This patch-set makes HFS file-system stop using the VFS '->write_supers()'
call-back and the '->s_dirt' superblock field because I plan to remove them
once all users are gone.
Like the other similar patch-sets, we switch to a delayed job for writing out
the superblock instead of using the 's_dirt' flag. Additionally, this patch-set
includes several clean-ups.
Reminder:
The goal is to get rid of the 'sync_supers()' kernel thread. This kernel thread
wakes up every 5 seconds (by default) and calls '->write_super()' for all
mounted file-systems. And the bad thing is that this is done even if all the
superblocks are clean. Moreover, many file-systems do not even need this end
they do not register the '->write_super()' method at all (e.g., btrfs).
So 'sync_supers()' most often just generates useless wake-ups and wastes power.
I am trying to make all file-systems independent of '->write_super()' and plan
to remove 'sync_supers()' and '->write_super()' completely once there are no
more users.
Al, in the past I was trying to upstream patches which optimized 'sync_super()',
but you wanted me to kill it completely instead, which I am trying to do
now, see http://lkml.org/lkml/2010/7/22/96
Tested using the fsstress test from the LTP project.
======
Overall status:
1. ext4: patches submitted, waiting for review from Ted Ts'o:
https://lkml.org/lkml/2012/4/2/111
2. udf: patch submitted, should be in Jan Kara's tree:
https://lkml.org/lkml/2012/6/4/233
3 exofs: patch submitted, not sure if it will go to the exofs tree:
https://lkml.org/lkml/2012/6/4/211
4. affs: patches submitted, should be in Al Viro's tree:
https://lkml.org/lkml/2012/6/6/400
4. ext2: done, see commit f72cf5e223a28d3b3ea7dc9e40464fd534e359e8
5. vfat: done, see commit 78491189ddb6d84d4a4abae992ed891a236d0263
6. jffs2: done, see commit 208b14e507c00ff7f108e1a388dd3d8cc805a443
7. reiserfs: done, see commit 033369d1af1264abc23bea2e174aa47cdd212f6f
TODO: hfsplus, sysv, ufs
======
fs/hfs/extent.c | 2 +-
fs/hfs/hfs_fs.h | 15 +++++++----
fs/hfs/inode.c | 14 +++-------
fs/hfs/mdb.c | 15 +++++++++++
fs/hfs/super.c | 73 +++++++++++++++++++++++++++----------------------------
5 files changed, 65 insertions(+), 54 deletions(-)
Thanks,
Artem.
From: Artem Bityutskiy <[email protected]>
HFS calls 'hfs_write_super()' from 'hfs_put_super()' in order to write the MDB
to the media. However, it is not needed because VFS calls '->sync_fs()' before
calling '->put_super()' - so by the time we are in 'hfs_write_super()', the MDB
is already synchronized.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/super.c | 2 --
1 files changed, 0 insertions(+), 2 deletions(-)
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index f7c06bb..47e4119 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -74,8 +74,6 @@ static int hfs_sync_fs(struct super_block *sb, int wait)
*/
static void hfs_put_super(struct super_block *sb)
{
- if (sb->s_dirt)
- hfs_write_super(sb);
hfs_mdb_close(sb);
/* release the MDB's resources */
hfs_mdb_put(sb);
--
1.7.7.6
From: Artem Bityutskiy <[email protected]>
Add an 'sb' VFS superblock back-reference to the 'struct hfs_sb_info' data
structure - we will need to find the VFS superblock from a
'struct hfs_sb_info' object in the next patch, so this change is jut a
preparation.
Remove few useless newlines while on it.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/hfs_fs.h | 6 +-----
fs/hfs/super.c | 1 +
2 files changed, 2 insertions(+), 5 deletions(-)
diff --git a/fs/hfs/hfs_fs.h b/fs/hfs/hfs_fs.h
index 1bf967c..3515612 100644
--- a/fs/hfs/hfs_fs.h
+++ b/fs/hfs/hfs_fs.h
@@ -137,16 +137,12 @@ struct hfs_sb_info {
gid_t s_gid; /* The gid of all files */
int session, part;
-
struct nls_table *nls_io, *nls_disk;
-
struct mutex bitmap_lock;
-
unsigned long flags;
-
u16 blockoffset;
-
int fs_div;
+ struct super_block *sb;
};
#define HFS_FLG_BITMAP_DIRTY 0
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index 0730135..99c6239 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -380,6 +380,7 @@ static int hfs_fill_super(struct super_block *sb, void *data, int silent)
if (!sbi)
return -ENOMEM;
+ sbi->sb = sb;
sb->s_fs_info = sbi;
res = -EINVAL;
--
1.7.7.6
From: Artem Bityutskiy <[email protected]>
This patch makes hfs stop using the VFS '->write_super()' method along with
the 's_dirt' superblock flag, because they are on their way out.
