[PATCH] ext3_readdir: use generic readahead

Linus points out that ext3_readdir's readahead only cuts in when ext3_readdir() is operating at the very start of the directory. So for large directories we end up performing no readahead at all and we suck. So take it all out and use the core VM's page_cache_readahead(). This means that ext3 directory reads will use all of readahead's dynamic sizing goop. Note that we're using the directory's filp->f_ra to hold the readahead state, but readahead is actually being performed against the underlying blockdev's address_space. Fortunately the readahead code is all set up to handle this. Tested with printk. It works. I was struggling to find a real workload which actually cared. (The patch also exports page_cache_readahead() to GPL modules) Cc: "Stephen C. Tweedie" <sct@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] ext3_readdir: use generic readahead
Linus points out that ext3_readdir's readahead only cuts in when ext3_readdir() is operating at the very start of the directory. So for large directories we end up performing no readahead at all and we suck. So take it all out and use the core VM's page_cache_readahead(). This means that ext3 directory reads will use all of readahead's dynamic sizing goop. Note that we're using the directory's filp->f_ra to hold the readahead state, but readahead is actually being performed against the underlying blockdev's address_space. Fortunately the readahead code is all set up to handle this. Tested with printk. It works. I was struggling to find a real workload which actually cared. (The patch also exports page_cache_readahead() to GPL modules) Cc: "Stephen C. Tweedie" <sct@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Andrew Morton · Linus Torvalds
1 parent b8e31edc10
Showing 4 changed files with 32 additions and 32 deletions Side-by-side Diff
fs/ext3/dir.c
fs/ext3/inode.c
include/linux/ext3_fs.h
mm/readahead.c
@@ -95,11 +95,10 @@
 			 void * dirent, filldir_t filldir)
 {
 	int error = 0;
-	unsigned long offset, blk;
-	int i, num, stored;
-	struct buffer_head * bh, * tmp, * bha[16];
-	struct ext3_dir_entry_2 * de;
-	struct super_block * sb;
+	unsigned long offset;
+	int i, stored;
+	struct ext3_dir_entry_2 *de;
+	struct super_block *sb;
 	int err;
 	struct inode *inode = filp->f_dentry->d_inode;
 	int ret = 0;
  
  
@@ -124,38 +123,35 @@
 	}
 #endif
 	stored = 0;
-	bh = NULL;
 	offset = filp->f_pos & (sb->s_blocksize - 1);
  
 	while (!error && !stored && filp->f_pos < inode->i_size) {
-		blk = (filp->f_pos) >> EXT3_BLOCK_SIZE_BITS(sb);
-		bh = ext3_bread(NULL, inode, blk, 0, &err);
+		unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
+		struct buffer_head map_bh;
+		struct buffer_head *bh = NULL;
+
+		map_bh.b_state = 0;
+		err = ext3_get_block_handle(NULL, inode, blk, &map_bh, 0, 0);
+		if (!err) {
+			page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
+				&filp->f_ra,
+				filp,
+				map_bh.b_blocknr >>
+					(PAGE_CACHE_SHIFT - inode->i_blkbits),
+				1);
+			bh = ext3_bread(NULL, inode, blk, 0, &err);
+		}
+
+		/*
+		 * We ignore I/O errors on directories so users have a chance
+		 * of recovering data when there's a bad sector
+		 */
 		if (!bh) {
 			ext3_error (sb, "ext3_readdir",
 				"directory #%lu contains a hole at offset %lu",
 				inode->i_ino, (unsigned long)filp->f_pos);
 			filp->f_pos += sb->s_blocksize - offset;
 			continue;
-		}
-
-		/*
-		 * Do the readahead
-		 */
-		if (!offset) {
-			for (i = 16 >> (EXT3_BLOCK_SIZE_BITS(sb) - 9), num = 0;
-			     i > 0; i--) {
-				tmp = ext3_getblk (NULL, inode, ++blk, 0, &err);
-				if (tmp && !buffer_uptodate(tmp) &&
-						!buffer_locked(tmp))
-					bha[num++] = tmp;
-				else
-					brelse (tmp);
-			}
-			if (num) {
-				ll_rw_block (READA, num, bha);
-				for (i = 0; i < num; i++)
-					brelse (bha[i]);
-			}
 		}
  
 revalidate:
@@ -671,7 +671,7 @@
  * The BKL may not be held on entry here.  Be sure to take it early.
  */
  
-static int
+int
 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
 		struct buffer_head *bh_result, int create, int extend_disksize)
 {
@@ -772,9 +772,12 @@
  
  
 /* inode.c */
-extern int ext3_forget(handle_t *, int, struct inode *, struct buffer_head *, int);
-extern struct buffer_head * ext3_getblk (handle_t *, struct inode *, long, int, int *);
-extern struct buffer_head * ext3_bread (handle_t *, struct inode *, int, int, int *);
+int ext3_forget(handle_t *, int, struct inode *, struct buffer_head *, int);
+struct buffer_head * ext3_getblk (handle_t *, struct inode *, long, int, int *);
+struct buffer_head * ext3_bread (handle_t *, struct inode *, int, int, int *);
+int ext3_get_block_handle(handle_t *handle, struct inode *inode,
+	sector_t iblock, struct buffer_head *bh_result, int create,
+	int extend_disksize);
  
 extern void ext3_read_inode (struct inode *);
 extern int  ext3_write_inode (struct inode *, int);
@@ -555,6 +555,7 @@
 out:
 	return ra->prev_page + 1;
 }
+EXPORT_SYMBOL_GPL(page_cache_readahead);
  
