Re: [PATCH] xfs: fix array bounds checking in log recovery

[Carlos Maiolino]

On Tue, Sep 01, 2026 at 05:39:57PM +0800, Hongling Zeng wrote:
> 
> 在 2026年09月01日 15:06, Carlos Maiolino 写道:
> > On Mon, Aug 24, 2026 at 05:45:35PM +0800, Hongling Zeng wrote:
> > > The log recovery code increments array indices based on bits set in
> > > blf_data_map without verifying that the indices stay within the bounds
> > > of the item->ri_buf array.
> > > 
> > > Since blf_data_map is from untrusted log data, a malicious log could set
> > > many bits while ri_total is small, causing array index overflow.
> > > 
> > > This can result in an out-of-bounds access during log recovery, causing
> > > a kernel crash or memory corruption.
> > > 
> > > This patch adds array bounds checking for ri_buf access in
> > > xlog_recover_do_reg_buffer() and xlog_recover_do_inode_buffer().
> > > 
> > > The check in xlog_recover_do_inode_buffer() is placed right before the
> > > actual array access, not after incrementing the index, to avoid false
> > > positives when processing the last valid region.
> > > 
> > > Fixes: 1094d3f12363 ("xfs: refactor log recovery buffer item dispatch for pass2 commit functions")
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Signed-off-by: Hongling Zeng <zenghongling@xxxxxxxxxx>
> > Hi,
> > 
> > is this a re-send? A v2? Please specify why you are sending the same
> > patche again. This saves a lot of time trying to understand why there
> > are two patches with the same subject in the list.
> Hi ,
> This is the same patch version, but the previous submission failed to
> apply because my local repository was out of date. I updated my tree and
> resubmitted the patch with the same fix.
> 
> Sorry for the confusion.

So this is not the same version, it's a rebase against top of the tree
and that should be properly described.

I'll be updating for-next soon this week, please rebase it again on top
of for-next once I publish it and re-send it.

Cheers,
Carlos

> 
> Thanks,
> Hongling
> > > ---
> > >   fs/xfs/xfs_buf_item_recover.c | 26 ++++++++++++++++++++++++++
> > >   1 file changed, 26 insertions(+)
> > > 
> > > diff --git a/fs/xfs/xfs_buf_item_recover.c b/fs/xfs/xfs_buf_item_recover.c
> > > index 57929f115055..7148716366ba 100644
> > > --- a/fs/xfs/xfs_buf_item_recover.c
> > > +++ b/fs/xfs/xfs_buf_item_recover.c
> > > @@ -485,6 +485,19 @@ xlog_recover_do_reg_buffer(
> > >   			break;
> > >   		nbits = xfs_contig_bits(buf_f->blf_data_map,
> > >   					buf_f->blf_map_size, bit);
> > > +
> > > +		/*
> > > +		 * The bitmap can have more bits set than there are regions
> > > +		 * in ri_buf, so we must check array bounds before using the
> > > +		 * index to access ri_buf[i].
> > > +		 */
> > > +		if (XFS_IS_CORRUPT(mp, i >= item->ri_total)) {
> > > +			xfs_alert(mp,
> > > +		"Buffer log item index (%d) exceeds allocated regions (%d).",
> > > +					i, item->ri_total);
> > > +			return -EFSCORRUPTED;
> > > +		}
> > > +
> > >   		ASSERT(nbits > 0);
> > >   		ASSERT(item->ri_buf[i].iov_base != NULL);
> > >   		ASSERT(item->ri_buf[i].iov_len % XFS_BLF_CHUNK == 0);
> > > @@ -688,6 +701,19 @@ xlog_recover_do_inode_buffer(
> > >   		if (next_unlinked_offset < reg_buf_offset)
> > >   			continue;
> > > +		/*
> > > +		 * Check array bounds here (right before accessing ri_buf)
> > > +		 * rather than after incrementing item_index. This avoids
> > > +		 * incorrectly rejecting logs when item_index reaches
> > > +		 * ri_total after processing the final valid region.
> > > +		 */
> > > +		if (XFS_IS_CORRUPT(mp, item_index >= item->ri_total)) {
> > > +			xfs_alert(mp,
> > > +		"Inode buffer log item index (%d) exceeds allocated regions (%d).",
> > > +				item_index, item->ri_total);
> > > +			return -EFSCORRUPTED;
> > > +		}
> > > +
> > >   		ASSERT(item->ri_buf[item_index].iov_base != NULL);
> > >   		ASSERT((item->ri_buf[item_index].iov_len % XFS_BLF_CHUNK) == 0);
> > >   		ASSERT((reg_buf_offset + reg_buf_bytes) <= BBTOB(bp->b_length));
> > > -- 
> > > 2.25.1
> > > 
> > > 
> 
>