Skip to contentSkip to conversation

f2fs: avoid readahead race condition

PATCH In review Jaegeuk Kim 4 participants 14 messages Jun 24, 2020 – Jul 3, 2020 F2FS file system

lore
Share & tools
lore

Jaegeuk Kim

author maintainer
#0 lore raw

PATCH 1/1 f2fs: avoid readahead race condition

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
 fs/f2fs/data.c  | 15 +++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 18 insertions(+)
fs/f2fs/data.c +15 -0
index dfd3225153570..1886d83bc5f15 100644
@@ -2292,6 +2292,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2302,6 +2303,17 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (F2FS_I(inode)->ra_offset == readahead_index(rac))
+			drop_ra = true;
+		else
+			F2FS_I(inode)->ra_offset = readahead_index(rac);
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
@@ -2368,6 +2380,9 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	}
 	if (bio)
 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
+
+	if (rac && readahead_count(rac) && !drop_ra)
+		F2FS_I(inode)->ra_offset = -1;
 	return ret;
 }
 
fs/f2fs/f2fs.h +1 -0
index 7fb2a1a334388..753782426feac 100644
@@ -809,6 +809,7 @@ struct f2fs_inode_info {
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct task_struct *inmem_task;	/* store inmemory task */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
+	pgoff_t ra_offset;		/* ongoing readahead offset */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
 
 	/* avoid racing between foreground op and gc */
fs/f2fs/super.c +2 -0
index 7326522057378..80cb7cd358f84 100644
@@ -1015,6 +1015,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
 	/* Will be used by directory only */
 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
 
+	fi->ra_offset = -1;
+
 	return &fi->vfs_inode;
 }
 
Signature · 1 line
2.27.0.111.gc72c7da667-goog
#1 lore raw
Quoted 25 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
 fs/f2fs/data.c  | 15 +++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 18 insertions(+)
fs/f2fs/data.c +1 -0
index dfd3225153570..1886d83bc5f15 100644
@@ -2292,6 +2292,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2312

Quoted 10 lines
fs/f2fs/data.c +7 -0
@@ -2302,6 +2303,17 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (F2FS_I(inode)->ra_offset == readahead_index(rac))
+			drop_ra = true;

I guess you missed to return at somewhere when drop_ra is true?

thanks,

Quoted 43 lines
fs/f2fs/data.c +7 -0
+		else
+			F2FS_I(inode)->ra_offset = readahead_index(rac);
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
@@ -2368,6 +2380,9 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	}
 	if (bio)
 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
+
+	if (rac && readahead_count(rac) && !drop_ra)
+		F2FS_I(inode)->ra_offset = -1;
 	return ret;
 }
 
fs/f2fs/f2fs.h +1 -0
index 7fb2a1a334388..753782426feac 100644
@@ -809,6 +809,7 @@ struct f2fs_inode_info {
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct task_struct *inmem_task;	/* store inmemory task */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
+	pgoff_t ra_offset;		/* ongoing readahead offset */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
 
 	/* avoid racing between foreground op and gc */
fs/f2fs/super.c +2 -0
index 7326522057378..80cb7cd358f84 100644
@@ -1015,6 +1015,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
 	/* Will be used by directory only */
 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
 
+	fi->ra_offset = -1;
+
 	return &fi->vfs_inode;
 }
 
author maintainer
#2 lore raw
In reply to Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 +++++++++++++++++-
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 20 insertions(+), 1 deletion(-)
fs/f2fs/data.c +17 -1
index d6094b9f3916..9b69a159cc6c 100644
@@ -2403,6 +2403,7 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 #endif
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2413,13 +2414,25 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (pages && is_readahead) {
+		page = list_last_entry(pages, struct page, lru);
+		if (F2FS_I(inode)->ra_offset == page_index(page))
+			drop_ra = true;
+		else
+			F2FS_I(inode)->ra_offset = page_index(page);
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (pages) {
 			page = list_last_entry(pages, struct page, lru);
 
 			prefetchw(&page->flags);
 			list_del(&page->lru);
-			if (add_to_page_cache_lru(page, mapping,
+			if (drop_ra || add_to_page_cache_lru(page, mapping,
 						  page_index(page),
 						  readahead_gfp_mask(mapping)))
 				goto next_page;
@@ -2484,6 +2497,9 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 	BUG_ON(pages && !list_empty(pages));
 	if (bio)
 		__f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
+
+	if (pages && is_readahead && !drop_ra)
+		F2FS_I(inode)->ra_offset = -1;
 	return pages ? 0 : ret;
 }
 
fs/f2fs/f2fs.h +1 -0
index 35afa13124b8..a95f84d72a55 100644
@@ -806,6 +806,7 @@ struct f2fs_inode_info {
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct task_struct *inmem_task;	/* store inmemory task */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
+	pgoff_t ra_offset;		/* ongoing readahead offset */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
 
