce3942d9ae1c89578157795..e0b09c33fc0f91b748e574dfdcf3164d9888f7d7 100644 (file)
@@ -539,6 +539,7 @@ static int __allocate_data_blocks(struct inode *inode, loff_t offset,
 
        map.m_lblk = F2FS_BYTES_TO_BLK(offset);
        map.m_len = F2FS_BYTES_TO_BLK(count);
+       map.m_next_pgofs = NULL;
 
        return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_DIO);
 }
@@ -586,8 +587,12 @@ next_dnode:
        set_new_dnode(&dn, inode, NULL, NULL, 0);
        err = get_dnode_of_data(&dn, pgofs, mode);
        if (err) {
-               if (err == -ENOENT)
+               if (err == -ENOENT) {
                        err = 0;
+                       if (map->m_next_pgofs)
+                               *map->m_next_pgofs =
+                                       get_next_page_offset(&dn, pgofs);
+               }
                goto unlock_out;
        }
 
@@ -609,6 +614,11 @@ next_block:
                        map->m_flags = F2FS_MAP_NEW;
                        blkaddr = dn.data_blkaddr;
                } else {
+                       if (flag == F2FS_GET_BLOCK_FIEMAP &&
+                                               blkaddr == NULL_ADDR) {
+                               if (map->m_next_pgofs)
+                                       *map->m_next_pgofs = pgofs + 1;
+                       }
                        if (flag != F2FS_GET_BLOCK_FIEMAP ||
                                                blkaddr != NEW_ADDR) {
                                if (flag == F2FS_GET_BLOCK_BMAP)
@@ -669,13 +679,15 @@ out:
 }
 
 static int __get_data_block(struct inode *inode, sector_t iblock,
-                       struct buffer_head *bh, int create, int flag)
+                       struct buffer_head *bh, int create, int flag,
+                       pgoff_t *next_pgofs)
 {
        struct f2fs_map_blocks map;
        int ret;
 
        map.m_lblk = iblock;
        map.m_len = bh->b_size >> inode->i_blkbits;
+       map.m_next_pgofs = next_pgofs;
 
        ret = f2fs_map_blocks(inode, &map, create, flag);
        if (!ret) {
@@ -687,16 +699,18 @@ static int __get_data_block(struct inode *inode, sector_t iblock,
 }
 
 static int get_data_block(struct inode *inode, sector_t iblock,
-                       struct buffer_head *bh_result, int create, int flag)
+                       struct buffer_head *bh_result, int create, int flag,
+                       pgoff_t *next_pgofs)
 {
-       return __get_data_block(inode, iblock, bh_result, create, flag);
+       return __get_data_block(inode, iblock, bh_result, create,
+                                                       flag, next_pgofs);
 }
 
 static int get_data_block_dio(struct inode *inode, sector_t iblock,
                        struct buffer_head *bh_result, int create)
 {
        return __get_data_block(inode, iblock, bh_result, create,
-                                               F2FS_GET_BLOCK_DIO);
+                                               F2FS_GET_BLOCK_DIO, NULL);
 }
 
 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
@@ -707,7 +721,7 @@ static int get_data_block_bmap(struct inode *inode, sector_t iblock,
                return -EFBIG;
 
        return __get_data_block(inode, iblock, bh_result, create,
-                                               F2FS_GET_BLOCK_BMAP);
+                                               F2FS_GET_BLOCK_BMAP, NULL);
 }
 
 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
@@ -725,6 +739,7 @@ int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 {
        struct buffer_head map_bh;
        sector_t start_blk, last_blk;
+       pgoff_t next_pgofs;
        loff_t isize;
        u64 logical = 0, phys = 0, size = 0;
        u32 flags = 0;
@@ -760,14 +775,15 @@ next:
        map_bh.b_size = len;
 
        ret = get_data_block(inode, start_blk, &map_bh, 0,
-                                       F2FS_GET_BLOCK_FIEMAP);
+                                       F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
        if (ret)
                goto out;
 
        /* HOLE */
        if (!buffer_mapped(&map_bh)) {
+               start_blk = next_pgofs;
                /* Go through holes util pass the EOF */
-               if (blk_to_logical(inode, start_blk++) < isize)
+               if (blk_to_logical(inode, start_blk) < isize)
                        goto prep_next;
                /* Found a hole beyond isize means no more extents.
                 * Note that the premise is that filesystems don't
@@ -835,6 +851,7 @@ static int f2fs_mpage_readpages(struct address_space *mapping,
        map.m_lblk = 0;
        map.m_len = 0;
        map.m_flags = 0;
+       map.m_next_pgofs = NULL;
 
        for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 
@@ -873,7 +890,7 @@ static int f2fs_mpage_readpages(struct address_space *mapping,
                        map.m_len = last_block - block_in_file;
 
                        if (f2fs_map_blocks(inode, &map, 0,
-                                                       F2FS_GET_BLOCK_READ))
+                                               F2FS_GET_BLOCK_READ))
                                goto set_error_page;
                }
 got_it:
index d86b52169d6e7574a4f90d00efb1b2bba79b4960..469a1f49a3cdd3a3cc407717b43034057f79ebb3 100644 (file)
@@ -383,6 +383,7 @@ struct f2fs_map_blocks {
        block_t m_lblk;
        unsigned int m_len;
        unsigned int m_flags;
+       pgoff_t *m_next_pgofs;          /* point next possible non-hole pgofs */
 };
 
 /* for flag in get_data_block */
index 8c51fc7d9396d28dc424d744d8801d51c3b7a255..50fa296efa67961479c1da816cceb95386e07460 100644 (file)
@@ -1647,7 +1647,7 @@ static int f2fs_defragment_range(struct f2fs_sb_info *sbi,
                                        struct f2fs_defragment *range)
 {
        struct inode *inode = file_inode(filp);
-       struct f2fs_map_blocks map;
+       struct f2fs_map_blocks map = { .m_next_pgofs = NULL };
        struct extent_info ei;
        pgoff_t pg_start, pg_end;
        unsigned int blk_per_seg = sbi->blocks_per_seg;