Skip to contentSkip to conversation

udf: add extent cache support in case of file reading

PATCH In review Namjae Jeon 3 participants 10 messages Jan 19, 2013 – Feb 4, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 udf: add extent cache support in case of file reading

From: Namjae Jeon <[email protected]>

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
---
 fs/udf/ialloc.c  |    4 +++
 fs/udf/inode.c   |   79 +++++++++++++++++++++++++++++++++++++++++++++++++-----
 fs/udf/udf_i.h   |   16 +++++++++++
 fs/udf/udfdecl.h |   10 +++----
 4 files changed, 98 insertions(+), 11 deletions(-)
fs/udf/ialloc.c +4 -0
index 7e5aae4..0cb208e 100644
@@ -117,6 +117,10 @@ struct inode *udf_new_inode(struct inode *dir, umode_t mode, int *err)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_use = 0;
 	iinfo->i_checkpoint = 1;
+	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+	spin_lock_init(&(iinfo->i_extent_cache_lock));
+	/* Mark extent cache as invalid for now */
+	iinfo->cached_extent.lstart = -1;
 	if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_AD_IN_ICB))
 		iinfo->i_alloc_type = ICBTAG_FLAG_AD_IN_ICB;
 	else if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_SHORT_AD))
fs/udf/inode.c +73 -6

131 lines not on this page · raw message

fs/udf/udf_i.h +16 -0
index bb8309d..b5cd8ed 100644
@@ -1,6 +1,19 @@
 #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 @@ struct udf_inode_info {
 		__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;
 };
 
fs/udf/udfdecl.h +5 -5
index de038da..db4a22e 100644
@@ -113,11 +113,6 @@ struct ustr {
 	uint8_t u_len;
 };
 
-struct extent_position {
-	struct buffer_head *bh;
-	uint32_t offset;
-	struct kernel_lb_addr block;
-};
 
 /* super.c */
 
@@ -164,6 +159,11 @@ extern int8_t udf_next_aext(struct inode *, struct extent_position *,
 			    struct kernel_lb_addr *, uint32_t *, int);
 extern int8_t udf_current_aext(struct inode *, struct extent_position *,
 			       struct kernel_lb_addr *, uint32_t *, int);
+int udf_read_extent_cache(struct inode *inode, loff_t bcount, loff_t *lbcount,
+			  struct extent_position *pos);
+void udf_update_extent_cache(struct inode *inode, loff_t estart,
+			     struct extent_position *pos, int next_epos);
+void udf_clear_extent_cache(struct udf_inode_info *iinfo);
 
 /* misc.c */
 extern struct buffer_head *udf_tgetblk(struct super_block *, int);
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 17 lines · Namjae Jeon

From: Namjae Jeon <[email protected]>

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
Quoted 5 lines

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

did you have any test on lots of small files?

- cong

Quoted 240 lines
---
 fs/udf/ialloc.c  |    4 +++
 fs/udf/inode.c   |   79 +++++++++++++++++++++++++++++++++++++++++++++++++-----
 fs/udf/udf_i.h   |   16 +++++++++++
 fs/udf/udfdecl.h |   10 +++----
 4 files changed, 98 insertions(+), 11 deletions(-)
fs/udf/ialloc.c +4 -0
index 7e5aae4..0cb208e 100644
@@ -117,6 +117,10 @@ struct inode *udf_new_inode(struct inode *dir, umode_t mode, int *err)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_use = 0;
 	iinfo->i_checkpoint = 1;
+	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+	spin_lock_init(&(iinfo->i_extent_cache_lock));
+	/* Mark extent cache as invalid for now */
+	iinfo->cached_extent.lstart = -1;
 	if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_AD_IN_ICB))
 		iinfo->i_alloc_type = ICBTAG_FLAG_AD_IN_ICB;
 	else if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_SHORT_AD))
fs/udf/inode.c +73 -6
index e78ef48..86e0469 100644
@@ -91,6 +91,7 @@ void udf_evict_inode(struct inode *inode)
 	}
 	kfree(iinfo->i_ext.i_data);
 	iinfo->i_ext.i_data = NULL;
+	udf_clear_extent_cache(iinfo);
 	if (want_delete) {
 		udf_free_inode(inode);
 	}
@@ -106,6 +107,7 @@ static void udf_write_failed(struct address_space *mapping, loff_t to)
 		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(iinfo);
 			udf_truncate_extents(inode);
 			up_write(&iinfo->i_data_sem);
 		}
@@ -373,7 +375,7 @@ static int udf_get_block(struct inode *inode, sector_t block,
 		iinfo->i_next_alloc_goal++;
 	}
 