 /*
  * handle_ra_miss() is called when it is known that a page which should have
...	...	@@ -95,11 +95,10 @@
95	95	void * dirent, filldir_t filldir)
96	96	{
97	97	int error = 0;
98		- unsigned long offset, blk;
99		- int i, num, stored;
100		- struct buffer_head * bh, * tmp, * bha[16];
101		- struct ext3_dir_entry_2 * de;
102		- struct super_block * sb;
	98	+ unsigned long offset;
	99	+ int i, stored;
	100	+ struct ext3_dir_entry_2 *de;
	101	+ struct super_block *sb;
103	102	int err;
104	103	struct inode *inode = filp->f_dentry->d_inode;
105	104	int ret = 0;
106	105
107	106
...	...	@@ -124,38 +123,35 @@
124	123	}
125	124	#endif
126	125	stored = 0;
127		- bh = NULL;
128	126	offset = filp->f_pos & (sb->s_blocksize - 1);
129	127
130	128	while (!error && !stored && filp->f_pos < inode->i_size) {
131		- blk = (filp->f_pos) >> EXT3_BLOCK_SIZE_BITS(sb);
132		- bh = ext3_bread(NULL, inode, blk, 0, &err);
	129	+ unsigned long blk = filp->f_pos >> EXT3_BLOCK_SIZE_BITS(sb);
	130	+ struct buffer_head map_bh;
	131	+ struct buffer_head *bh = NULL;
	132	+
	133	+ map_bh.b_state = 0;
	134	+ err = ext3_get_block_handle(NULL, inode, blk, &map_bh, 0, 0);
	135	+ if (!err) {
	136	+ page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
	137	+ &filp->f_ra,
	138	+ filp,
	139	+ map_bh.b_blocknr >>
	140	+ (PAGE_CACHE_SHIFT - inode->i_blkbits),
	141	+ 1);
	142	+ bh = ext3_bread(NULL, inode, blk, 0, &err);
	143	+ }
	144	+
	145	+ /*
	146	+ * We ignore I/O errors on directories so users have a chance
	147	+ * of recovering data when there's a bad sector
	148	+ */
133	149	if (!bh) {
134	150	ext3_error (sb, "ext3_readdir",
135	151	"directory #%lu contains a hole at offset %lu",
136	152	inode->i_ino, (unsigned long)filp->f_pos);
137	153	filp->f_pos += sb->s_blocksize - offset;
138	154	continue;
139		- }
140		-
141		- /*
142		- * Do the readahead
143		- */
144		- if (!offset) {
145		- for (i = 16 >> (EXT3_BLOCK_SIZE_BITS(sb) - 9), num = 0;
146		- i > 0; i--) {
147		- tmp = ext3_getblk (NULL, inode, ++blk, 0, &err);
148		- if (tmp && !buffer_uptodate(tmp) &&
149		- !buffer_locked(tmp))
150		- bha[num++] = tmp;
151		- else
152		- brelse (tmp);
153		- }
154		- if (num) {
155		- ll_rw_block (READA, num, bha);
156		- for (i = 0; i < num; i++)
157		- brelse (bha[i]);
158		- }
159	155	}
160	156
161	157	revalidate:
...	...	@@ -671,7 +671,7 @@
671	671	* The BKL may not be held on entry here. Be sure to take it early.
672	672	*/
673	673
674		-static int
	674	+int
675	675	ext3_get_block_handle(handle_t handle, struct inode inode, sector_t iblock,
676	676	struct buffer_head *bh_result, int create, int extend_disksize)
677	677	{
...	...	@@ -772,9 +772,12 @@
772	772
773	773
774	774	/* inode.c */
775		-extern int ext3_forget(handle_t , int, struct inode , struct buffer_head *, int);
776		-extern struct buffer_head * ext3_getblk (handle_t , struct inode , long, int, int *);
777		-extern struct buffer_head * ext3_bread (handle_t , struct inode , int, int, int *);
	775	+int ext3_forget(handle_t , int, struct inode , struct buffer_head *, int);
	776	+struct buffer_head * ext3_getblk (handle_t , struct inode , long, int, int *);
	777	+struct buffer_head * ext3_bread (handle_t , struct inode , int, int, int *);
	778	+int ext3_get_block_handle(handle_t handle, struct inode inode,
	779	+ sector_t iblock, struct buffer_head *bh_result, int create,
	780	+ int extend_disksize);
778	781
779	782	extern void ext3_read_inode (struct inode *);
780	783	extern int ext3_write_inode (struct inode *, int);
...	...	@@ -555,6 +555,7 @@
555	555	out:
556	556	return ra->prev_page + 1;
557	557	}
	558	+EXPORT_SYMBOL_GPL(page_cache_readahead);
558	559
559	560	/*
560	561	* handle_ra_miss() is called when it is known that a page which should have