 	/* avoid racing between foreground op and gc */
fs/f2fs/super.c +2 -0
index 0e860186a9c5..6fd2ad43d9e4 100644
@@ -1011,6 +1011,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
 	/* Will be used by directory only */
 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
 
+	fi->ra_offset = -1;
+
 	return &fi->vfs_inode;
 }
 
Signature · 8 lines
2.27.0.111.gc72c7da667-goog



_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
#3 lore raw
Quoted 32 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 +++++++++++++++++-
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 20 insertions(+), 1 deletion(-)
fs/f2fs/data.c +2 -0
index d6094b9f3916..9b69a159cc6c 100644
@@ -2403,6 +2403,7 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 #endif
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2413,13 +2414,25 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*

fs/f2fs/data.c int f2fs_mpage_readpages(struct address_space *mapping, · line 2427

Quoted 10 lines
fs/f2fs/data.c +10 -0
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (pages && is_readahead) {
+		page = list_last_entry(pages, struct page, lru);
+		if (F2FS_I(inode)->ra_offset == page_index(page))
+			drop_ra = true;
+		else
+			F2FS_I(inode)->ra_offset = page_index(page);
+	}

This is a data race because ra_offset can be read/written by different threads concurrently.

It either needs locking, or READ_ONCE() and WRITE_ONCE() if races are okay.

- Eric

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

author maintainer
#4 lore raw
Quoted 47 lines · Eric Biggers
Quoted 32 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 +++++++++++++++++-
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 20 insertions(+), 1 deletion(-)
fs/f2fs/data.c +2 -0
index d6094b9f3916..9b69a159cc6c 100644
@@ -2403,6 +2403,7 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 #endif
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2413,13 +2414,25 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*

fs/f2fs/data.c int f2fs_mpage_readpages(struct address_space *mapping, · line 2427

Quoted 10 lines
fs/f2fs/data.c +10 -0
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (pages && is_readahead) {
+		page = list_last_entry(pages, struct page, lru);
+		if (F2FS_I(inode)->ra_offset == page_index(page))
+			drop_ra = true;
+		else
+			F2FS_I(inode)->ra_offset = page_index(page);
+	}

This is a data race because ra_offset can be read/written by different threads concurrently.

Quoted 1 line

It either needs locking, or READ_ONCE() and WRITE_ONCE() if races are okay.

I just wanted to keep zero overhead, since it doesn't matter either cases of skipping readahead or not.

Quoted 2 lines

- Eric

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

#5 lore raw
Quoted 48 lines · Jaegeuk Kim
Quoted 47 lines · Eric Biggers
Quoted 32 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 +++++++++++++++++-
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 20 insertions(+), 1 deletion(-)
fs/f2fs/data.c +2 -0
index d6094b9f3916..9b69a159cc6c 100644
@@ -2403,6 +2403,7 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 #endif
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2413,13 +2414,25 @@ int f2fs_mpage_readpages(struct address_space *mapping,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*

fs/f2fs/data.c int f2fs_mpage_readpages(struct address_space *mapping, · line 2427

Quoted 10 lines
fs/f2fs/data.c +10 -0
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (pages && is_readahead) {
+		page = list_last_entry(pages, struct page, lru);
+		if (F2FS_I(inode)->ra_offset == page_index(page))
+			drop_ra = true;
+		else
+			F2FS_I(inode)->ra_offset = page_index(page);
+	}

This is a data race because ra_offset can be read/written by different threads concurrently.

Quoted 5 lines
Quoted 1 line

It either needs locking, or READ_ONCE() and WRITE_ONCE() if races are okay.

I just wanted to keep zero overhead, since it doesn't matter either cases of skipping readahead or not.

Okay, then it should use READ_ONCE() and WRITE_ONCE().