The whole "superblock write-out" VFS infrastructure is served by the
'sync_supers()' kernel thread, which wakes up every 5 (by default) seconds and
writes out all dirty superblocks using the '->write_super()' call-back. But the
problem with this thread is that it wastes power by waking up the system every
5 seconds, even if there are no diry superblocks, or there are no client
file-systems which would need this (e.g., btrfs does not use
'->write_super()'). So we want to kill it completely and thus, we need to make
file-systems to stop using the '->write_super()' VFS service, and then remove
it together with the kernel thread.
Tested using fsstress from the LTP project.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/extent.c | 2 +-
fs/hfs/hfs_fs.h | 9 ++++++-
fs/hfs/inode.c | 11 +++------
fs/hfs/mdb.c | 2 +
fs/hfs/super.c | 65 +++++++++++++++++++++++++++++-------------------------
5 files changed, 50 insertions(+), 39 deletions(-)
diff --git a/fs/hfs/extent.c b/fs/hfs/extent.c
index 2c16316..a67955a 100644
--- a/fs/hfs/extent.c
+++ b/fs/hfs/extent.c
@@ -432,7 +432,7 @@ out:
if (inode->i_ino < HFS_FIRSTUSER_CNID)
set_bit(HFS_FLG_ALT_MDB_DIRTY, &HFS_SB(sb)->flags);
set_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags);
- sb->s_dirt = 1;
+ hfs_mark_mdb_dirty(sb);
}
return res;
diff --git a/fs/hfs/hfs_fs.h b/fs/hfs/hfs_fs.h
index 3515612..8275175 100644
--- a/fs/hfs/hfs_fs.h
+++ b/fs/hfs/hfs_fs.h
@@ -14,6 +14,7 @@
#include <linux/mutex.h>
#include <linux/buffer_head.h>
#include <linux/fs.h>
+#include <linux/workqueue.h>
#include <asm/byteorder.h>
#include <asm/uaccess.h>
@@ -143,6 +144,9 @@ struct hfs_sb_info {
u16 blockoffset;
int fs_div;
struct super_block *sb;
+ int work_queued; /* non-zero delayed work is queued */
+ struct delayed_work mdb_work; /* MDB flush delayed work */
+ spinlock_t work_lock; /* protects mdb_work and work_queued */
};
#define HFS_FLG_BITMAP_DIRTY 0
@@ -222,6 +226,9 @@ extern int hfs_compare_dentry(const struct dentry *parent,
extern void hfs_asc2mac(struct super_block *, struct hfs_name *, struct qstr *);
extern int hfs_mac2asc(struct super_block *, char *, const struct hfs_name *);
+/* super.c */
+extern void hfs_mark_mdb_dirty(struct super_block *sb);
+
extern struct timezone sys_tz;
/*
@@ -249,7 +256,7 @@ static inline const char *hfs_mdb_name(struct super_block *sb)
static inline void hfs_bitmap_dirty(struct super_block *sb)
{
set_bit(HFS_FLG_BITMAP_DIRTY, &HFS_SB(sb)->flags);
- sb->s_dirt = 1;
+ hfs_mark_mdb_dirty(sb);
}
#define sb_bread512(sb, sec, data) ({ \
diff --git a/fs/hfs/inode.c b/fs/hfs/inode.c
index 629eba4..cb78138 100644
--- a/fs/hfs/inode.c
+++ b/fs/hfs/inode.c
@@ -220,7 +220,7 @@ struct inode *hfs_new_inode(struct inode *dir, struct qstr *name, umode_t mode)
insert_inode_hash(inode);
mark_inode_dirty(inode);
set_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags);
- sb->s_dirt = 1;
+ hfs_mark_mdb_dirty(sb);
return inode;
}
@@ -235,7 +235,7 @@ void hfs_delete_inode(struct inode *inode)
if (HFS_I(inode)->cat_key.ParID == cpu_to_be32(HFS_ROOT_CNID))
