In fs/xfs/xfs_rtalloc.c::xfs_growfs_rt() there's a completely useless
conditional at the error_exit label.
The 'if (nmp)' check is pointless and might as well be removed for two
reasons.
1) if 'nmp' is NULL then kmem_free() will end up calling kfree() with a NULL
argument - which in turn will just cause a return from kfree(). No harm
done.
2) At the beginning of the function there's an assignment; '*nmp = *mp;' so
if 'nmp' was NULL we'd already have blown up due to dereferencing a NULL
pointer.
This patch gets rid of the pointless check.
Signed-off-by: Jesper Juhl <[email protected]>
---
fs/xfs/xfs_rtalloc.c | 3 +--
1 files changed, 1 insertion(+), 2 deletions(-)
--- linux-2.6.18-rc4-orig/fs/xfs/xfs_rtalloc.c 2006-08-11 00:11:13.000000000 +0200
+++ linux-2.6.18-rc4/fs/xfs/xfs_rtalloc.c 2006-08-13 00:07:43.000000000 +0200
@@ -2107,8 +2107,7 @@ xfs_growfs_rt(
* Error paths come here.
*/
error_exit:
- if (nmp)
- kmem_free(nmp, sizeof(*nmp));
+ kmem_free(nmp, sizeof(*nmp));
xfs_trans_cancel(tp, cancelflags);
return error;
}
On Sun, Aug 13, 2006 at 12:16:50AM +0200, Jesper Juhl wrote:
> In fs/xfs/xfs_rtalloc.c::xfs_growfs_rt() there's a completely useless
> conditional at the error_exit label.
> The 'if (nmp)' check is pointless and might as well be removed for two
> reasons.
>
> 1) if 'nmp' is NULL then kmem_free() will end up calling kfree() with a NULL
> argument - which in turn will just cause a return from kfree(). No harm
> done.
Thats valid.
> 2) At the beginning of the function there's an assignment; '*nmp = *mp;' so
Thats not. Theres no assignment at the start of the function;
theres one inside the main body of the loop 20+ lines into it,
and right after a mem alloc with flags requiring no failure.
Later that local variable is freed then set to NULL inside the
loop, before continuing the next iteration...
Really this code would be better if reworked slightly to just
allocate nmp once before entering the loop, and then free it
once at the end... we wouldn't need a goto, just a few breaks
in the loop and a conditional transaction cancel.
> This patch gets rid of the pointless check.
Hmm, seems like code churn that makes the code slightly less
obvious, but thats just me... I'd prefer a tested patch that
implements the above suggestion, to be honest. :)
cheers.
--
Nathan
On 14/08/06, Nathan Scott <[email protected]> wrote:
> On Sun, Aug 13, 2006 at 12:16:50AM +0200, Jesper Juhl wrote:
> > In fs/xfs/xfs_rtalloc.c::xfs_growfs_rt() there's a completely useless
> > conditional at the error_exit label.
> > The 'if (nmp)' check is pointless and might as well be removed for two
> > reasons.
> >
> > 1) if 'nmp' is NULL then kmem_free() will end up calling kfree() with a NULL
> > argument - which in turn will just cause a return from kfree(). No harm
> > done.
>
> Thats valid.
>
> > 2) At the beginning of the function there's an assignment; '*nmp = *mp;' so
>
> Thats not. Theres no assignment at the start of the function;
> theres one inside the main body of the loop 20+ lines into it,
> and right after a mem alloc with flags requiring no failure.
> Later that local variable is freed then set to NULL inside the
> loop, before continuing the next iteration...
>
> Really this code would be better if reworked slightly to just
> allocate nmp once before entering the loop, and then free it
> once at the end... we wouldn't need a goto, just a few breaks
> in the loop and a conditional transaction cancel.
>
> > This patch gets rid of the pointless check.
>
> Hmm, seems like code churn that makes the code slightly less
> obvious, but thats just me... I'd prefer a tested patch that
> implements the above suggestion, to be honest. :)
>
Ok, I'll see what I can come up with.
--
Jesper Juhl <[email protected]>
Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please http://www.expita.com/nomime.html
On Monday 14 August 2006 09:25, Jesper Juhl wrote:
> On 14/08/06, Nathan Scott <[email protected]> wrote:
> > On Sun, Aug 13, 2006 at 12:16:50AM +0200, Jesper Juhl wrote:
> >
> > Really this code would be better if reworked slightly to just
> > allocate nmp once before entering the loop, and then free it
> > once at the end... we wouldn't need a goto, just a few breaks
> > in the loop and a conditional transaction cancel.
> >
> > > This patch gets rid of the pointless check.
