udf: add extent cache support in case of file reading

This patch implements extent caching in case of file reading. While reading a file, currently, UDF reads metadata serially which takes a lot of time depending on the number of extents present in the file. Caching last accessd extent improves metadata read time. Instead of reading file metadata from start, now we read from the cached extent. This patch considerably improves the time spent by CPU in kernel mode. For example, while reading a 10.9 GB file using dd: Time before applying patch: 11677022208 bytes (10.9GB) copied, 1529.748921 seconds, 7.3MB/s real 25m 29.85s user 0m 12.41s sys 15m 34.75s Time after applying patch: 11677022208 bytes (10.9GB) copied, 1469.338231 seconds, 7.6MB/s real 24m 29.44s user 0m 15.73s sys 3m 27.61s [JK: Fix bh refcounting issues, simplify initialization] Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com> Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com> Signed-off-by: Bonggil Bak <bgbak@samsung.com> Signed-off-by: Jan Kara <jack@suse.cz>

udf: add extent cache support in case of file reading
This patch implements extent caching in case of file reading. While reading a file, currently, UDF reads metadata serially which takes a lot of time depending on the number of extents present in the file. Caching last accessd extent improves metadata read time. Instead of reading file metadata from start, now we read from the cached extent. This patch considerably improves the time spent by CPU in kernel mode. For example, while reading a 10.9 GB file using dd: Time before applying patch: 11677022208 bytes (10.9GB) copied, 1529.748921 seconds, 7.3MB/s real 25m 29.85s user 0m 12.41s sys 15m 34.75s Time after applying patch: 11677022208 bytes (10.9GB) copied, 1469.338231 seconds, 7.6MB/s real 24m 29.44s user 0m 15.73s sys 3m 27.61s [JK: Fix bh refcounting issues, simplify initialization] Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com> Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com> Signed-off-by: Bonggil Bak <bgbak@samsung.com> Signed-off-by: Jan Kara <jack@suse.cz>
Namjae Jeon · Jan Kara
1 parent 9734c971aa
Showing 4 changed files with 98 additions and 11 deletions Side-by-side Diff
fs/udf/inode.c
fs/udf/super.c
fs/udf/udf_i.h
fs/udf/udfdecl.h
@@ -67,7 +67,75 @@
 			       struct extent_position *);
 static int udf_get_block(struct inode *, sector_t, struct buffer_head *, int);
  
+static void __udf_clear_extent_cache(struct inode *inode)
+{
+	struct udf_inode_info *iinfo = UDF_I(inode);
  
+	if (iinfo->cached_extent.lstart != -1) {
+		brelse(iinfo->cached_extent.epos.bh);
+		iinfo->cached_extent.lstart = -1;
+	}
+}
+
+/* Invalidate extent cache */
+static void udf_clear_extent_cache(struct inode *inode)
+{
+	struct udf_inode_info *iinfo = UDF_I(inode);
+
+	spin_lock(&iinfo->i_extent_cache_lock);
+	__udf_clear_extent_cache(inode);
+	spin_unlock(&iinfo->i_extent_cache_lock);
+}
+
+/* Return contents of extent cache */
+static int udf_read_extent_cache(struct inode *inode, loff_t bcount,
+				 loff_t *lbcount, struct extent_position *pos)
+{
+	struct udf_inode_info *iinfo = UDF_I(inode);
+	int ret = 0;
+
+	spin_lock(&iinfo->i_extent_cache_lock);
+	if ((iinfo->cached_extent.lstart <= bcount) &&
+	    (iinfo->cached_extent.lstart != -1)) {
+		/* Cache hit */
+		*lbcount = iinfo->cached_extent.lstart;
+		memcpy(pos, &iinfo->cached_extent.epos,
+		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
+		ret = 1;
+	}
+	spin_unlock(&iinfo->i_extent_cache_lock);
+	return ret;
+}
+
+/* Add extent to extent cache */
+static void udf_update_extent_cache(struct inode *inode, loff_t estart,
+				    struct extent_position *pos, int next_epos)
+{
+	struct udf_inode_info *iinfo = UDF_I(inode);
+
+	spin_lock(&iinfo->i_extent_cache_lock);
+	/* Invalidate previously cached extent */
+	__udf_clear_extent_cache(inode);
+	if (pos->bh)
+		get_bh(pos->bh);
+	memcpy(&iinfo->cached_extent.epos, pos,
+	       sizeof(struct extent_position));
+	iinfo->cached_extent.lstart = estart;
+	if (next_epos)
+		switch (iinfo->i_alloc_type) {
+		case ICBTAG_FLAG_AD_SHORT:
+			iinfo->cached_extent.epos.offset -=
+			sizeof(struct short_ad);
+			break;
+		case ICBTAG_FLAG_AD_LONG:
+			iinfo->cached_extent.epos.offset -=
+			sizeof(struct long_ad);
+		}
+	spin_unlock(&iinfo->i_extent_cache_lock);
+}
+
 void udf_evict_inode(struct inode *inode)
 {
 	struct udf_inode_info *iinfo = UDF_I(inode);
@@ -90,6 +158,7 @@
 	}
 	kfree(iinfo->i_ext.i_data);
 	iinfo->i_ext.i_data = NULL;
+	udf_clear_extent_cache(inode);
 	if (want_delete) {
 		udf_free_inode(inode);
 	}
@@ -105,6 +174,7 @@
 		truncate_pagecache(inode, to, isize);
 		if (iinfo->i_alloc_type != ICBTAG_FLAG_AD_IN_ICB) {
 			down_write(&iinfo->i_data_sem);
+			udf_clear_extent_cache(inode);
 			udf_truncate_extents(inode);
 			up_write(&iinfo->i_data_sem);
 		}
@@ -372,7 +442,7 @@
 		iinfo->i_next_alloc_goal++;
 	}
  
