Skip to contentSkip to conversation

UBIFS: kill BKL

PATCH In review Artem Bityutskiy 4 participants 8 messages Jul 19, 2009 – Jul 20, 2009

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

PATCH 1/1 UBIFS: kill BKL

From: Artem Bityutskiy <[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

---
 fs/ubifs/super.c |   15 ++-------------
 1 files changed, 2 insertions(+), 13 deletions(-)
fs/ubifs/super.c +2 -13
index 26d2e0d..4ad992f 100644
@@ -1726,8 +1726,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
 		  c->vi.vol_id);
 
-	lock_kernel();
-
 	/*
 	 * The following asserts are only valid if there has not been a failure
 	 * of the media. For example, there will be dirty inodes if we failed
@@ -1792,8 +1790,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubi_close_volume(c->ubi);
 	mutex_unlock(&c->umount_mutex);
 	kfree(c);
-
-	unlock_kernel();
 }
 
 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
@@ -1809,24 +1805,18 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 		return err;
 	}
 
-	lock_kernel();
 	if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
 		if (c->ro_media) {
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;
 		}
 		err = ubifs_remount_rw(c);
-		if (err) {
-			unlock_kernel();
+		if (err)
 			return err;
-		}
 	} else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY)) {
-		if (c->ro_media) {
+		if (c->ro_media)
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;
-		}
 		ubifs_remount_ro(c);
 	}
 
@@ -1839,7 +1829,6 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 	}
 
 	ubifs_assert(c->lst.taken_empty_lebs > 0);
-	unlock_kernel();
 	return 0;
 }
 
Signature · 1 line
1.6.0.6
#1 lore raw
Quoted 9 lines · Artem Bityutskiy

From: Artem Bityutskiy <[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

---

Looks like protection is always provided by sb->s_umount

Missing {} below, btw.

Quoted 35 lines
 fs/ubifs/super.c |   15 ++-------------
 1 files changed, 2 insertions(+), 13 deletions(-)
fs/ubifs/super.c +0 -6
index 26d2e0d..4ad992f 100644
@@ -1726,8 +1726,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
 		  c->vi.vol_id);
 
-	lock_kernel();
-
 	/*
 	 * The following asserts are only valid if there has not been a failure
 	 * of the media. For example, there will be dirty inodes if we failed
@@ -1792,8 +1790,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubi_close_volume(c->ubi);
 	mutex_unlock(&c->umount_mutex);
 	kfree(c);
-
-	unlock_kernel();
 }
 
 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
@@ -1809,24 +1805,18 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 		return err;
 	}
 
-	lock_kernel();
 	if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
 		if (c->ro_media) {
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;

fs/ubifs/super.c static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data) · line 1817

Quoted 10 lines
fs/ubifs/super.c +2 -4
 		}
 		err = ubifs_remount_rw(c);
-		if (err) {
-			unlock_kernel();
+		if (err)
 			return err;
-		}
 	} else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY)) {
-		if (c->ro_media) {
+		if (c->ro_media)

Missing {}

Quoted 15 lines
fs/ubifs/super.c +0 -3
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;
-		}
 		ubifs_remount_ro(c);
 	}
 
@@ -1839,7 +1829,6 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 	}
 
 	ubifs_assert(c->lst.taken_empty_lebs > 0);
-	unlock_kernel();
 	return 0;
 }
 
#2 lore raw
Quoted 8 lines · Adrian Hunter
Quoted 7 lines · Artem Bityutskiy

From: Artem Bityutskiy <[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

Quoted 6 lines
Quoted 2 lines

---

Looks like protection is always provided by sb->s_umount

Missing {} below, btw.

Ugrh. Right. Will fix shortly.