-
+	udf_clear_extent_cache(iinfo);
 	phys = inode_getblk(inode, block, &err, &new);
 	if (!phys)
 		goto abort;
@@ -1172,6 +1174,7 @@ set_size:
 	} else {
 		if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
 			down_write(&iinfo->i_data_sem);
+			udf_clear_extent_cache(iinfo);
 			memset(iinfo->i_ext.i_data + iinfo->i_lenEAttr + newsize,
 			       0x00, bsize - newsize -
 			       udf_file_entry_alloc_offset(inode));
@@ -1185,6 +1188,7 @@ set_size:
 		if (err)
 			return err;
 		down_write(&iinfo->i_data_sem);
+		udf_clear_extent_cache(iinfo);
 		truncate_setsize(inode, newsize);
 		udf_truncate_extents(inode);
 		up_write(&iinfo->i_data_sem);
@@ -1302,6 +1306,9 @@ static void udf_fill_inode(struct inode *inode, struct buffer_head *bh)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_next_alloc_block = 0;
 	iinfo->i_next_alloc_goal = 0;
+	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+	spin_lock_init(&(iinfo->i_extent_cache_lock));
+	iinfo->cached_extent.lstart = -1;
 	if (fe->descTag.tagIdent == cpu_to_le16(TAG_IDENT_EFE)) {
 		iinfo->i_efe = 1;
 		iinfo->i_use = 0;
@@ -2157,11 +2164,12 @@ int8_t inode_bmap(struct inode *inode, sector_t block,
 	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) {
@@ -2171,7 +2179,8 @@ int8_t inode_bmap(struct inode *inode, sector_t block,
 		}
 		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;
@@ -2201,3 +2210,61 @@ long udf_block_map(struct inode *inode, sector_t block)
 	else
 		return ret;
 }
+
+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);
+
+	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));
+		spin_unlock(&iinfo->i_extent_cache_lock);
+		return 1;
+	} else
+		udf_clear_extent_cache(iinfo);
+	spin_unlock(&iinfo->i_extent_cache_lock);
+	return 0;
+}
+
+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);
+	if (pos->bh != NULL)
+		/* Increase ref count */
+		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);
+}
+
+/* This function should be called after i_data_sem is
+ * held for writing OR i_extent_cache_lock is taken.
+ */
+void udf_clear_extent_cache(struct udf_inode_info *iinfo)
+{
+	if (iinfo->cached_extent.lstart != -1) {
+		brelse(iinfo->cached_extent.epos.bh);
+		memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+		iinfo->cached_extent.lstart = -1;
+	}
+}
+
fs/udf/udf_i.h +16 -0
index bb8309d..b5cd8ed 100644
@@ -1,6 +1,19 @@
 #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 @@ struct udf_inode_info {
 		__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;
 };
 
fs/udf/udfdecl.h +5 -5
index de038da..db4a22e 100644
@@ -113,11 +113,6 @@ struct ustr {
 	uint8_t u_len;
 };
 
-struct extent_position {
-	struct buffer_head *bh;
-	uint32_t offset;
-	struct kernel_lb_addr block;
-};
 
 /* super.c */
 
@@ -164,6 +159,11 @@ extern int8_t udf_next_aext(struct inode *, struct extent_position *,
 			    struct kernel_lb_addr *, uint32_t *, int);
 extern int8_t udf_current_aext(struct inode *, struct extent_position *,
 			       struct kernel_lb_addr *, uint32_t *, int);
+int udf_read_extent_cache(struct inode *inode, loff_t bcount, loff_t *lbcount,
+			  struct extent_position *pos);
+void udf_update_extent_cache(struct inode *inode, loff_t estart,
+			     struct extent_position *pos, int next_epos);
+void udf_clear_extent_cache(struct udf_inode_info *iinfo);
 
 /* misc.c */
 extern struct buffer_head *udf_tgetblk(struct super_block *, int);
Signature · 7 lines
1.7.9.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
#2 lore raw

2013/1/19, Cong Ding <[email protected]>:

Quoted 18 lines · Namjae Jeon
Quoted 17 lines

From: Namjae Jeon <[email protected]>

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
Quoted 6 lines
Quoted 5 lines

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

did you have any test on lots of small files?