-
+	udf_clear_extent_cache(inode);
 	phys = inode_getblk(inode, block, &err, &new);
 	if (!phys)
 		goto abort;
@@ -1171,6 +1241,7 @@
 	} else {
 		if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
 			down_write(&iinfo->i_data_sem);
+			udf_clear_extent_cache(inode);
 			memset(iinfo->i_ext.i_data + iinfo->i_lenEAttr + newsize,
 			       0x00, bsize - newsize -
 			       udf_file_entry_alloc_offset(inode));
@@ -1184,6 +1255,7 @@
 		if (err)
 			return err;
 		down_write(&iinfo->i_data_sem);
+		udf_clear_extent_cache(inode);
 		truncate_setsize(inode, newsize);
 		udf_truncate_extents(inode);
 		up_write(&iinfo->i_data_sem);
  
@@ -2156,11 +2228,12 @@
 	struct udf_inode_info *iinfo;
  
 	iinfo = UDF_I(inode);
-	pos->offset = 0;
-	pos->block = iinfo->i_location;
-	pos->bh = NULL;
+	if (!udf_read_extent_cache(inode, bcount, &lbcount, pos)) {
+		pos->offset = 0;
+		pos->block = iinfo->i_location;
+		pos->bh = NULL;
+	}
 	*elen = 0;
-
 	do {
 		etype = udf_next_aext(inode, pos, eloc, elen, 1);
 		if (etype == -1) {
@@ -2170,7 +2243,8 @@
 		}
 		lbcount += *elen;
 	} while (lbcount <= bcount);
-
+	/* update extent cache */
+	udf_update_extent_cache(inode, lbcount - *elen, pos, 1);
 	*offset = (bcount + *elen - lbcount) >> blocksize_bits;
  
 	return etype;
@@ -134,6 +134,8 @@
 	ei->i_next_alloc_goal = 0;
 	ei->i_strat4096 = 0;
 	init_rwsem(&ei->i_data_sem);
+	ei->cached_extent.lstart = -1;
+	spin_lock_init(&ei->i_extent_cache_lock);
  
 	return &ei->vfs_inode;
 }
 #ifndef _UDF_I_H
 #define _UDF_I_H
  
+struct extent_position {
+	struct buffer_head *bh;
+	uint32_t offset;
+	struct kernel_lb_addr block;
+};
+
+struct udf_ext_cache {
+	/* Extent position */
+	struct extent_position epos;
+	/* Start logical offset in bytes */
+	loff_t lstart;
+};
+
 /*
  * The i_data_sem and i_mutex serve for protection of allocation information
  * of a regular files and symlinks. This includes all extents belonging to
@@ -35,6 +48,9 @@
 		__u8		*i_data;
 	} i_ext;
 	struct rw_semaphore	i_data_sem;
+	struct udf_ext_cache cached_extent;
+	/* Spinlock for protecting extent cache */
+	spinlock_t i_extent_cache_lock;
 	struct inode vfs_inode;
 };
  