Signature · 6 lines
Best Regards,
Artem Bityutskiy (Артём Битюцкий)
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#3 lore raw
In reply to Adrian Hunter
Quoted 8 lines · Hunter Adrian (Nokia-D/Helsinki)
Quoted 7 lines · Artem Bityutskiy

From: Artem Bityutskiy<[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

Quoted 4 lines
Quoted 2 lines

---

Looks like protection is always provided by sb->s_umount

Err, then what would be a possible reason we would need BKL?

Signature · 6 lines
Best Regards,
Artem Bityutskiy (Артём Битюцкий)
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#4 lore raw
Quoted 9 lines · Bityutskiy Artem (Nokia-D/Helsinki)
Quoted 8 lines · Hunter Adrian (Nokia-D/Helsinki)
Quoted 7 lines · Artem Bityutskiy

From: Artem Bityutskiy<[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

Quoted 5 lines
Quoted 3 lines
Quoted 2 lines

---

Looks like protection is always provided by sb->s_umount

Err, then what would be a possible reason we would need BKL?

Say if we had used BKL instead of c->umount_mutex. i.e. if we had relied on BKL in some other code to provide synchronisation with unmounting/remounting etc

#5 lore raw
In reply to Adrian Hunter
Quoted 8 lines · Adrian Hunter
Quoted 7 lines · Artem Bityutskiy

From: Artem Bityutskiy <[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

Quoted 4 lines
Quoted 2 lines

---

Looks like protection is always provided by sb->s_umount

Not for 2 or more instances of fs mounted/umounted/remounted at the same time, which I meant in my comment. But I've fixed the comment.

Quoted 1 line

Missing {} below, btw.

New version is below.

Quoted 1 line

From 347a38db88429400f0f479dc4d7de2b673999433 Mon Sep 17 00:00:00 2001

From: Artem Bityutskiy <[email protected]>
Date: Sun, 19 Jul 2009 13:51:04 +0300
Subject: [PATCH] UBIFS: kill BKL

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount and prevented more than one instance of the same file-system from doing mount/umount/remount at the same time. But it is OK for UBIFS and it does not need any additional locking for these cases. Thus, kick the BKL out of UBIFS.

---
Changelog: fixed parentheses
remove #include <linux/smp_lock.h>

 fs/ubifs/super.c |   13 +------------
 1 files changed, 1 insertions(+), 12 deletions(-)
fs/ubifs/super.c +1 -12
index 26d2e0d..13e7ed4 100644
@@ -36,7 +36,6 @@
 #include <linux/mount.h>
 #include <linux/math64.h>
 #include <linux/writeback.h>
-#include <linux/smp_lock.h>
 #include "ubifs.h"
 
 /*
@@ -1726,8 +1725,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubifs_msg("un-mount UBI device %d, volume %d", c->vi.ubi_num,
 		  c->vi.vol_id);
 
-	lock_kernel();
-
 	/*
 	 * The following asserts are only valid if there has not been a failure
 	 * of the media. For example, there will be dirty inodes if we failed
@@ -1792,8 +1789,6 @@ static void ubifs_put_super(struct super_block *sb)
 	ubi_close_volume(c->ubi);
 	mutex_unlock(&c->umount_mutex);
 	kfree(c);
-
-	unlock_kernel();
 }
 
 static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
@@ -1809,22 +1804,17 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 		return err;
 	}
 
-	lock_kernel();
 	if ((sb->s_flags & MS_RDONLY) && !(*flags & MS_RDONLY)) {
 		if (c->ro_media) {
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;
 		}
 		err = ubifs_remount_rw(c);
-		if (err) {
-			unlock_kernel();
+		if (err)
 			return err;
-		}
 	} else if (!(sb->s_flags & MS_RDONLY) && (*flags & MS_RDONLY)) {
 		if (c->ro_media) {
 			ubifs_msg("cannot re-mount due to prior errors");
-			unlock_kernel();
 			return -EROFS;
 		}
 		ubifs_remount_ro(c);
@@ -1839,7 +1829,6 @@ static int ubifs_remount_fs(struct super_block *sb, int *flags, char *data)
 	}
 
 	ubifs_assert(c->lst.taken_empty_lebs > 0);
-	unlock_kernel();
 	return 0;
 }
 
Signature · 1 line
1.6.0.6
#6 lore raw
In reply to Artem Bityutskiy
Quoted 6 lines · Artem Bityutskiy

From: Artem Bityutskiy <[email protected]>

The BKL was pushed down from VFS to the file-systems. It used to serialize mount/unmount/remount. UBIFS must be safe if several file-systems are mounted/unmounted/re-mounted at the same time, so kill kick the BKL out of UBIFS.