Hi. Cong.

I created 2048 files of each 512KB size for testing performance drpping by extent cache feature.

Used this script to read every file:
index=0
while [ $index != 2048 ]
do
dd if=file.$index of=/dev/zero 1> /dev/null 2>/dev/null index=$(($index + 1))
done

Performance without patch:
VDLinux#> echo 3 > /proc/sys/vm/drop_caches
VDLinux#> time ./script2.sh

real    0m 55.13s
user    0m 1.40s
sys     0m 25.17s

Performace with patch =>
VDLinux#> time ./script2.sh

real    0m 53.70s
user    0m 1.60s
sys     0m 25.11s

I can not find any performance dropping with extent cache patch. Thanks.

Quoted 256 lines

- cong

Quoted 253 lines
---
 fs/udf/ialloc.c  |    4 +++
 fs/udf/inode.c   |   79
+++++++++++++++++++++++++++++++++++++++++++++++++-----
 fs/udf/udf_i.h   |   16 +++++++++++
 fs/udf/udfdecl.h |   10 +++----
 4 files changed, 98 insertions(+), 11 deletions(-)
fs/udf/ialloc.c +4 -0
index 7e5aae4..0cb208e 100644
@@ -117,6 +117,10 @@ struct inode *udf_new_inode(struct inode *dir,
 umode_t mode, int *err)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_use = 0;
 	iinfo->i_checkpoint = 1;
+	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+	spin_lock_init(&(iinfo->i_extent_cache_lock));
+	/* Mark extent cache as invalid for now */
+	iinfo->cached_extent.lstart = -1;
 	if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_AD_IN_ICB))
 		iinfo->i_alloc_type = ICBTAG_FLAG_AD_IN_ICB;
fs/udf/ialloc.c +0 -0
 	else if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_SHORT_AD))
fs/udf/inode.c +2 -0
index e78ef48..86e0469 100644
@@ -91,6 +91,7 @@ void udf_evict_inode(struct inode *inode)
 	}
 	kfree(iinfo->i_ext.i_data);
 	iinfo->i_ext.i_data = NULL;
+	udf_clear_extent_cache(iinfo);
 	if (want_delete) {
 		udf_free_inode(inode);
 	}
@@ -106,6 +107,7 @@ static void udf_write_failed(struct address_space
 *mapping, loff_t to)
 		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(iinfo);
 			udf_truncate_extents(inode);
 			up_write(&iinfo->i_data_sem);
fs/udf/inode.c +1 -1
 		}
@@ -373,7 +375,7 @@ static int udf_get_block(struct inode *inode, sector_t
 block,
 		iinfo->i_next_alloc_goal++;
 	}
 
-
+	udf_clear_extent_cache(iinfo);
 	phys = inode_getblk(inode, block, &err, &new);
 	if (!phys)
fs/udf/inode.c +5 -0
 		goto abort;
@@ -1172,6 +1174,7 @@ set_size:
 	} else {
 		if (iinfo->i_alloc_type == ICBTAG_FLAG_AD_IN_ICB) {
 			down_write(&iinfo->i_data_sem);
+			udf_clear_extent_cache(iinfo);
 			memset(iinfo->i_ext.i_data + iinfo->i_lenEAttr + newsize,
 			       0x00, bsize - newsize -
 			       udf_file_entry_alloc_offset(inode));
@@ -1185,6 +1188,7 @@ set_size:
 		if (err)
 			return err;
 		down_write(&iinfo->i_data_sem);
+		udf_clear_extent_cache(iinfo);
 		truncate_setsize(inode, newsize);
 		udf_truncate_extents(inode);
 		up_write(&iinfo->i_data_sem);
@@ -1302,6 +1306,9 @@ static void udf_fill_inode(struct inode *inode,
 struct buffer_head *bh)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_next_alloc_block = 0;
 	iinfo->i_next_alloc_goal = 0;
+	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+	spin_lock_init(&(iinfo->i_extent_cache_lock));
+	iinfo->cached_extent.lstart = -1;
 	if (fe->descTag.tagIdent == cpu_to_le16(TAG_IDENT_EFE)) {
 		iinfo->i_efe = 1;
fs/udf/inode.c +5 -4
 		iinfo->i_use = 0;
@@ -2157,11 +2164,12 @@ int8_t inode_bmap(struct inode *inode, sector_t
 block,
 	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);
fs/udf/inode.c +2 -1
 		if (etype == -1) {
@@ -2171,7 +2179,8 @@ int8_t inode_bmap(struct inode *inode, sector_t
 block,
 		}
 		lbcount += *elen;
 	} while (lbcount <= bcount);
-
+	/* update extent cache */
+	udf_update_extent_cache(inode, lbcount - *elen, pos, 1);
 	*offset = (bcount + *elen - lbcount) >> blocksize_bits;
 