@@ -113,11 +113,6 @@
 	uint8_t u_len;
 };
  
-struct extent_position {
-	struct buffer_head *bh;
-	uint32_t offset;
-	struct kernel_lb_addr block;
-};
  
 /* super.c */
...	...	@@ -67,7 +67,75 @@
67	67	struct extent_position *);
68	68	static int udf_get_block(struct inode , sector_t, struct buffer_head , int);
69	69
	70	+static void __udf_clear_extent_cache(struct inode *inode)
	71	+{
	72	+ struct udf_inode_info *iinfo = UDF_I(inode);
70	73
	74	+ if (iinfo->cached_extent.lstart != -1) {
	75	+ brelse(iinfo->cached_extent.epos.bh);
	76	+ iinfo->cached_extent.lstart = -1;
	77	+ }
	78	+}
	79	+
	80	+/* Invalidate extent cache */
	81	+static void udf_clear_extent_cache(struct inode *inode)
	82	+{
	83	+ struct udf_inode_info *iinfo = UDF_I(inode);
	84	+
	85	+ spin_lock(&iinfo->i_extent_cache_lock);
	86	+ __udf_clear_extent_cache(inode);
	87	+ spin_unlock(&iinfo->i_extent_cache_lock);
	88	+}
	89	+
	90	+/* Return contents of extent cache */
	91	+static int udf_read_extent_cache(struct inode *inode, loff_t bcount,
	92	+ loff_t lbcount, struct extent_position pos)
	93	+{
	94	+ struct udf_inode_info *iinfo = UDF_I(inode);
	95	+ int ret = 0;
	96	+
	97	+ spin_lock(&iinfo->i_extent_cache_lock);
	98	+ if ((iinfo->cached_extent.lstart <= bcount) &&
	99	+ (iinfo->cached_extent.lstart != -1)) {
	100	+ /* Cache hit */
	101	+ *lbcount = iinfo->cached_extent.lstart;
	102	+ memcpy(pos, &iinfo->cached_extent.epos,
	103	+ sizeof(struct extent_position));
	104	+ if (pos->bh)
	105	+ get_bh(pos->bh);
	106	+ ret = 1;
	107	+ }
	108	+ spin_unlock(&iinfo->i_extent_cache_lock);
	109	+ return ret;
	110	+}
	111	+
	112	+/* Add extent to extent cache */
	113	+static void udf_update_extent_cache(struct inode *inode, loff_t estart,
	114	+ struct extent_position *pos, int next_epos)
	115	+{
	116	+ struct udf_inode_info *iinfo = UDF_I(inode);
	117	+
	118	+ spin_lock(&iinfo->i_extent_cache_lock);
	119	+ /* Invalidate previously cached extent */
	120	+ __udf_clear_extent_cache(inode);
	121	+ if (pos->bh)
	122	+ get_bh(pos->bh);
	123	+ memcpy(&iinfo->cached_extent.epos, pos,
	124	+ sizeof(struct extent_position));
	125	+ iinfo->cached_extent.lstart = estart;
	126	+ if (next_epos)
	127	+ switch (iinfo->i_alloc_type) {
	128	+ case ICBTAG_FLAG_AD_SHORT:
	129	+ iinfo->cached_extent.epos.offset -=
	130	+ sizeof(struct short_ad);
	131	+ break;
	132	+ case ICBTAG_FLAG_AD_LONG:
	133	+ iinfo->cached_extent.epos.offset -=
	134	+ sizeof(struct long_ad);
	135	+ }
	136	+ spin_unlock(&iinfo->i_extent_cache_lock);
	137	+}
	138	+
71	139	void udf_evict_inode(struct inode *inode)
72	140	{
73	141	struct udf_inode_info *iinfo = UDF_I(inode);
...	...	@@ -90,6 +158,7 @@
90	158	}
91	159	kfree(iinfo->i_ext.i_data);
92	160	iinfo->i_ext.i_data = NULL;
	161	+ udf_clear_extent_cache(inode);
93	162	if (want_delete) {
94	163	udf_free_inode(inode);
95	164	}
...	...	@@ -105,6 +174,7 @@
105	174	truncate_pagecache(inode, to, isize);
106	175	if (iinfo->i_alloc_type != ICBTAG_FLAG_AD_IN_ICB) {
107	176	down_write(&iinfo->i_data_sem);
	177	+ udf_clear_extent_cache(inode);
