ext4: make xattr inode reads faster
authorTahsin Erdogan <tahsin@google.com>
Sun, 6 Aug 2017 04:07:01 +0000 (00:07 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Sun, 6 Aug 2017 04:07:01 +0000 (00:07 -0400)
ext4_xattr_inode_read() currently reads each block sequentially while
waiting for io operation to complete before moving on to the next
block. This prevents request merging in block layer.

Add a ext4_bread_batch() function that starts reads for all blocks
then optionally waits for them to complete. A similar logic is used
in ext4_find_entry(), so update that code to use the new function.

Signed-off-by: Tahsin Erdogan <tahsin@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
fs/ext4/ext4.h
fs/ext4/inode.c
fs/ext4/namei.c
fs/ext4/xattr.c

index d50229e92c55b7508c0fa5bf7378f71db0e54485..a2bb7d2870e487de3f8601107cdf67c05b022034 100644 (file)
@@ -2461,6 +2461,8 @@ extern void ext4_process_freed_data(struct super_block *sb, tid_t commit_tid);
 int ext4_inode_is_fast_symlink(struct inode *inode);
 struct buffer_head *ext4_getblk(handle_t *, struct inode *, ext4_lblk_t, int);
 struct buffer_head *ext4_bread(handle_t *, struct inode *, ext4_lblk_t, int);
+int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
+                    bool wait, struct buffer_head **bhs);
 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
                             struct buffer_head *bh_result, int create);
 int ext4_get_block(struct inode *inode, sector_t iblock,
index 2e6c02258ee264826ae0d3c8187e026ef4976819..56bca45bcdf45a1996ab10d23edc640c8fdd53a4 100644 (file)
@@ -1015,6 +1015,50 @@ struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
        return ERR_PTR(-EIO);
 }
 
+/* Read a contiguous batch of blocks. */
+int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
+                    bool wait, struct buffer_head **bhs)
+{
+       int i, err;
+
+       for (i = 0; i < bh_count; i++) {
+               bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
+               if (IS_ERR(bhs[i])) {
+                       err = PTR_ERR(bhs[i]);
+                       bh_count = i;
+                       goto out_brelse;
+               }
+       }
+
+       for (i = 0; i < bh_count; i++)
+               /* Note that NULL bhs[i] is valid because of holes. */
+               if (bhs[i] && !buffer_uptodate(bhs[i]))
+                       ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
+                                   &bhs[i]);
+
+       if (!wait)
+               return 0;
+
+       for (i = 0; i < bh_count; i++)
+               if (bhs[i])
+                       wait_on_buffer(bhs[i]);
+
+       for (i = 0; i < bh_count; i++) {
+               if (bhs[i] && !buffer_uptodate(bhs[i])) {
+                       err = -EIO;
+                       goto out_brelse;
+               }
+       }
+       return 0;
+
+out_brelse:
+       for (i = 0; i < bh_count; i++) {
+               brelse(bhs[i]);
+               bhs[i] = NULL;
+       }
+       return err;
+}
+
 int ext4_walk_page_buffers(handle_t *handle,
                           struct buffer_head *head,
                           unsigned from,
index cc986b8241817aea5ba782edc24fe1033d508ea0..c1cf020d18895ccedca1690431e009c1f05ad846 100644 (file)
@@ -1342,13 +1342,12 @@ static struct buffer_head * ext4_find_entry (struct inode *dir,
        struct super_block *sb;
        struct buffer_head *bh_use[NAMEI_RA_SIZE];
        struct buffer_head *bh, *ret = NULL;
-       ext4_lblk_t start, block, b;
+       ext4_lblk_t start, block;
        const u8 *name = d_name->name;
-       int ra_max = 0;         /* Number of bh's in the readahead
+       size_t ra_max = 0;      /* Number of bh's in the readahead
                                   buffer, bh_use[] */
-       int ra_ptr = 0;         /* Current index into readahead
+       size_t ra_ptr = 0;      /* Current index into readahead
                                   buffer */
-       int num = 0;
        ext4_lblk_t  nblocks;
        int i, namelen, retval;
        struct ext4_filename fname;
@@ -1411,31 +1410,17 @@ restart:
                if (ra_ptr >= ra_max) {
                        /* Refill the readahead buffer */
                        ra_ptr = 0;
-                       b = block;
-                       for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
-                               /*
-                                * Terminate if we reach the end of the
-                                * directory and must wrap, or if our
-                                * search has finished at this block.
-                                */
-                               if (b >= nblocks || (num && block == start)) {
-                                       bh_use[ra_max] = NULL;
-                                       break;
-                               }
-                               num++;
-                               bh = ext4_getblk(NULL, dir, b++, 0);
-                               if (IS_ERR(bh)) {
-                                       if (ra_max == 0) {
-                                               ret = bh;
-                                               goto cleanup_and_exit;
-                                       }
-                                       break;
-                               }
-                               bh_use[ra_max] = bh;
-                               if (bh)
-                                       ll_rw_block(REQ_OP_READ,
-                                                   REQ_META | REQ_PRIO,
-                                                   1, &bh);
+                       if (block < start)
+                               ra_max = start - block;
+                       else
+                               ra_max = nblocks - block;
+                       ra_max = min(ra_max, ARRAY_SIZE(bh_use));
+                       retval = ext4_bread_batch(dir, block, ra_max,
+                                                 false /* wait */, bh_use);
+                       if (retval) {
+                               ret = ERR_PTR(retval);
+                               ra_max = 0;
+                               goto cleanup_and_exit;
                        }
                }
                if ((bh = bh_use[ra_ptr++]) == NULL)
index 4025666c59912661a2cf9f799bf4cce9bb6a411d..5fa912e5d2a6c53d8ee49987aeee9f78ab18945b 100644 (file)
@@ -317,28 +317,41 @@ static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
  */
 static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 {
-       unsigned long block = 0;
-       struct buffer_head *bh;
-       int blocksize = ea_inode->i_sb->s_blocksize;
-       size_t csize, copied = 0;
-       void *copy_pos = buf;
-
-       while (copied < size) {
-               csize = (size - copied) > blocksize ? blocksize : size - copied;
-               bh = ext4_bread(NULL, ea_inode, block, 0);
-               if (IS_ERR(bh))
-                       return PTR_ERR(bh);
-               if (!bh)
-                       return -EFSCORRUPTED;
+       int blocksize = 1 << ea_inode->i_blkbits;
+       int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
+       int tail_size = (size % blocksize) ?: blocksize;
+       struct buffer_head *bhs_inline[8];
+       struct buffer_head **bhs = bhs_inline;
+       int i, ret;
+
+       if (bh_count > ARRAY_SIZE(bhs_inline)) {
+               bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
+               if (!bhs)
+                       return -ENOMEM;
+       }
 
-               memcpy(copy_pos, bh->b_data, csize);
-               brelse(bh);
+       ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
+                              true /* wait */, bhs);
+       if (ret)
+               goto free_bhs;
 
-               copy_pos += csize;
-               block += 1;
-               copied += csize;
+       for (i = 0; i < bh_count; i++) {
+               /* There shouldn't be any holes in ea_inode. */
+               if (!bhs[i]) {
+                       ret = -EFSCORRUPTED;
+                       goto put_bhs;
+               }
+               memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
+                      i < bh_count - 1 ? blocksize : tail_size);
        }
-       return 0;
+       ret = 0;
+put_bhs:
+       for (i = 0; i < bh_count; i++)
+               brelse(bhs[i]);
+free_bhs:
+       if (bhs != bhs_inline)
+               kfree(bhs);
+       return ret;
 }
 
 static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,