error in ext4_mb_release_inode_pa

error in ext4_mb_release_inode_pa

[Eric Sandeen]

This was on #linuxfs last night, and I think I've seen at least one
other report of it:

[22:44]  <shehjart> any ideas why i get the following two lines on the
serial console when writing to ext4 over software raid0:
[22:44]  <shehjart> pa e00001004112d450: logic 11928, phys. 47003288,
len 360
[22:45]  <shehjart> EXT4-fs error (device md0):
ext4_mb_release_inode_pa: free 176, pa_free 174

Just FYI,
-Eric

Re: error in ext4_mb_release_inode_pa

[Andreas Dilger]

On Jul 07, 2008  10:29 -0500, Eric Sandeen wrote:
> This was on #linuxfs last night, and I think I've seen at least one
> other report of it:
> 
> [22:44]  <shehjart> any ideas why i get the following two lines on the
> serial console when writing to ext4 over software raid0:
> [22:44]  <shehjart> pa e00001004112d450: logic 11928, phys. 47003288,
> len 360
> [22:45]  <shehjart> EXT4-fs error (device md0):
> ext4_mb_release_inode_pa: free 176, pa_free 174

The bug that I recalled from Lustre is unlikely to be the same.  It is
https://bugzilla.lustre.org/show_bug.cgi?id=15932

	"error: N blocks in bitmap, M in gd"

There was a second bug in ext3_mb_use_best_found() hit on > 8TB filesystems:
https://bugzilla.lustre.org/show_bug.cgi?id=16101

	BUG_ON(ac->ac_b_ex.fe_group != e3b->bd_group);


Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

Re: error in ext4_mb_release_inode_pa

[Eric Sandeen]

Andreas Dilger wrote:
> On Jul 07, 2008  10:29 -0500, Eric Sandeen wrote:
>> This was on #linuxfs last night, and I think I've seen at least one
>> other report of it:
>>
>> [22:44]  <shehjart> any ideas why i get the following two lines on the
>> serial console when writing to ext4 over software raid0:
>> [22:44]  <shehjart> pa e00001004112d450: logic 11928, phys. 47003288,
>> len 360
>> [22:45]  <shehjart> EXT4-fs error (device md0):
>> ext4_mb_release_inode_pa: free 176, pa_free 174
> 
> The bug that I recalled from Lustre is unlikely to be the same.  It is
> https://bugzilla.lustre.org/show_bug.cgi?id=15932
> 
> 	"error: N blocks in bitmap, M in gd"
> 
> There was a second bug in ext3_mb_use_best_found() hit on > 8TB filesystems:
> https://bugzilla.lustre.org/show_bug.cgi?id=16101
> 
> 	BUG_ON(ac->ac_b_ex.fe_group != e3b->bd_group);
> 

Thanks Andreas -

In an effort to not lose track of it I filed a bug:

http://bugzilla.kernel.org/show_bug.cgi?id=11053

-Eric

Re: error in ext4_mb_release_inode_pa

[Aneesh Kumar K.V]

On Mon, Jul 07, 2008 at 05:20:22PM -0600, Andreas Dilger wrote:
> On Jul 07, 2008  10:29 -0500, Eric Sandeen wrote:
> > This was on #linuxfs last night, and I think I've seen at least one
> > other report of it:
> > 
> > [22:44]  <shehjart> any ideas why i get the following two lines on the
> > serial console when writing to ext4 over software raid0:
> > [22:44]  <shehjart> pa e00001004112d450: logic 11928, phys. 47003288,
> > len 360
> > [22:45]  <shehjart> EXT4-fs error (device md0):
> > ext4_mb_release_inode_pa: free 176, pa_free 174
> 
> The bug that I recalled from Lustre is unlikely to be the same.  It is
> https://bugzilla.lustre.org/show_bug.cgi?id=15932
> 
> 	"error: N blocks in bitmap, M in gd"

The first part of the fix is not needed. I guess we are initializing
block bitmap properly. The second part which states "We cannot trust
find_next_bit() to return a value < max. So we must check its
return for overflow." can be done as below 

How about ?

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index a1e58fb..d2c61eb 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -381,22 +381,28 @@ static inline void mb_clear_bit_atomic(spinlock_t *lock, int bit, void *addr)
 
 static inline int mb_find_next_zero_bit(void *addr, int max, int start)
 {
-	int fix = 0;
+	int fix = 0, ret, tmpmax;
 	addr = mb_correct_addr_and_bit(&fix, addr);
-	max += fix;
+	tmpmax = max + fix;
 	start += fix;
 
-	return ext4_find_next_zero_bit(addr, max, start) - fix;
+	ret = ext4_find_next_zero_bit(addr, tmpmax, start) - fix;
+	if (ret > max)
+		return max;
+	return ret;
 }
 
 static inline int mb_find_next_bit(void *addr, int max, int start)
 {
-	int fix = 0;
+	int fix = 0, ret, tmpmax;
 	addr = mb_correct_addr_and_bit(&fix, addr);
-	max += fix;
+	tmpmax = max + fix;
 	start += fix;
 
-	return ext4_find_next_bit(addr, max, start) - fix;
+	ret = ext4_find_next_bit(addr, tmpmax, start) - fix;
+	if (ret > max)
+		return max;
+	return ret;
 }
 
 static void *mb_find_buddy(struct ext4_buddy *e4b, int order, int *max)
@@ -3633,8 +3639,6 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
 		if (bit >= end)
 			break;
 		next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
-		if (next > end)
-			next = end;
 		start = group * EXT4_BLOCKS_PER_GROUP(sb) + bit +
 				le32_to_cpu(sbi->s_es->s_first_data_block);
 		mb_debug("    free preallocated %u/%u in group %u\n",


> 
> There was a second bug in ext3_mb_use_best_found() hit on > 8TB filesystems:
> https://bugzilla.lustre.org/show_bug.cgi?id=16101
> 
> 	BUG_ON(ac->ac_b_ex.fe_group != e3b->bd_group);
> 

This fix is not needed I guess because we use the ext4_group_t for group
I don't know why the bd_blkbits change is needed

-+	__u16 bd_blkbits;
-+	__u16 bd_group;
++	unsigned bd_group;
++	unsigned bd_blkbits;
 +};


-aneesh