On something slightly related - anohter part of the super_ops rework in 2.6.31 was that we now always push data out before calling into ->sync_fs to synchronize the metadata.

This means the generic_sync_sb_inodes call in ubifs_sync_fs should go away now.

#7 lore raw
Quoted 6 lines · Christoph Hellwig

On something slightly related - anohter part of the super_ops rework in 2.6.31 was that we now always push data out before calling into ->sync_fs to synchronize the metadata.

This means the generic_sync_sb_inodes call in ubifs_sync_fs should go away now.

Yes, indeed '__sync_filesystem()' invokes 'sync_inodes_sb()' before '->sync_sb()'. Thanks for pointing this out. Below is the patch.

Quoted 1 line

From c14ed92bf8639c8533f8e923859b136d0671db1b Mon Sep 17 00:00:00 2001

From: Artem Bityutskiy <[email protected]>
Date: Mon, 20 Jul 2009 17:56:19 +0300
Subject: [PATCH] UBIFS: remove unneeded call from ubifs_sync_fs

Nowadays VFS always synchronizes all dirty inodes and pages before calling '->sync_fs()', so remove unneeded 'generic_sync_sb_inodes()' from 'ubifs_sync_fs()'. It used to be needed, but not any longer.

---
 fs/ubifs/super.c |   24 +++++++-----------------
 1 files changed, 7 insertions(+), 17 deletions(-)
fs/ubifs/super.c +7 -17
index 13e7ed4..b541bd7 100644
@@ -437,12 +437,6 @@ static int ubifs_sync_fs(struct super_block *sb, int wait)
 {
 	int i, err;
 	struct ubifs_info *c = sb->s_fs_info;
-	struct writeback_control wbc = {
-		.sync_mode   = WB_SYNC_ALL,
-		.range_start = 0,
-		.range_end   = LLONG_MAX,
-		.nr_to_write = LONG_MAX,
-	};
 
 	/*
 	 * Zero @wait is just an advisory thing to help the file system shove
@@ -453,17 +447,6 @@ static int ubifs_sync_fs(struct super_block *sb, int wait)
 		return 0;
 
 	/*
-	 * VFS calls '->sync_fs()' before synchronizing all dirty inodes and
-	 * pages, so synchronize them first, then commit the journal. Strictly
-	 * speaking, it is not necessary to commit the journal here,
-	 * synchronizing write-buffers would be enough. But committing makes
-	 * UBIFS free space predictions much more accurate, so we want to let
-	 * the user be able to get more accurate results of 'statfs()' after
-	 * they synchronize the file system.
-	 */
-	generic_sync_sb_inodes(sb, &wbc);
-
-	/*
 	 * Synchronize write buffers, because 'ubifs_run_commit()' does not
 	 * do this if it waits for an already running commit.
 	 */
@@ -473,6 +456,13 @@ static int ubifs_sync_fs(struct super_block *sb, int wait)
 			return err;
 	}
 
+	/*
+	 * Strictly speaking, it is not necessary to commit the journal here,
+	 * synchronizing write-buffers would be enough. But committing makes
+	 * UBIFS free space predictions much more accurate, so we want to let
+	 * the user be able to get more accurate results of 'statfs()' after
+	 * they synchronize the file system.
+	 */
 	err = ubifs_run_commit(c);
 	if (err)
 		return err;
Signature · 10 lines
1.6.0.6

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)

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