fs/udf/inode.c +58 -0
 	return etype;
@@ -2201,3 +2210,61 @@ long udf_block_map(struct inode *inode, sector_t
 block)
 	else
 		return ret;
 }
+
+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);
+
+	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));
+		spin_unlock(&iinfo->i_extent_cache_lock);
+		return 1;
+	} else
+		udf_clear_extent_cache(iinfo);
+	spin_unlock(&iinfo->i_extent_cache_lock);
+	return 0;
+}
+
+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);
+	if (pos->bh != NULL)
+		/* Increase ref count */
+		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);
+}
+
+/* This function should be called after i_data_sem is
+ * held for writing OR i_extent_cache_lock is taken.
+ */
+void udf_clear_extent_cache(struct udf_inode_info *iinfo)
+{
+	if (iinfo->cached_extent.lstart != -1) {
+		brelse(iinfo->cached_extent.epos.bh);
+		memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
+		iinfo->cached_extent.lstart = -1;
+	}
+}
+
fs/udf/udf_i.h +13 -0
index bb8309d..b5cd8ed 100644
@@ -1,6 +1,19 @@
 #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

fs/udf/udf_i.h +3 -0
@@ -35,6 +48,9 @@ struct udf_inode_info {
 		__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;
 };
 
fs/udf/udfdecl.h +5 -5
index de038da..db4a22e 100644
@@ -113,11 +113,6 @@ struct ustr {
 	uint8_t u_len;
 };
 
-struct extent_position {
-	struct buffer_head *bh;
-	uint32_t offset;
-	struct kernel_lb_addr block;
-};
 
 /* super.c */
 
@@ -164,6 +159,11 @@ extern int8_t udf_next_aext(struct inode *, struct
 extent_position *,
 			    struct kernel_lb_addr *, uint32_t *, int);
 extern int8_t udf_current_aext(struct inode *, struct extent_position *,
 			       struct kernel_lb_addr *, uint32_t *, int);
+int udf_read_extent_cache(struct inode *inode, loff_t bcount, loff_t
 *lbcount,
+			  struct extent_position *pos);
+void udf_update_extent_cache(struct inode *inode, loff_t estart,
+			     struct extent_position *pos, int next_epos);
+void udf_clear_extent_cache(struct udf_inode_info *iinfo);
 

/* misc.c */
extern struct buffer_head *udf_tgetblk(struct super_block *, int);
--
1.7.9.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

reviewer
#3 lore raw
In reply to Namjae Jeon
Quoted 19 lines · Namjae Jeon

From: Namjae Jeon <[email protected]>

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

Quoted 7 lines
real    24m 29.44s
user    0m 15.73s
sys     3m 27.61s

Thanks for the patch Namjae. I did a few more changes to the patch.
Please check them whether you think they are OK.