> >
> > Hmm, seems like code churn that makes the code slightly less
> > obvious, but thats just me... I'd prefer a tested patch that
> > implements the above suggestion, to be honest. :)
> >
> Ok, I'll see what I can come up with.
>
How this?
Compile tested only since I'm at home and don't have any XFS filesystems to
play with atm.
Rework fs/xfs/xfs_rtalloc.c::xfs_growfs_rt() to allocate and free 'nmp' just
once and make the error handling a bit clearer.
Signed-off-by: Jesper Juhl <[email protected]>
---
fs/xfs/xfs_rtalloc.c | 37 +++++++++++++++----------------------
1 files changed, 15 insertions(+), 22 deletions(-)
--- linux-2.6.18-rc4-orig/fs/xfs/xfs_rtalloc.c 2006-08-11 00:11:13.000000000 +0200
+++ linux-2.6.18-rc4/fs/xfs/xfs_rtalloc.c 2006-08-16 22:36:03.000000000 +0200
@@ -1976,7 +1976,11 @@ xfs_growfs_rt(
if ((error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks,
mp->m_sb.sb_rsumino)))
return error;
- nmp = NULL;
+
+ /*
+ * Allocate a new (fake) mount/sb.
+ */
+ nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
/*
* Loop over the bitmap blocks.
* We will do everything one bitmap block at a time.
@@ -1987,10 +1991,6 @@ xfs_growfs_rt(
((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
bmbno < nrbmblocks;
bmbno++) {
- /*
- * Allocate a new (fake) mount/sb.
- */
- nmp = kmem_alloc(sizeof(*nmp), KM_SLEEP);
*nmp = *mp;
nsbp = &nmp->m_sb;
/*
@@ -2018,13 +2018,13 @@ xfs_growfs_rt(
cancelflags = 0;
if ((error = xfs_trans_reserve(tp, 0,
XFS_GROWRTFREE_LOG_RES(nmp), 0, 0, 0)))
- goto error_exit;
+ break;
/*
* Lock out other callers by grabbing the bitmap inode lock.
*/
if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0,
XFS_ILOCK_EXCL, &ip)))
- goto error_exit;
+ break;
ASSERT(ip == mp->m_rbmip);
/*
* Update the bitmap inode's size.
@@ -2038,7 +2038,7 @@ xfs_growfs_rt(
*/
if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rsumino, 0,
XFS_ILOCK_EXCL, &ip)))
- goto error_exit;
+ break;
ASSERT(ip == mp->m_rsumip);
/*
* Update the summary inode's size.
@@ -2053,7 +2053,7 @@ xfs_growfs_rt(
mp->m_rsumlevels != nmp->m_rsumlevels) {
error = xfs_rtcopy_summary(mp, nmp, tp);
if (error)
- goto error_exit;
+ break;
}
/*
* Update superblock fields.
@@ -2080,18 +2080,13 @@ xfs_growfs_rt(
error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
if (error)
- goto error_exit;
+ break;
/*
* Mark more blocks free in the superblock.
*/
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
nsbp->sb_rextents - sbp->sb_rextents);
/*
- * Free the fake mp structure.
- */
- kmem_free(nmp, sizeof(*nmp));
- nmp = NULL;
- /*
* Update mp values into the real mp structure.
*/
mp->m_rsumlevels = nrsumlevels;
@@ -2101,15 +2096,13 @@ xfs_growfs_rt(
*/
xfs_trans_commit(tp, 0, NULL);
}
- return 0;
-
+ if (error)
+ xfs_trans_cancel(tp, cancelflags);
/*
- * Error paths come here.
+ * Free the fake mp structure.
*/
-error_exit:
- if (nmp)
- kmem_free(nmp, sizeof(*nmp));
- xfs_trans_cancel(tp, cancelflags);
+ kmem_free(nmp, sizeof(*nmp));
+
return error;
}
On Wed, Aug 16, 2006 at 10:44:19PM +0200, Jesper Juhl wrote:
> On Monday 14 August 2006 09:25, Jesper Juhl wrote:
> > On 14/08/06, Nathan Scott <[email protected]> wrote:
> > > On Sun, Aug 13, 2006 at 12:16:50AM +0200, Jesper Juhl wrote:
> > >
> > > > This patch gets rid of the pointless check.
> > >
> > > Hmm, seems like code churn that makes the code slightly less
> > > obvious, but thats just me... I'd prefer a tested patch that
> > > implements the above suggestion, to be honest. :)
> > >
> > Ok, I'll see what I can come up with.
> >
>
> How this?
>
> Compile tested only since I'm at home and don't have any XFS filesystems to
> play with atm.
It looks good. I've tested it, and it works fine - I'll merge it
into my tree shortly. Thanks for following up on this.
cheers.
--
Nathan