HFS_SB(sb)->root_dirs--;
set_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags);
- sb->s_dirt = 1;
+ hfs_mark_mdb_dirty(sb);
return;
}
HFS_SB(sb)->file_count--;
@@ -248,7 +248,7 @@ void hfs_delete_inode(struct inode *inode)
}
}
set_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags);
- sb->s_dirt = 1;
+ hfs_mark_mdb_dirty(sb);
}
void hfs_inode_read_fork(struct inode *inode, struct hfs_extent *ext,
@@ -644,10 +644,7 @@ static int hfs_file_fsync(struct file *filp, loff_t start, loff_t end,
/* sync the superblock to buffers */
sb = inode->i_sb;
- if (sb->s_dirt) {
- sb->s_dirt = 0;
- hfs_mdb_commit(sb);
- }
+ flush_delayed_work_sync(&HFS_SB(sb)->mdb_work);
/* .. finally sync the buffers to disk */
err = sync_blockdev(sb->s_bdev);
if (!ret)
diff --git a/fs/hfs/mdb.c b/fs/hfs/mdb.c
index 5fd51a5..a5cd355 100644
--- a/fs/hfs/mdb.c
+++ b/fs/hfs/mdb.c
@@ -263,6 +263,8 @@ void hfs_mdb_commit(struct super_block *sb)
if (sb->s_flags & MS_RDONLY)
return;
+ cancel_delayed_work(&HFS_SB(sb)->mdb_work);
+
lock_buffer(HFS_SB(sb)->mdb_bh);
if (test_and_clear_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags)) {
/* These parameters may have been modified, so write them back */
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index 99c6239..14528ee 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -29,38 +29,9 @@ static struct kmem_cache *hfs_inode_cachep;
MODULE_LICENSE("GPL");
-/*
- * hfs_write_super()
- *
- * Description:
- * This function is called by the VFS only. When the filesystem
- * is mounted r/w it updates the MDB on disk.
- * Input Variable(s):
- * struct super_block *sb: Pointer to the hfs superblock
- * Output Variable(s):
- * NONE
- * Returns:
- * void
- * Preconditions:
- * 'sb' points to a "valid" (struct super_block).
- * Postconditions:
- * The MDB is marked 'unsuccessfully unmounted' by clearing bit 8 of drAtrb
- * (hfs_put_super() must set this flag!). Some MDB fields are updated
- * and the MDB buffer is written to disk by calling hfs_mdb_commit().
- */
-static void hfs_write_super(struct super_block *sb)
-{
- sb->s_dirt = 0;
-
- /* sync everything to the buffers */
- hfs_mdb_commit(sb);
-}
-
static int hfs_sync_fs(struct super_block *sb, int wait)
{
hfs_mdb_commit(sb);
- sb->s_dirt = 0;
-
return 0;
}
@@ -73,11 +44,44 @@ static int hfs_sync_fs(struct super_block *sb, int wait)
*/
static void hfs_put_super(struct super_block *sb)
{
+ cancel_delayed_work_sync(&HFS_SB(sb)->mdb_work);
hfs_mdb_close(sb);
/* release the MDB's resources */
hfs_mdb_put(sb);
}
+static void flush_mdb(struct work_struct *work)
+{
+ struct hfs_sb_info *sbi;
+ struct super_block *sb;
+
+ sbi = container_of(work, struct hfs_sb_info, mdb_work.work);
+ sb = sbi->sb;
+
+ spin_lock(&sbi->work_lock);
+ sbi->work_queued = 0;
+ spin_unlock(&sbi->work_lock);
+
+ hfs_mdb_commit(sb);
+}
+
+void hfs_mark_mdb_dirty(struct super_block *sb)
+{
+ struct hfs_sb_info *sbi = HFS_SB(sb);
+ unsigned long delay;
+
+ if (sb->s_flags & MS_RDONLY)
+ return;
+
+ spin_lock(&sbi->work_lock);