fs/udf/ialloc.c +0 -4
index 0cb208e..7e5aae4 100644
@@ -117,10 +117,6 @@ struct inode *udf_new_inode(struct inode *dir, umode_t mode, int *err)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_use = 0;
 	iinfo->i_checkpoint = 1;
-	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
-	spin_lock_init(&(iinfo->i_extent_cache_lock));
-	/* Mark extent cache as invalid for now */
-	iinfo->cached_extent.lstart = -1;
 	if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_AD_IN_ICB))
 		iinfo->i_alloc_type = ICBTAG_FLAG_AD_IN_ICB;
 	else if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_SHORT_AD))
fs/udf/inode.c +0 -3
index 8494b8c..fb0c4c4 100644
@@ -1305,9 +1305,6 @@ static void udf_fill_inode(struct inode *inode, struct buffer_head *bh)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_next_alloc_block = 0;
 	iinfo->i_next_alloc_goal = 0;
-	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
-	spin_lock_init(&(iinfo->i_extent_cache_lock));
-	iinfo->cached_extent.lstart = -1;
 	if (fe->descTag.tagIdent == cpu_to_le16(TAG_IDENT_EFE)) {
 		iinfo->i_efe = 1;
 		iinfo->i_use = 0;

Initialization now happens in udf_alloc_inode(). Also it's not necessary
to initialized cached_extent.epos when lstart == -1 - noone should look at that.

@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode, loff_t bcount,
*lbcount = iinfo->cached_extent.lstart;
memcpy(pos, &iinfo->cached_extent.epos,
sizeof(struct extent_position));
+ if (pos->bh)
+ get_bh(pos->bh);
spin_unlock(&iinfo->i_extent_cache_lock);
return 1;
} else
This is the most important - we should give buffer reference to pos->bh.
Caller will eventually free it right?

@@ -2236,8 +2235,7 @@ void udf_update_extent_cache(struct inode *inode, loff_t estart,
struct udf_inode_info *iinfo = UDF_I(inode);

spin_lock(&iinfo->i_extent_cache_lock);
- if (pos->bh != NULL)
- /* Increase ref count */
+ if (pos->bh)
get_bh(pos->bh);
memcpy(&iinfo->cached_extent.epos, pos,
sizeof(struct extent_position));
@@ -2266,4 +2264,3 @@ void udf_clear_extent_cache(struct udf_inode_info *iinfo)
iinfo->cached_extent.lstart = -1;
}
}
-

fs/udf/super.c +2 -0
index 186adbf..da8ce9f 100644
@@ -134,6 +134,8 @@ static struct inode *udf_alloc_inode(struct super_block *sb)
 	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;
 }

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#4 lore raw

2013/1/21, Jan Kara <[email protected]>:

Quoted 60 lines · Namjae Jeon
Quoted 19 lines

From: Namjae Jeon <[email protected]>

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

Quoted 7 lines
real    24m 29.44s
user    0m 15.73s
sys     3m 27.61s

Thanks for the patch Namjae. I did a few more changes to the patch.
Please check them whether you think they are OK.

fs/udf/ialloc.c +0 -4
index 0cb208e..7e5aae4 100644
@@ -117,10 +117,6 @@ struct inode *udf_new_inode(struct inode *dir, umode_t
 mode, int *err)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_use = 0;
 	iinfo->i_checkpoint = 1;
-	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
-	spin_lock_init(&(iinfo->i_extent_cache_lock));
-	/* Mark extent cache as invalid for now */
-	iinfo->cached_extent.lstart = -1;
 	if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_AD_IN_ICB))
 		iinfo->i_alloc_type = ICBTAG_FLAG_AD_IN_ICB;
fs/udf/ialloc.c +0 -0
 	else if (UDF_QUERY_FLAG(inode->i_sb, UDF_FLAG_USE_SHORT_AD))
fs/udf/inode.c +0 -3
index 8494b8c..fb0c4c4 100644
@@ -1305,9 +1305,6 @@ static void udf_fill_inode(struct inode *inode, struct
 buffer_head *bh)
 	iinfo->i_lenAlloc = 0;
 	iinfo->i_next_alloc_block = 0;
 	iinfo->i_next_alloc_goal = 0;
-	memset(&iinfo->cached_extent, 0, sizeof(struct udf_ext_cache));
-	spin_lock_init(&(iinfo->i_extent_cache_lock));
-	iinfo->cached_extent.lstart = -1;
 	if (fe->descTag.tagIdent == cpu_to_le16(TAG_IDENT_EFE)) {
 		iinfo->i_efe = 1;
Quoted 2 lines

iinfo->i_use = 0;

Hi Jan.