- Eric

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

author maintainer
#6 lore raw
In reply to Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +18 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;
 		}
 
 #ifdef CONFIG_F2FS_FS_COMPRESSION
@@ -2372,6 +2387,9 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	}
 	if (bio)
 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
+
+	if (rac && readahead_count(rac) && !drop_ra)
+		WRITE_ONCE(F2FS_I(inode)->ra_offset, -1);
 	return ret;
 }
 
fs/f2fs/f2fs.h +1 -0
index 6a655edeb522f..e6e47618a3576 100644
@@ -809,6 +809,7 @@ struct f2fs_inode_info {
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct task_struct *inmem_task;	/* store inmemory task */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
+	pgoff_t ra_offset;		/* ongoing readahead offset */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
 
 	/* avoid racing between foreground op and gc */
fs/f2fs/super.c +2 -0
index 7326522057378..80cb7cd358f84 100644
@@ -1015,6 +1015,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
 	/* Will be used by directory only */
 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
 
+	fi->ra_offset = -1;
+
 	return &fi->vfs_inode;
 }
 
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog
#7 lore raw
Quoted 41 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +9 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2327

Quoted 10 lines
fs/f2fs/data.c +6 -0
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;

When CONFIG_F2FS_FS_COMPRESSION is not set (i.e. x86_64 defconfig + CONFIG_F2FS_FS=y):

$ make -skj"$(nproc)" O=out distclean defconfig fs/f2fs/data.o
../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined
 2327 |     goto next_page;
      |     ^~~~
...

Cheers,
Nathan

author maintainer
#8 lore raw
Quoted 56 lines · Nathan Chancellor
Quoted 41 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +9 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2327

Quoted 10 lines
fs/f2fs/data.c +6 -0
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;

When CONFIG_F2FS_FS_COMPRESSION is not set (i.e. x86_64 defconfig + CONFIG_F2FS_FS=y):

Quoted 6 lines
$ make -skj"$(nproc)" O=out distclean defconfig fs/f2fs/data.o
../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined
 2327 |     goto next_page;
      |     ^~~~
...

Thanks. I pushed the fix for -next.
https://lore.kernel.org/linux-f2fs-devel/[email protected]/T/#t

Thanks,

Quoted 9 lines

Cheers,
Nathan

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Reviewed-by
#9 lore raw
Quoted 64 lines · Jaegeuk Kim
Quoted 56 lines · Nathan Chancellor
Quoted 41 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +9 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2327

Quoted 10 lines
fs/f2fs/data.c +6 -0
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;

When CONFIG_F2FS_FS_COMPRESSION is not set (i.e. x86_64 defconfig + CONFIG_F2FS_FS=y):

Quoted 6 lines
$ make -skj"$(nproc)" O=out distclean defconfig fs/f2fs/data.o
../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined
 2327 |     goto next_page;
      |     ^~~~
...
Quoted 2 lines

Thanks. I pushed the fix for -next.
https://lore.kernel.org/linux-f2fs-devel/[email protected]/T/#t

Thanks,

Quoted 19 lines

Thanks,

Quoted 9 lines

Cheers,
Nathan

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

#10 lore raw
Quoted 65 lines · Chao Yu
Quoted 64 lines · Jaegeuk Kim
Quoted 62 lines · Nathan Chancellor
Quoted 41 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +9 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2327

Quoted 10 lines
fs/f2fs/data.c +6 -0
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;

When CONFIG_F2FS_FS_COMPRESSION is not set (i.e. x86_64 defconfig + CONFIG_F2FS_FS=y):

$ make -skj"$(nproc)" O=out distclean defconfig fs/f2fs/data.o
../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined
 2327 |     goto next_page;
      |     ^~~~
...
Quoted 2 lines
Quoted 2 lines

Thanks. I pushed the fix for -next.
https://lore.kernel.org/linux-f2fs-devel/[email protected]/T/#t

It will hang the kernel because we missed to unlock those cached pages, I changed to 'goto set_error_page', the issue was gone.

Thanks,

Quoted 31 lines

Thanks,

Quoted 14 lines

Thanks,

Quoted 9 lines

Cheers,
Nathan

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Quoted 5 lines

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

author maintainer
#11 lore raw
Quoted 66 lines · Chao Yu
Quoted 65 lines · Chao Yu
Quoted 64 lines · Jaegeuk Kim
Quoted 62 lines · Nathan Chancellor
Quoted 41 lines · Jaegeuk Kim

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

---
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 18 ++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 21 insertions(+)
fs/f2fs/data.c +9 -0
index 995cf78b23c5e..360b4c9080d97 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,24 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else

fs/f2fs/data.c static int f2fs_mpage_readpages(struct inode *inode, · line 2327

Quoted 10 lines
fs/f2fs/data.c +6 -0
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra)
+				goto next_page;

When CONFIG_F2FS_FS_COMPRESSION is not set (i.e. x86_64 defconfig + CONFIG_F2FS_FS=y):

$ make -skj"$(nproc)" O=out distclean defconfig fs/f2fs/data.o
../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined
 2327 |     goto next_page;
      |     ^~~~
...
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines

Thanks. I pushed the fix for -next.
https://lore.kernel.org/linux-f2fs-devel/[email protected]/T/#t

It will hang the kernel because we missed to unlock those cached pages, I changed to 'goto set_error_page', the issue was gone.

