Re: [PATCH] ext4: Try to initialize all groups we can in case of failure on ppc64

[Andreas Dilger <adilger@dilger.ca>]

On May 27, 2015, at 9:44 AM, Eric Sandeen <sandeen@redhat.com> wrote:
> 
> On 5/27/15 5:25 AM, Lukas Czerner wrote:
>> Currently on the machines with page size > block size when initializing
>> block group buddy cache we initialize it for all the block group bitmaps
>> in the page. However in the case of read error, checksum error, or if
>> a single bitmap is in any way corrupted we would fail to initialize all
>> of the bitmaps. This is problematic because we will not have access to
>> the other allocation groups even though those might be perfectly fine
>> and usable.
>> 
>> Fix this by reading all the bitmaps instead of error out on the first
>> problem and simply skip the bitmaps which were either not read properly,
>> or are not valid.
>> 
>> Signed-off-by: Lukas Czerner <lczerner@redhat.com>
>> ---
>> fs/ext4/mballoc.c | 11 ++++++++---
>> 1 file changed, 8 insertions(+), 3 deletions(-)
>> 
>> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
>> index 8d1e602..7e28007 100644
>> --- a/fs/ext4/mballoc.c
>> +++ b/fs/ext4/mballoc.c
>> @@ -882,10 +882,8 @@ static int ext4_mb_init_cache(struct page *page, char *incore)
>> 
>> 	/* wait for I/O completion */
>> 	for (i = 0, group = first_group; i < groups_per_page; i++, group++) {
>> -		if (bh[i] && ext4_wait_block_bitmap(sb, group, bh[i])) {
>> +		if (bh[i] && ext4_wait_block_bitmap(sb, group, bh[i]))
>> 			err = -EIO;
>> -			goto out;
>> -		}
>> 	}
>> 
>> 	first_block = page->index * blocks_per_page;
>> @@ -898,6 +896,13 @@ static int ext4_mb_init_cache(struct page *page, char *incore)
>> 			/* skip initialized uptodate buddy */
>> 			continue;
>> 
>> +		if (!buffer_verified(bh[group - first_group]) ||
>> +		    !buffer_uptodate(bh[group - first_group]))
>> +			/* Skip faulty bitmaps */
>> +			continue;
>> +		else
>> +			err = 0;
>> +
> 
> Hm, ext4_wait_block_bitmap() can fail 3 ways (buffer not update, or
> not verified, but also if ext4_get_group_desc fails), but this only
> checks 2 of those, right?
> 
> If ext4_get_group_desc fails, we'll have a bh which is new, may or
> may not be uptodate, and ... I guess it won't be verified in that
> case either, will it.  So that's probably ok.
> 
> In fact, is the (!verified) test enough, here? (IOWs, could it
> possibly be verified, but not uptodate?) 
> 
> 
> In the bigger picture - if the filesystem is corrupt (or the device
> is bad) in this way, do we really *want* to continue?  What are the
> consequences of doing so, and have you tested with a filesystem
> partially-initialized like this?
> 
> Thinking more about it ... (sorry for the stream of consciousness
> here) - if validation fails, encountering this sort of error will
> trigger remount,ro or panic unless errors=continue, right?  But I
> guess that's still the default, so maybe we do expect to continue.
> So I go back to the question of: have you tested with partial init
> like this, to be sure we don't fall off some cliff later?

There is also the check in ext4_validate_block_bitmap() that will
fail if the bitmap is bad in some way.  At least that is marking
the bitmap bad in the group descriptor so that it is skipped for
later allocations.  Probably the same needs to be done here.

It also makes sense to fail the mount outright if too many of the
bitmaps are bad, or only mount it read-only, since it will otherwise
be unusable.

Cheers, Andreas