Quoted 3 lines

Initialization now happens in udf_alloc_inode(). Also it's not necessary
to initialized cached_extent.epos when lstart == -1 - noone should look at that.

This change is okay.

Quoted 10 lines
fs/udf/inode.c +2 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode, loff_t
 bcount,
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
Quoted 3 lines

} else
This is the most important - we should give buffer reference to pos->bh.
Caller will eventually free it right?

This change is not required as we give buffer reference to pos->bh at the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented. Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Thanks for review and change!

Quoted 38 lines
fs/udf/inode.c +1 -2
@@ -2236,8 +2235,7 @@ void udf_update_extent_cache(struct inode *inode,
 loff_t estart,
 	struct udf_inode_info *iinfo = UDF_I(inode);
 
 	spin_lock(&iinfo->i_extent_cache_lock);
-	if (pos->bh != NULL)
-		/* Increase ref count */
+	if (pos->bh)
 		get_bh(pos->bh);
 	memcpy(&iinfo->cached_extent.epos, pos,
fs/udf/inode.c +0 -1
 	       sizeof(struct extent_position));
@@ -2266,4 +2264,3 @@ void udf_clear_extent_cache(struct udf_inode_info
 *iinfo)
 		iinfo->cached_extent.lstart = -1;
 	}
 }
-
fs/udf/super.c +2 -0
index 186adbf..da8ce9f 100644
@@ -134,6 +134,8 @@ static struct inode *udf_alloc_inode(struct super_block
 *sb)
 	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;

}

Honza
--
Jan Kara <[email protected]>
SUSE Labs, CR

reviewer
#5 lore raw
Quoted 13 lines · Namjae Jeon

2013/1/21, Jan Kara <[email protected]>:

Quoted 9 lines
(quoted diff) +2 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode, loff_t
 bcount,
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
Quoted 3 lines

} else
This is the most important - we should give buffer reference to pos->bh.
Caller will eventually free it right?

Quoted 8 lines

This change is not required as we give buffer reference to pos->bh at the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented. Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Yes, udf_update_extent_cache() gets its own reference to bh but that is
dropped in udf_clear_extent_cache(). So I think udf_read_extent_cache() needs to get a reference to the caller (as the caller will eventually free the bh via brelse(epos.bh) e.g. in udf_extend_file(). Also I realized udf_update_extent_cache() needs to first clear the cache if it is valid. Otherwise it just overwrites bh pointer and reference is leaked. Is it clearer now?

I've also changed locking of udf_clear_extent_cache() so that
i_extent_cache_lock is always taken for that function - it makes the locking rules obvious at the first sight.

Attached is the patch I currently carry.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
Quoted 1 line

From 99600051b04bc4ec8bd4d16a8bf993ca54042db6 Mon Sep 17 00:00:00 2001

From: Namjae Jeon <[email protected]>
Date: Sat, 19 Jan 2013 11:17:14 +0900
Subject: [PATCH] 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]

---
 fs/udf/inode.c   |   86 ++++++++++++++++++++++++++++++++++++++++++++++++++----
 fs/udf/super.c   |    2 +
 fs/udf/udf_i.h   |   16 ++++++++++
 fs/udf/udfdecl.h |    5 ---
 4 files changed, 98 insertions(+), 11 deletions(-)
fs/udf/inode.c +80 -6
index cbae1ed..7a12e48 100644
@@ -67,6 +67,74 @@ static void udf_update_extents(struct inode *,
 			       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)
 {
@@ -90,6 +158,7 @@ void udf_evict_inode(struct inode *inode)
 	}
 	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 @@ static void udf_write_failed(struct address_space *mapping, loff_t to)
 		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 @@ static int udf_get_block(struct inode *inode, sector_t block,
 		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 @@ set_size:
 	} 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 @@ set_size:
 		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 @@ int8_t inode_bmap(struct inode *inode, sector_t block,
 	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 @@ int8_t inode_bmap(struct inode *inode, sector_t block,
 		}
 		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;
fs/udf/super.c +2 -0
index 186adbf..da8ce9f 100644
@@ -134,6 +134,8 @@ static struct inode *udf_alloc_inode(struct super_block *sb)
 	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;
 }
fs/udf/udf_i.h +16 -0
index bb8309d..b5cd8ed 100644
@@ -1,6 +1,19 @@
 #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 @@ struct udf_inode_info {
 		__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;
 };
 
fs/udf/udfdecl.h +0 -5
index de038da..be7dabb 100644
@@ -113,11 +113,6 @@ struct ustr {
 	uint8_t u_len;
 };
 
-struct extent_position {
-	struct buffer_head *bh;
-	uint32_t offset;
-	struct kernel_lb_addr block;
-};
 
 /* super.c */
 
Signature · 1 line
1.7.1
#6 lore raw

2013/1/22, Jan Kara <[email protected]>:

Quoted 24 lines · Namjae Jeon
Quoted 15 lines

2013/1/21, Jan Kara <[email protected]>:

Quoted 12 lines
(quoted diff) +0 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode,

loff_t
bcount,

(quoted diff) +2 -0
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
 	} else
  This is the most important - we should give buffer reference to
Quoted 2 lines

pos->bh.
Caller will eventually free it right?

Quoted 8 lines

This change is not required as we give buffer reference to pos->bh at the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented. Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Quoted 7 lines

Yes, udf_update_extent_cache() gets its own reference to bh but that is
dropped in udf_clear_extent_cache(). So I think udf_read_extent_cache() needs to get a reference to the caller (as the caller will eventually free the bh via brelse(epos.bh) e.g. in udf_extend_file(). Also I realized udf_update_extent_cache() needs to first clear the cache if it is valid. Otherwise it just overwrites bh pointer and reference is leaked. Is it clearer now?

Yes, you're right. Also, this patch looks good to me.

Quoted 4 lines

I've also changed locking of udf_clear_extent_cache() so that
i_extent_cache_lock is always taken for that function - it makes the locking rules obvious at the first sight.

Yes, right. it is needed.
When we test with this patch, working fine.
Thanks Jan!

Quoted 9 lines

Attached is the patch I currently carry.

Honza

--
Jan Kara <[email protected]>
SUSE Labs, CR

#7 lore raw

Hi. Jan.

Sorry for interrupt.
Have you taken this patch to your tree ? I can not find it.. or Is there any issue regarding this patch ?

Thanks!

2013/1/22, Namjae Jeon <[email protected]>:

Quoted 51 lines

2013/1/22, Jan Kara <[email protected]>:

Quoted 24 lines · Namjae Jeon
Quoted 15 lines

2013/1/21, Jan Kara <[email protected]>:

Quoted 12 lines
(quoted diff) +0 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode,

loff_t
bcount,

(quoted diff) +2 -0
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
 	} else
  This is the most important - we should give buffer reference to
Quoted 2 lines

pos->bh.
Caller will eventually free it right?

Quoted 8 lines

This change is not required as we give buffer reference to pos->bh at the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented. Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Quoted 8 lines

Yes, udf_update_extent_cache() gets its own reference to bh but that is
dropped in udf_clear_extent_cache(). So I think udf_read_extent_cache() needs to get a reference to the caller (as the caller will eventually free
the bh via brelse(epos.bh) e.g. in udf_extend_file(). Also I realized udf_update_extent_cache() needs to first clear the cache if it is valid. Otherwise it just overwrites bh pointer and reference is leaked. Is it clearer now?

Yes, you're right. Also, this patch looks good to me.

Quoted 4 lines

I've also changed locking of udf_clear_extent_cache() so that
i_extent_cache_lock is always taken for that function - it makes the locking rules obvious at the first sight.

Yes, right. it is needed.
When we test with this patch, working fine.
Thanks Jan!

Quoted 9 lines

Attached is the patch I currently carry.

Honza

--
Jan Kara <[email protected]>
SUSE Labs, CR

reviewer
#8 lore raw
Quoted 5 lines · Namjae Jeon

Hi. Jan.

Sorry for interrupt.
Have you taken this patch to your tree ? I can not find it.. or Is there any issue regarding this patch ?

I had it in my tree but not in the for_next branch. Did it now so you
should see the patch in tomorrow's linux-next.

