[PATCH] xfs_repair: junk last entry in sf dir if name starts beyond dir size

[Rui Gomes]

Hello Eric, 

Sorry for the late late reply, I didn't had the time to dig in to this earlier.

Actually gdb was lying us, the segfault doesn't happen at:
  for (i = 0; i < be32_to_cpu(btp->count); i++) 

but a bit later at:

if (be32_to_cpu(lep[i].address) == addr &&
        be32_to_cpu(lep[i].hashval) == hash)


And the cause of the segfault is lep[i] 

So I tried:

(gdb) print lep
$1 = (xfs_dir2_leaf_entry_t *) 0xfffffffc9ac12788

p lep[0].address
Cannot access memory at address 0xfffffffc9ac12794

For what I can see the lep[0] struct doesn't exist!

The inode where this happen bellow:

[root at icess8a xfsprogs-dev]# xfs_db -c "inode 620507648" -c "p" /dev/sdb1
Metadata corruption detected at block 0x4ffed6d08/0x1000                                                                                                                                                                                     
xfs_db: cannot init perag data (117). Continuing anyway.                                                                                                                                                                                     
core.magic = 0x494e                                                                                                                                                                                                                          
core.mode = 040755                                                                                                                                                                                                                           
core.version = 2                                                                                                                                                                                                                             
core.format = 2 (extents)
core.nlinkv2 = 3
core.onlink = 0
core.projid_lo = 0
core.projid_hi = 0
core.uid = 0
core.gid = 0
core.flushiter = 2
core.atime.sec = Fri May 16 12:21:52 2014
core.atime.nsec = 779442171
core.mtime.sec = Tue Mar 24 12:03:59 2009
core.mtime.nsec = 000000000
core.ctime.sec = Fri Feb 28 19:54:03 2014
core.ctime.nsec = 736630717
core.size = 4096
core.nblocks = 1
core.extsize = 0
core.nextents = 1
core.naextents = 0
core.forkoff = 0
core.aformat = 2 (extents)
core.dmevmask = 0
core.dmstate = 0
core.newrtbm = 0
core.prealloc = 0
core.realtime = 0
core.immutable = 0
core.append = 0
core.sync = 0
core.noatime = 0
core.nodump = 0
core.rtinherit = 0
core.projinherit = 0
core.nosymlinks = 0
core.extsz = 0
core.extszinherit = 0
core.nodefrag = 0
core.filestream = 0
core.gen = 3064228498
next_unlinked = null
u.bmx[0] = [startoff,startblock,blockcount,extentflag] 0:[0,38781727,1,0]


Regards 

------------------------------- 
Rui Gomes 
CTO 


RVX - Reykjavik Visual Effects 
Seljavegur 2, 
101 Reykjavik 
Iceland 


Tel: + 354 527 3330 
Mob: + 354 663 3360

----- Original Message -----
From: "Rui Gomes" <rgomes at rvx.is>
To: "Eric Sandeen" <sandeen at sandeen.net>
Cc: "omar" <omar at rvx.is>, "xfs" <xfs at oss.sgi.com>
Sent: Wednesday, 11 March, 2015 16:01:10
Subject: Re: [PATCH] xfs_repair: junk last entry in sf dir if name starts beyond dir size

Hi,

Thank you for pointing out where to look, I will try to dissect this a bit further and report back to you. 

Regards 

------------------------------- 
Rui Gomes 
CTO 


RVX - Reykjavik Visual Effects 
Seljavegur 2, 
101 Reykjavik 
Iceland 


Tel: + 354 527 3330 
Mob: + 354 663 3360

----- Original Message -----
From: "Eric Sandeen" <sandeen at sandeen.net>
To: "Rui Gomes" <rgomes at rvx.is>
Cc: "omar" <omar at rvx.is>, "xfs" <xfs at oss.sgi.com>
Sent: Wednesday, 11 March, 2015 15:46:50
Subject: Re: [PATCH] xfs_repair: junk last entry in sf dir if name starts beyond dir size

On 3/11/15 11:04 AM, Rui Gomes wrote:
> Program received signal SIGSEGV, Segmentation fault.
> 0x000000000044dbcd in __xfs_dir3_data_check (dp=dp at entry=0x0, bp=bp at entry=0x1538e810) at xfs_dir2_data.c:148
> 148                             for (i = 0; i < be32_to_cpu(btp->count); i++) {
> (gdb) bt
> #0  0x000000000044dbcd in __xfs_dir3_data_check (dp=dp at entry=0x0, bp=bp at entry=0x1538e810) at xfs_dir2_data.c:148
> #1  0x000000000044b9ea in xfs_dir3_block_verify (bp=bp at entry=0x1538e810) at xfs_dir2_block.c:62

Can you do a little digging around in gdb to sort out more
about why it segfaulted?

We got a xfs_dir2_data_hdr_t from the passed-in bp w/ valid magic:

hdr = bp->b_addr;

and from that got btp:

        switch (hdr->magic) {
        case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
                btp = xfs_dir2_block_tail_p(mp, hdr);

and this just finds an offset from hdr:

                ((char *)hdr + mp->m_dirblksize)) - 1;

but then apparently blew up when we tried to use btp:

for (i = 0; i < be32_to_cpu(btp->count); i++) {

I don't see offhand how the hdr is ok, with good magic, but an
offset from the hdr (btp) is causing a segfault.  Can you dig around
a bit more in gdb?

-Eric