108	178	udf_truncate_extents(inode);
109	179	up_write(&iinfo->i_data_sem);
110	180	}
...	...	@@ -372,7 +442,7 @@
372	442	iinfo->i_next_alloc_goal++;
373	443	}
374	444
375		-
	445	+ udf_clear_extent_cache(inode);
376	446	phys = inode_getblk(inode, block, &err, &new);
377	447	if (!phys)
378	448	goto abort;
...	...	@@ -1171,6 +1241,7 @@
1171	1241	} else {
1172	1242	if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
1173	1243	down_write(&iinfo->i_data_sem);
	1244	+ udf_clear_extent_cache(inode);
1174	1245	memset(iinfo->i_ext.i_data + iinfo->i_lenEAttr + newsize,
1175	1246	0x00, bsize - newsize -
1176	1247	udf_file_entry_alloc_offset(inode));
...	...	@@ -1184,6 +1255,7 @@
1184	1255	if (err)
1185	1256	return err;
1186	1257	down_write(&iinfo->i_data_sem);
	1258	+ udf_clear_extent_cache(inode);
1187	1259	truncate_setsize(inode, newsize);
1188	1260	udf_truncate_extents(inode);
1189	1261	up_write(&iinfo->i_data_sem);
1190	1262
...	...	@@ -2156,11 +2228,12 @@
2156	2228	struct udf_inode_info *iinfo;
2157	2229
2158	2230	iinfo = UDF_I(inode);
2159		- pos->offset = 0;
2160		- pos->block = iinfo->i_location;
2161		- pos->bh = NULL;
	2231	+ if (!udf_read_extent_cache(inode, bcount, &lbcount, pos)) {
	2232	+ pos->offset = 0;
	2233	+ pos->block = iinfo->i_location;
	2234	+ pos->bh = NULL;
	2235	+ }
2162	2236	*elen = 0;
2163		-
2164	2237	do {
2165	2238	etype = udf_next_aext(inode, pos, eloc, elen, 1);
2166	2239	if (etype == -1) {
...	...	@@ -2170,7 +2243,8 @@
2170	2243	}
2171	2244	lbcount += *elen;
2172	2245	} while (lbcount <= bcount);
2173		-
	2246	+ /* update extent cache */
	2247	+ udf_update_extent_cache(inode, lbcount - *elen, pos, 1);
2174	2248	offset = (bcount + elen - lbcount) >> blocksize_bits;
2175	2249
2176	2250	return etype;
...	...	@@ -134,6 +134,8 @@
134	134	ei->i_next_alloc_goal = 0;
135	135	ei->i_strat4096 = 0;
136	136	init_rwsem(&ei->i_data_sem);
	137	+ ei->cached_extent.lstart = -1;
	138	+ spin_lock_init(&ei->i_extent_cache_lock);
137	139
138	140	return &ei->vfs_inode;
139	141	}
1	1	#ifndef _UDF_I_H
2	2	#define _UDF_I_H
3	3
	4	+struct extent_position {
	5	+ struct buffer_head *bh;
	6	+ uint32_t offset;
	7	+ struct kernel_lb_addr block;
	8	+};
	9	+
	10	+struct udf_ext_cache {
	11	+ /* Extent position */
	12	+ struct extent_position epos;
	13	+ /* Start logical offset in bytes */
	14	+ loff_t lstart;
	15	+};
	16	+
4	17	/*
5	18	* The i_data_sem and i_mutex serve for protection of allocation information
6	19	* of a regular files and symlinks. This includes all extents belonging to
...	...	@@ -35,6 +48,9 @@
35	48	__u8 *i_data;
36	49	} i_ext;
37	50	struct rw_semaphore i_data_sem;
	51	+ struct udf_ext_cache cached_extent;
	52	+ /* Spinlock for protecting extent cache */
	53	+ spinlock_t i_extent_cache_lock;
38	54	struct inode vfs_inode;
39	55	};
40	56
...	...	@@ -113,11 +113,6 @@
113	113	uint8_t u_len;
114	114	};
115	115
116		-struct extent_position {
117		- struct buffer_head *bh;
118		- uint32_t offset;
119		- struct kernel_lb_addr block;
120		-};
121	116
122	117	/* super.c */
123	118