How about v4?

Quoted 34 lines

Thanks,

Quoted 31 lines

Thanks,

Quoted 14 lines

Thanks,

Quoted 9 lines

Cheers,
Nathan

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Quoted 5 lines

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

#12 lore raw
Quoted 1 line · Jaegeuk Kim

How about v4?

It looks good to me.

Thanks,

author maintainer Reviewed-by Chao Yu
#13 lore raw
In reply to Jaegeuk Kim

From 3634864095bd1aafbb60ff49dac7d13ce157b658 Mon Sep 17 00:00:00 2001
From: Jaegeuk Kim <[email protected]>
Date: Mon, 22 Jun 2020 23:01:05 -0700
Subject: [PATCH] f2fs: avoid readahead race condition
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

If two readahead threads having same offset enter in readpages, every read IOs are split and issued to the disk which giving lower bandwidth.

This patch tries to avoid redundant readahead calls.

Fixes one build error reported by Randy.
Fix build error when F2FS_FS_COMPRESSION is not set/enabled. This label is needed in either case.

../fs/f2fs/data.c: In function ‘f2fs_mpage_readpages’:
../fs/f2fs/data.c:2327:5: error: label ‘next_page’ used but not defined

     goto next_page;

---
v4:
- fix missing unlock_page
v3:
- use READ|WRITE_ONCE
v2:
- add missing code to bypass read

 fs/f2fs/data.c  | 20 ++++++++++++++++++++
 fs/f2fs/f2fs.h  |  1 +
 fs/f2fs/super.c |  2 ++
 3 files changed, 23 insertions(+)
fs/f2fs/data.c +20 -0
index 995cf78b23c5e..066d29938c03a 100644
@@ -2296,6 +2296,7 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	unsigned nr_pages = rac ? readahead_count(rac) : 1;
 	unsigned max_nr_pages = nr_pages;
 	int ret = 0;
+	bool drop_ra = false;
 
 	map.m_pblk = 0;
 	map.m_lblk = 0;
@@ -2306,10 +2307,26 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	map.m_seg_type = NO_CHECK_TYPE;
 	map.m_may_create = false;
 
+	/*
+	 * Two readahead threads for same address range can cause race condition
+	 * which fragments sequential read IOs. So let's avoid each other.
+	 */
+	if (rac && readahead_count(rac)) {
+		if (READ_ONCE(F2FS_I(inode)->ra_offset) == readahead_index(rac))
+			drop_ra = true;
+		else
+			WRITE_ONCE(F2FS_I(inode)->ra_offset,
+						readahead_index(rac));
+	}
+
 	for (; nr_pages; nr_pages--) {
 		if (rac) {
 			page = readahead_page(rac);
 			prefetchw(&page->flags);
+			if (drop_ra) {
+				f2fs_put_page(page, 1);
+				continue;
+			}
 		}
 
 #ifdef CONFIG_F2FS_FS_COMPRESSION
@@ -2372,6 +2389,9 @@ static int f2fs_mpage_readpages(struct inode *inode,
 	}
 	if (bio)
 		__submit_bio(F2FS_I_SB(inode), bio, DATA);
+
+	if (rac && readahead_count(rac) && !drop_ra)
+		WRITE_ONCE(F2FS_I(inode)->ra_offset, -1);
 	return ret;
 }
 
fs/f2fs/f2fs.h +1 -0
index 6a655edeb522f..e6e47618a3576 100644
@@ -809,6 +809,7 @@ struct f2fs_inode_info {
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct task_struct *inmem_task;	/* store inmemory task */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
+	pgoff_t ra_offset;		/* ongoing readahead offset */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
 
 	/* avoid racing between foreground op and gc */
fs/f2fs/super.c +2 -0
index 7326522057378..80cb7cd358f84 100644
@@ -1015,6 +1015,8 @@ static struct inode *f2fs_alloc_inode(struct super_block *sb)
 	/* Will be used by directory only */
 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
 
+	fi->ra_offset = -1;
+
 	return &fi->vfs_inode;
 }
 
Signature · 1 line
2.27.0.212.ge8ba1cc988-goog