+ if (!sbi->work_queued) {
+ delay = msecs_to_jiffies(dirty_writeback_interval * 10);
+ queue_delayed_work(system_long_wq, &sbi->mdb_work, delay);
+ sbi->work_queued = 1;
+ }
+ spin_unlock(&sbi->work_lock);
+}
+
/*
* hfs_statfs()
*
@@ -177,7 +181,6 @@ static const struct super_operations hfs_super_operations = {
.write_inode = hfs_write_inode,
.evict_inode = hfs_evict_inode,
.put_super = hfs_put_super,
- .write_super = hfs_write_super,
.sync_fs = hfs_sync_fs,
.statfs = hfs_statfs,
.remount_fs = hfs_remount,
@@ -382,6 +385,8 @@ static int hfs_fill_super(struct super_block *sb, void *data, int silent)
sbi->sb = sb;
sb->s_fs_info = sbi;
+ spin_lock_init(&sbi->work_lock);
+ INIT_DELAYED_WORK(&sbi->mdb_work, flush_mdb);
res = -EINVAL;
if (!parse_options((char *)data, sbi)) {
--
1.7.7.6
From: Artem Bityutskiy <[email protected]>
We have the following pattern in 2 places in HFS
if (!RDONLY)
hfs_mdb_commit();
This patch pushes the RDONLY check down to 'hfs_mdb_commit()'. This will
make the following patches a bit simpler.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/inode.c | 3 +--
fs/hfs/mdb.c | 3 +++
fs/hfs/super.c | 3 +--
3 files changed, 5 insertions(+), 4 deletions(-)
diff --git a/fs/hfs/inode.c b/fs/hfs/inode.c
index daf3d63..629eba4 100644
--- a/fs/hfs/inode.c
+++ b/fs/hfs/inode.c
@@ -646,8 +646,7 @@ static int hfs_file_fsync(struct file *filp, loff_t start, loff_t end,
sb = inode->i_sb;
if (sb->s_dirt) {
sb->s_dirt = 0;
- if (!(sb->s_flags & MS_RDONLY))
- hfs_mdb_commit(sb);
+ hfs_mdb_commit(sb);
}
/* .. finally sync the buffers to disk */
err = sync_blockdev(sb->s_bdev);
diff --git a/fs/hfs/mdb.c b/fs/hfs/mdb.c
index 7a32240..5fd51a5 100644
--- a/fs/hfs/mdb.c
+++ b/fs/hfs/mdb.c
@@ -260,6 +260,9 @@ void hfs_mdb_commit(struct super_block *sb)
{
struct hfs_mdb *mdb = HFS_SB(sb)->mdb;
+ if (sb->s_flags & MS_RDONLY)
+ return;
+
lock_buffer(HFS_SB(sb)->mdb_bh);
if (test_and_clear_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags)) {
/* These parameters may have been modified, so write them back */
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index 47e4119..0730135 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -53,8 +53,7 @@ static void hfs_write_super(struct super_block *sb)
sb->s_dirt = 0;
/* sync everything to the buffers */
- if (!(sb->s_flags & MS_RDONLY))
- hfs_mdb_commit(sb);
+ hfs_mdb_commit(sb);
}
static int hfs_sync_fs(struct super_block *sb, int wait)
--
1.7.7.6
From: Artem Bityutskiy <[email protected]>
HFS uses 'lock_super()'/'unlock_super()' around 'hfs_mdb_commit()' in order
to serialize MDB (Master Directory Block) changes. Push it down to
'hfs_mdb_commit()' in order to simplify the code a bit.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/inode.c | 2 --
fs/hfs/mdb.c | 2 ++
fs/hfs/super.c | 4 ----
3 files changed, 2 insertions(+), 6 deletions(-)
diff --git a/fs/hfs/inode.c b/fs/hfs/inode.c
index 761ec06..daf3d63 100644
--- a/fs/hfs/inode.c
+++ b/fs/hfs/inode.c
@@ -645,11 +645,9 @@ static int hfs_file_fsync(struct file *filp, loff_t start, loff_t end,
/* sync the superblock to buffers */
sb = inode->i_sb;
if (sb->s_dirt) {
- lock_super(sb);
sb->s_dirt = 0;
if (!(sb->s_flags & MS_RDONLY))
hfs_mdb_commit(sb);
- unlock_super(sb);
}
/* .. finally sync the buffers to disk */
err = sync_blockdev(sb->s_bdev);
diff --git a/fs/hfs/mdb.c b/fs/hfs/mdb.c
index 1563d5c..3f558d58 100644
--- a/fs/hfs/mdb.c
+++ b/fs/hfs/mdb.c
@@ -260,6 +260,7 @@ void hfs_mdb_commit(struct super_block *sb)
{
struct hfs_mdb *mdb = HFS_SB(sb)->mdb;
+ lock_super(sb);
if (test_and_clear_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags)) {
/* These parameters may have been modified, so write them back */
mdb->drLsMod = hfs_mtime();
@@ -317,6 +318,7 @@ void hfs_mdb_commit(struct super_block *sb)
size -= len;
}
}
+ unlock_super(sb);
}
void hfs_mdb_close(struct super_block *sb)
diff --git a/fs/hfs/super.c b/fs/hfs/super.c
index 7b4c537..f7c06bb 100644
--- a/fs/hfs/super.c
+++ b/fs/hfs/super.c
@@ -50,21 +50,17 @@ MODULE_LICENSE("GPL");
*/
static void hfs_write_super(struct super_block *sb)
{
- lock_super(sb);
sb->s_dirt = 0;
/* sync everything to the buffers */
if (!(sb->s_flags & MS_RDONLY))
hfs_mdb_commit(sb);
- unlock_super(sb);
}
static int hfs_sync_fs(struct super_block *sb, int wait)
{
- lock_super(sb);
hfs_mdb_commit(sb);
sb->s_dirt = 0;
- unlock_super(sb);
return 0;
}
--
1.7.7.6
From: Artem Bityutskiy <[email protected]>
Stop using lock_super for serializing the MDB changes - use the buffer-head own
lock instead. Tested with fsstress.
Signed-off-by: Artem Bityutskiy <[email protected]>
---
fs/hfs/mdb.c | 12 ++++++++++--
1 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/fs/hfs/mdb.c b/fs/hfs/mdb.c
index 3f558d58..7a32240 100644
--- a/fs/hfs/mdb.c
+++ b/fs/hfs/mdb.c
@@ -260,7 +260,7 @@ void hfs_mdb_commit(struct super_block *sb)
{
struct hfs_mdb *mdb = HFS_SB(sb)->mdb;
- lock_super(sb);
+ lock_buffer(HFS_SB(sb)->mdb_bh);
if (test_and_clear_bit(HFS_FLG_MDB_DIRTY, &HFS_SB(sb)->flags)) {
/* These parameters may have been modified, so write them back */
mdb->drLsMod = hfs_mtime();
@@ -284,9 +284,13 @@ void hfs_mdb_commit(struct super_block *sb)
&mdb->drXTFlSize, NULL);
hfs_inode_write_fork(HFS_SB(sb)->cat_tree->inode, mdb->drCTExtRec,
&mdb->drCTFlSize, NULL);
+
+ lock_buffer(HFS_SB(sb)->alt_mdb_bh);
memcpy(HFS_SB(sb)->alt_mdb, HFS_SB(sb)->mdb, HFS_SECTOR_SIZE);
HFS_SB(sb)->alt_mdb->drAtrb |= cpu_to_be16(HFS_SB_ATTRIB_UNMNT);
HFS_SB(sb)->alt_mdb->drAtrb &= cpu_to_be16(~HFS_SB_ATTRIB_INCNSTNT);
+ unlock_buffer(HFS_SB(sb)->alt_mdb_bh);
+
mark_buffer_dirty(HFS_SB(sb)->alt_mdb_bh);
sync_dirty_buffer(HFS_SB(sb)->alt_mdb_bh);
}
@@ -309,7 +313,11 @@ void hfs_mdb_commit(struct super_block *sb)
break;
}
len = min((int)sb->s_blocksize - off, size);
+
+ lock_buffer(bh);
memcpy(bh->b_data + off, ptr, len);
+ unlock_buffer(bh);
+
mark_buffer_dirty(bh);
brelse(bh);
block++;
@@ -318,7 +326,7 @@ void hfs_mdb_commit(struct super_block *sb)
size -= len;
}
}
- unlock_super(sb);
+ unlock_buffer(HFS_SB(sb)->mdb_bh);
}
void hfs_mdb_close(struct super_block *sb)
--
1.7.7.6