Skip to contentSkip to conversation

xfs: clean up xfs_set_maxicount & use in growfs

PATCH In review Eric Sandeen 4 participants 6 messages Feb 25, 2014 – Feb 27, 2014 XFS filesystem

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 xfs: clean up xfs_set_maxicount & use in growfs

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

I rewrote it using mult_frac (so handy!) and rounddown, to do a nice no-overflow, no-precision-loss multiplication by (%/100), then round down to the inode allocation unit. Then check to see if we've overflowed the u64, and set to XFS_MAXINUMBER if so.

Also, growfs open-coded the setting of maxicount; call the handy helper instead.

This slightly changes growfs behavior, because we weren't rounding to the allocation multiple in that code. I checked that we get the same answers for non-growfs cases by running 100 mkfs.xfs's with imaxpct from 1 to 100, and it gives the same inode count in all cases.

Thanks to dchinner for pointing out that this could use fixing.

---

Ok Brian, have at it. ;)

fs/xfs/xfs_fsops.c +1 -6
index 02fb943..867e476 100644
@@ -471,12 +471,7 @@ xfs_growfs_data_private(
 	/* New allocation groups fully initialized, so update mount struct */
 	if (nagimax)
 		mp->m_maxagi = nagimax;
-	if (mp->m_sb.sb_imax_pct) {
-		__uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
-		do_div(icount, 100);
-		mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
-	} else
-		mp->m_maxicount = 0;
+	xfs_set_maxicount(mp);
 	xfs_set_low_space_thresholds(mp);
 
 	/* update secondary superblocks. */
fs/xfs/xfs_mount.c +8 -8
index 02df7b4..0ec2b19 100644
@@ -419,24 +419,24 @@ xfs_update_alignment(xfs_mount_t *mp)
 }
 
 /*
- * Set the maximum inode count for this filesystem
+ * Set the maximum inode count for this filesystem based on sb_imax_pct
  */
-STATIC void
+void
 xfs_set_maxicount(xfs_mount_t *mp)
 {
 	xfs_sb_t	*sbp = &(mp->m_sb);
-	__uint64_t	icount;
+	__uint64_t	iblocks;
 
 	if (sbp->sb_imax_pct) {
 		/*
 		 * Make sure the maximum inode count is a multiple
 		 * of the units we allocate inodes in.
 		 */
-		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
-		do_div(icount, 100);
-		do_div(icount, mp->m_ialloc_blks);
-		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
-				   sbp->sb_inopblog;
+		iblocks = mult_frac(sbp->sb_dblocks, sbp->sb_imax_pct, 100);
+		iblocks = rounddown(iblocks, mp->m_ialloc_blks);
+		mp->m_maxicount = iblocks << sbp->sb_inopblog;
+		if (mp->m_maxicount < iblocks)
+			mp->m_maxicount = XFS_MAXINUMBER;
 	} else {
 		mp->m_maxicount = 0;
 	}
fs/xfs/xfs_mount.h +1 -0
index a466c5e..6288a56 100644
@@ -388,6 +388,7 @@ extern int	xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *,
 extern int	xfs_mount_log_sb(xfs_mount_t *, __int64_t);
 extern struct xfs_buf *xfs_getsb(xfs_mount_t *, int);
 extern int	xfs_readsb(xfs_mount_t *, int);
+void		xfs_set_maxicount(xfs_mount_t *mp);
 extern void	xfs_freesb(xfs_mount_t *);
 extern int	xfs_fs_writable(xfs_mount_t *);
 extern int	xfs_sb_validate_fsb_count(struct xfs_sb *, __uint64_t);

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#1 lore raw
Quoted 5 lines · Eric Sandeen

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

Would be useful to get this test case into xfstests..

fs/xfs/xfs_mount.c xfs_update_alignment(xfs_mount_t *mp) · line 428

Quoted 7 lines
fs/xfs/xfs_mount.c +2 -2
-STATIC void
+void
 xfs_set_maxicount(xfs_mount_t *mp)
 {
 	xfs_sb_t	*sbp = &(mp->m_sb);
-	__uint64_t	icount;
+	__uint64_t	iblocks;

Seems like this could move into the if clause below.

fs/xfs/xfs_mount.h extern int xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *, · line 394

Quoted 8 lines
fs/xfs/xfs_mount.h +1 -0
@@ -388,6 +388,7 @@ extern int	xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *,
 extern int	xfs_mount_log_sb(xfs_mount_t *, __int64_t);
 extern struct xfs_buf *xfs_getsb(xfs_mount_t *, int);
 extern int	xfs_readsb(xfs_mount_t *, int);
+void		xfs_set_maxicount(xfs_mount_t *mp);
 extern void	xfs_freesb(xfs_mount_t *);
 extern int	xfs_fs_writable(xfs_mount_t *);
 extern int	xfs_sb_validate_fsb_count(struct xfs_sb *, __uint64_t);

A missing extern while all other prototypes around it have one seems rather odd.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#2 lore raw
Quoted 8 lines · Christoph Hellwig
Quoted 5 lines · Eric Sandeen

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

Would be useful to get this test case into xfstests..

Your'e bot, aren't you. ;)

yeah you're right. I went down this path by starting a testcase for 3-patches-ago, and got here. That one can be sent soon, and sure, it'd be worth doing this too.