Honza

Quoted 52 lines

2013/1/22, Namjae Jeon <[email protected]>:

Quoted 51 lines

2013/1/22, Jan Kara <[email protected]>:

Quoted 24 lines · Namjae Jeon
Quoted 15 lines

2013/1/21, Jan Kara <[email protected]>:

Quoted 12 lines
(quoted diff) +0 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode *inode,

loff_t
bcount,

(quoted diff) +2 -0
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
 	} else
  This is the most important - we should give buffer reference to
Quoted 2 lines

pos->bh.
Caller will eventually free it right?

Quoted 8 lines

This change is not required as we give buffer reference to pos->bh at the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented. Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Quoted 8 lines

Yes, udf_update_extent_cache() gets its own reference to bh but that is
dropped in udf_clear_extent_cache(). So I think udf_read_extent_cache() needs to get a reference to the caller (as the caller will eventually free
the bh via brelse(epos.bh) e.g. in udf_extend_file(). Also I realized udf_update_extent_cache() needs to first clear the cache if it is valid. Otherwise it just overwrites bh pointer and reference is leaked. Is it clearer now?

Yes, you're right. Also, this patch looks good to me.

Quoted 4 lines

I've also changed locking of udf_clear_extent_cache() so that
i_extent_cache_lock is always taken for that function - it makes the locking rules obvious at the first sight.

Yes, right. it is needed.
When we test with this patch, working fine.
Thanks Jan!

Quoted 9 lines

Attached is the patch I currently carry.

Honza

--
Jan Kara <[email protected]>
SUSE Labs, CR

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#9 lore raw

2013/2/4, Jan Kara <[email protected]>:

Quoted 8 lines · Namjae Jeon
Quoted 5 lines

Hi. Jan.

Sorry for interrupt.
Have you taken this patch to your tree ? I can not find it.. or Is there any issue regarding this patch ?

I had it in my tree but not in the for_next branch. Did it now so you
should see the patch in tomorrow's linux-next.

Okay, I see.
Thanks Jan!

Quoted 63 lines

Honza

Quoted 57 lines

2013/1/22, Namjae Jeon <[email protected]>:

Quoted 56 lines

2013/1/22, Jan Kara <[email protected]>:

Quoted 26 lines · Namjae Jeon
Quoted 17 lines

2013/1/21, Jan Kara <[email protected]>:

Quoted 13 lines
(quoted diff) +0 -0
@@ -2222,6 +2219,8 @@ int udf_read_extent_cache(struct inode
 *inode,

loff_t
bcount,

(quoted diff) +2 -0
 		*lbcount = iinfo->cached_extent.lstart;
 		memcpy(pos, &iinfo->cached_extent.epos,
 		       sizeof(struct extent_position));
+		if (pos->bh)
+			get_bh(pos->bh);
 		spin_unlock(&iinfo->i_extent_cache_lock);
 		return 1;
 	} else
  This is the most important - we should give buffer reference to
Quoted 2 lines

pos->bh.
Caller will eventually free it right?

This change is not required as we give buffer reference to pos->bh at

Quoted 8 lines

the time of cache update.
When we start reading a file, first we try to read the cache which will lead to cache miss.
So, we would really access the pos->bh in udf_update_extent_cache for the first time, and this is where the buffer reference is incremented.
Calling get_bh at 2 places will eventually lead to mem leak. Let me know your opinion.

Quoted 11 lines

Yes, udf_update_extent_cache() gets its own reference to bh but that
is
dropped in udf_clear_extent_cache(). So I think udf_read_extent_cache()
needs to get a reference to the caller (as the caller will eventually free
the bh via brelse(epos.bh) e.g. in udf_extend_file(). Also I realized udf_update_extent_cache() needs to first clear the cache if it is valid.
Otherwise it just overwrites bh pointer and reference is leaked. Is it clearer now?

Yes, you're right. Also, this patch looks good to me.

Quoted 4 lines

I've also changed locking of udf_clear_extent_cache() so that
i_extent_cache_lock is always taken for that function - it makes the locking rules obvious at the first sight.

Yes, right. it is needed.
When we test with this patch, working fine.
Thanks Jan!

Quoted 9 lines

Attached is the patch I currently carry.

Honza

--
Jan Kara <[email protected]>
SUSE Labs, CR

--
Jan Kara <[email protected]>
SUSE Labs, CR