2019-10-02 16:08:45

by Saiyam Doshi

[permalink] [raw]
Subject: [PATCH] staging: exfat: use bdev_sync function directly where needed

fs_sync() is wrapper to bdev_sync(). When fs_sync is called with
non-zero argument, bdev_sync gets called.

Most instances of fs_sync is called with false and very few with
true. Refactor this and makes direct call to bdev_sync() where
needed and removes fs_sync definition.

Signed-off-by: Saiyam Doshi <[email protected]>
---
drivers/staging/exfat/exfat_super.c | 19 +++----------------
1 file changed, 3 insertions(+), 16 deletions(-)

diff --git a/drivers/staging/exfat/exfat_super.c b/drivers/staging/exfat/exfat_super.c
index 229ecabe7a93..3d3b0f0eebdc 100644
--- a/drivers/staging/exfat/exfat_super.c
+++ b/drivers/staging/exfat/exfat_super.c
@@ -285,12 +285,6 @@ static const struct dentry_operations exfat_dentry_ops = {

static DEFINE_SEMAPHORE(z_sem);

-static inline void fs_sync(struct super_block *sb, bool do_sync)
-{
- if (do_sync)
- bdev_sync(sb);
-}
-
/*
* If ->i_mode can't hold S_IWUGO (i.e. ATTR_RO), we use ->i_attrs to
* save ATTR_RO instead of ->i_mode.
@@ -458,7 +452,6 @@ static int ffsUmountVol(struct super_block *sb)
/* acquire the lock for file system critical section */
down(&p_fs->v_sem);

- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);

if (p_fs->vol_type == EXFAT) {
@@ -527,7 +520,9 @@ static int ffsSyncVol(struct super_block *sb, bool do_sync)
down(&p_fs->v_sem);

/* synchronize the file system */
- fs_sync(sb, do_sync);
+ if (do_sync)
+ bdev_sync(sb);
+
fs_set_vol_flags(sb, VOL_CLEAN);

if (p_fs->dev_ejected)
@@ -667,7 +662,6 @@ static int ffsCreateFile(struct inode *inode, char *path, u8 mode,
ret = create_file(inode, &dir, &uni_name, mode, fid);

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1040,7 +1034,6 @@ static int ffsWriteFile(struct inode *inode, struct file_id_t *fid,
}

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1180,7 +1173,6 @@ static int ffsTruncateFile(struct inode *inode, u64 old_size, u64 new_size)
fid->rwoffset = fid->size;

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1328,7 +1320,6 @@ static int ffsMoveFile(struct inode *old_parent_inode, struct file_id_t *fid,
}
out:
#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1390,7 +1381,6 @@ static int ffsRemoveFile(struct inode *inode, struct file_id_t *fid)
fid->flags = (p_fs->vol_type == EXFAT) ? 0x03 : 0x01;

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1479,7 +1469,6 @@ static int ffsSetAttr(struct inode *inode, u32 attr)
}

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -1917,7 +1906,6 @@ static int ffsCreateDir(struct inode *inode, char *path, struct file_id_t *fid)
ret = create_dir(inode, &dir, &uni_name, fid);

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

@@ -2178,7 +2166,6 @@ static int ffsRemoveDir(struct inode *inode, struct file_id_t *fid)
fid->flags = (p_fs->vol_type == EXFAT) ? 0x03 : 0x01;

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

--
2.20.1


2019-10-02 19:06:07

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: [PATCH] staging: exfat: use bdev_sync function directly where needed

On Wed, 02 Oct 2019 20:47:03 +0530, Saiyam Doshi said:
> fs_sync() is wrapper to bdev_sync(). When fs_sync is called with
> non-zero argument, bdev_sync gets called.
>
> Most instances of fs_sync is called with false and very few with
> true. Refactor this and makes direct call to bdev_sync() where
> needed and removes fs_sync definition.

Did you do an actual analysis of where bdev_sync() *should*
be called?

The note in the TODO was intended to point out that many of the calls
are called with 'false' and probably should not be.

#ifdef CONFIG_EXFAT_DELAYED_SYNC
- fs_sync(sb, false);
fs_set_vol_flags(sb, VOL_CLEAN);
#endif

Consider - with this patch, we are now setting the volume state to clean - but
we've not actually flushed the filesystem to disk, which means it's almost
guaranteed that the file system is *NOT* in fact clean.

Changing all the 'false' that happen before a VOL_CLEAN to 'true' is
probably more correct - but still totally wrong if DELAYED_SYNC isn't defined
because then we *aren't* syncing to disk at all.



Attachments:
(No filename) (849.00 B)

2019-10-03 11:17:12

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH] staging: exfat: use bdev_sync function directly where needed

On Wed, Oct 02, 2019 at 08:47:03PM +0530, Saiyam Doshi wrote:
> fs_sync() is wrapper to bdev_sync(). When fs_sync is called with
> non-zero argument, bdev_sync gets called.
>
> Most instances of fs_sync is called with false and very few with
> true. Refactor this and makes direct call to bdev_sync() where
> needed and removes fs_sync definition.
>
> Signed-off-by: Saiyam Doshi <[email protected]>

Reviewed-by: Dan Carpenter <[email protected]>

regards,
dan carpenter

2019-10-03 11:48:08

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH] staging: exfat: use bdev_sync function directly where needed

I replied to your other thread and I added Saiyam Doshi to the CC list
there. Just to be clear this patch is a good cleanup and doesn't affect
runtime at all.

In the other thread, I suggested that we leave fs_sync() as a marker
even though it's dead code. But looking at it now, I think that it's
not really useful. Future auditors should look for places which call
fs_set_vol_flags(sb, VOL_CLEAN); instead. That's exactly the places
which call fs_sync().

regards,
dan carpenter

2019-10-03 11:50:16

by Dan Carpenter

[permalink] [raw]
Subject: Re: [PATCH] staging: exfat: use bdev_sync function directly where needed

Or we could apply your other patch which trumps both this patch and the
patch to the TODO.

regards,
dan carpenter