Quoted 10 lines

fs/xfs/xfs_mount.c xfs_update_alignment(xfs_mount_t *mp) · line 428

Quoted 7 lines
fs/xfs/xfs_mount.c +2 -2
-STATIC void
+void
 xfs_set_maxicount(xfs_mount_t *mp)
 {
 	xfs_sb_t	*sbp = &(mp->m_sb);
-	__uint64_t	icount;
+	__uint64_t	iblocks;

Seems like this could move into the if clause below.

Quoted 11 lines

fs/xfs/xfs_mount.h extern int xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *, · line 394

Quoted 8 lines
fs/xfs/xfs_mount.h +1 -0
@@ -388,6 +388,7 @@ extern int	xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *,
 extern int	xfs_mount_log_sb(xfs_mount_t *, __int64_t);
 extern struct xfs_buf *xfs_getsb(xfs_mount_t *, int);
 extern int	xfs_readsb(xfs_mount_t *, int);
+void		xfs_set_maxicount(xfs_mount_t *mp);
 extern void	xfs_freesb(xfs_mount_t *);
 extern int	xfs_fs_writable(xfs_mount_t *);
 extern int	xfs_sb_validate_fsb_count(struct xfs_sb *, __uint64_t);

A missing extern while all other prototypes around it have one seems rather odd.

Sure does!

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#3 lore raw
In reply to Christoph Hellwig
Quoted 8 lines · Christoph Hellwig
Quoted 5 lines · Eric Sandeen

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

Would be useful to get this test case into xfstests..

Ok so I was going on Dave's assertion about that. ;)

To overflow, we'd need dblocks * 100 to be > 2^64-1:

so dblocks would need to be > (2^64-1)/100

for 4k blocks that's 655 exabytes. Maybe not so possible after all ;)

Dave, maybe just removing the open-code is enough here.

-Eric

Quoted 22 lines

fs/xfs/xfs_mount.c xfs_update_alignment(xfs_mount_t *mp) · line 428

Quoted 7 lines
fs/xfs/xfs_mount.c +2 -2
-STATIC void
+void
 xfs_set_maxicount(xfs_mount_t *mp)
 {
 	xfs_sb_t	*sbp = &(mp->m_sb);
-	__uint64_t	icount;
+	__uint64_t	iblocks;

Seems like this could move into the if clause below.

fs/xfs/xfs_mount.h extern int xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *, · line 394

Quoted 8 lines
fs/xfs/xfs_mount.h +1 -0
@@ -388,6 +388,7 @@ extern int	xfs_mod_incore_sb_batch(xfs_mount_t *, xfs_mod_sb_t *,
 extern int	xfs_mount_log_sb(xfs_mount_t *, __int64_t);
 extern struct xfs_buf *xfs_getsb(xfs_mount_t *, int);
 extern int	xfs_readsb(xfs_mount_t *, int);
+void		xfs_set_maxicount(xfs_mount_t *mp);
 extern void	xfs_freesb(xfs_mount_t *);
 extern int	xfs_fs_writable(xfs_mount_t *);
 extern int	xfs_sb_validate_fsb_count(struct xfs_sb *, __uint64_t);

A missing extern while all other prototypes around it have one seems rather odd.

Quoted 5 lines

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#4 lore raw
Quoted 10 lines · Eric Sandeen
Quoted 8 lines · Christoph Hellwig
Quoted 5 lines · Eric Sandeen

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

Would be useful to get this test case into xfstests..

Quoted 7 lines

Ok so I was going on Dave's assertion about that. ;)

To overflow, we'd need dblocks * 100 to be > 2^64-1:

so dblocks would need to be > (2^64-1)/100

for 4k blocks that's 655 exabytes. Maybe not so possible after all ;)

Until the block count is corrupted by fsfuzzer? ;)

Quoted 1 line

Dave, maybe just removing the open-code is enough here.

Sure, but I still like the conversion to use mult_frac....

Cheers,

Dave.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#5 lore raw
Quoted 17 lines · Dave Chinner
Quoted 16 lines · Eric Sandeen
Quoted 8 lines · Christoph Hellwig
Quoted 5 lines · Eric Sandeen

xfs_set_maxicount() overflowed fairly easily for large filesystems and large maxicount; we started out by multiplying dblocks by the percentage, *then* dividing by 100, and never checked for an overflow. The calculations were also, IMHO, a little hard to follow.

Would be useful to get this test case into xfstests..

Ok so I was going on Dave's assertion about that. ;)

To overflow, we'd need dblocks * 100 to be > 2^64-1:

so dblocks would need to be > (2^64-1)/100

Quoted 7 lines
Quoted 1 line

for 4k blocks that's 655 exabytes. Maybe not so possible after all ;)

Until the block count is corrupted by fsfuzzer? ;)

Quoted 1 line

Dave, maybe just removing the open-code is enough here.

Sure, but I still like the conversion to use mult_frac....

Ok, I do too. Just a bit resistant to fixing what ain't broke. I'll see if I can write up a test to be sure we're doing sane things statfs/growfs with the change.

-Eric

Quoted 4 lines

Cheers,

Dave.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs