After a fair number of xfstests runs, xfs/182 started to fail
regularly with a corrupted directory - a directory read verifier was
failing after recovery because it found a block with a XARM magic
number (remote attribute block) rather than a directory data block.

The first time I saw this repeated failure I did /something/ and the
problem went away, so I was never able to find the underlying
problem. Test xfs/182 failed again today, and I found the root
cause before I did /something else/ that made it go away.

Tracing indicated that the block in question was being correctly
logged, the log was being flushed by sync, but the buffer was not
being written back before the shutdown occurred. Tracing also
indicated that log recovery was also reading the block, but then
never writing it before log recovery invalidated the cache,
indicating that it was not modified by log recovery.

More detailed analysis of the corpse indicated that the filesystem
had a uuid of "a4131074-1872-4cac-9323-2229adbcb886" but the XARM
block had a uuid of "8f32f043-c3c9-e7f8-f947-4e7f989c05d3", which
indicated it was a block from an older filesystem. The reason that
log recovery didn't replay it was that the LSN in the XARM block was
larger than the LSN of the transaction being replayed, and so the
block was not overwritten by log recovery.

Hence, log recovery cant blindly trust the magic number and LSN in
the block - it must verify that it belongs to the filesystem being
recovered before using the LSN. i.e. if the UUIDs don't match, we
need to unconditionally recovery the change held in the log.

This patch was first tested on a block device that was repeatedly
causing xfs/182 to fail with the same failure on the same block with
the same directory read corruption signature (i.e. XARM block). It
did not fail, and hasn't failed since.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Additional code in the error handler of xlog_recover_inode_pass2()
results in the following error:

static checker warning: "fs/xfs/xfs_log_recover.c:2999
xlog_recover_inode_pass2()
	 info: ignoring unreachable code."
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NMark Tinguely <tinguely@sgi.com>
Reviewed-by: Ben Myers <bpm@sgi.com
Signed-off-by: NBen Myers <bpm@sgi.com>

This is the recovery side of the btree block owner change operation
performed by swapext on CRC enabled filesystems. We detect that an
owner change is needed by the flag that has been placed on the inode
log format flag field. Because the inode recovery is being replayed
after the buffers that make up the BMBT in the given checkpoint, we
can walk all the buffers and directly modify them when we see the
flag set on an inode.

Because the inode can be relogged and hence present in multiple
chekpoints with the "change owner" flag set, we could do multiple
passes across the inode to do this change. While this isn't optimal,
we can't directly ignore the flag as there may be multiple
independent swap extent operations being replayed on the same inode
in different checkpoints so we can't ignore them.

Further, because the owner change operation uses ordered buffers, we
might have buffers that are newer on disk than the current
checkpoint and so already have the owner changed in them. Hence we
cannot just peek at a buffer in the tree and check that it has the
correct owner and assume that the change was completed.

So, for the moment just brute force the owner change every time we
see an inode with the flag set. Note that we have to be careful here
because the owner of the buffers may point to either the old owner
or the new owner. Currently the verifier can't verify the owner
directly, so there is no failure case here right now. If we verify
the owner exactly in future, then we'll have to take this into
account.

This was tested in terms of normal operation via xfstests - all of
the fsr tests now pass without failure. however, we really need to
modify xfs/227 to stress v3 inodes correctly to ensure we fully
cover this case for v5 filesystems.

In terms of recovery testing, I used a hacked version of xfs_fsr
that held the temp inode open for a few seconds before exiting so
that the filesystem could be shut down with an open owner change
recovery flags set on at least the temp inode. fsr leaves the temp
inode unlinked and in btree format, so this was necessary for the
owner change to be reliably replayed.

logprint confirmed the tmp inode in the log had the correct flag set:

INO: cnt:3 total:3 a:0x69e9e0 len:56 a:0x69ea20 len:176 a:0x69eae0 len:88
        INODE: #regs:3   ino:0x44  flags:0x209   dsize:88
	                                 ^^^^^

0x200 is set, indicating a data fork owner change needed to be
replayed on inode 0x44.  A printk in the revoery code confirmed that
the inode change was recovered:

XFS (vdc): Mounting Filesystem
XFS (vdc): Starting recovery (logdev: internal)
recovering owner change ino 0x44
XFS (vdc): Version 5 superblock detected. This kernel L support enabled!
Use of these features in this kernel is at your own risk!
XFS (vdc): Ending recovery (logdev: internal)

The script used to test this was:

$ cat ./recovery-fsr.sh
#!/bin/bash

dev=/dev/vdc
mntpt=/mnt/scratch
testfile=$mntpt/testfile

umount $mntpt
mkfs.xfs -f -m crc=1 $dev
mount $dev $mntpt
chmod 777 $mntpt

for i in `seq 10000 -1 0`; do
        xfs_io -f -d -c "pwrite $(($i * 4096)) 4096" $testfile > /dev/null 2>&1
done
xfs_bmap -vp $testfile |head -20

xfs_fsr -d -v $testfile &
sleep 10
/home/dave/src/xfstests-dev/src/godown -f $mntpt
wait
umount $mntpt

xfs_logprint -t $dev |tail -20
time mount $dev $mntpt
xfs_bmap -vp $testfile
umount $mntpt
$
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

sparse reports:

fs/xfs/xfs_log_recover.c:2017:24: sparse: cast to restricted __be64

Because I used the wrong structure for the on-disk superblock cast
in 50d5c8d8 ("xfs: check LSN ordering for v5 superblocks during
recovery"). Fix it.

Reported-by: kbuild test robot
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>

CRC enabled filesystems fail log recovery with 100% reliability on
xfstests xfs/085 with the following failure:

XFS (vdb): Mounting Filesystem
XFS (vdb): Starting recovery (logdev: internal)
XFS (vdb): Corruption detected. Unmount and run xfs_repair
XFS (vdb): bad inode magic/vsn daddr 144 #0 (magic=0)
XFS: Assertion failed: 0, file: fs/xfs/xfs_inode_buf.c, line: 95

The problem is that the inode buffer has not been recovered before
the readahead on the inode buffer is issued. The checkpoint being
recovered actually allocates the inode chunk we are doing readahead
from, so what comes from disk during readahead is essentially
random and the verifier barfs on it.

This inode buffer readahead problem affects non-crc filesystems,
too, but xfstests does not trigger it at all on such
configurations....
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Log recovery has some strict ordering requirements which unordered
or reordered metadata writeback can defeat. This can occur when an
item is logged in a transaction, written back to disk, and then
logged in a new transaction before the tail of the log is moved past
the original modification.

The result of this is that when we read an object off disk for
recovery purposes, the buffer that we read may not contain the
object type that recovery is expecting and hence at the end of the
checkpoint being recovered we have an invalid object in memory.

This isn't usually a problem, as recovery will then replay all the
other checkpoints and that brings the object back to a valid and
correct state, but the issue is that while the object is in the
invalid state it can be flushed to disk. This results in the object
verifier failing and triggering a corruption shutdown of log
recover. This is correct behaviour for the verifiers - the problem
is that we are not detecting that the object we've read off disk is
newer than the transaction we are replaying.

All metadata in v5 filesystems has the LSN of it's last modification
stamped in it. This enabled log recover to read that field and
determine the age of the object on disk correctly. If the LSN of the
object on disk is older than the transaction being replayed, then we
replay the modification. If the LSN of the object matches or is more
recent than the transaction's LSN, then we should avoid overwriting
the object as that is what leads to the transient corrupt state.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

xfstests xfs/087 fails 100% reliably with this assert:

XFS (vdb): Mounting Filesystem
XFS (vdb): Starting recovery (logdev: internal)
XFS: Assertion failed: bp->b_flags & XBF_STALE, file: fs/xfs/xfs_buf.c, line: 548

while trying to read a dquot buffer in xlog_recover_dquot_ra_pass2().

The issue is that the buffer length to read that is passed to
xfs_buf_readahead is in units of filesystem blocks, not disk blocks.
(i.e. FSB, not daddr). Fix it but putting the correct conversion in
place.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

When doing readhaead in log recovery, we check to see if buffers are
cancelled before doing readahead. If we find a cancelled buffer,
however, we always decrement the reference count we have on it, and
that means that readahead is causing a double decrement of the
cancelled buffer reference count.

This results in log recovery *replaying cancelled buffers* as the
actual recovery pass does not find the cancelled buffer entry in the
commit phase of the second pass across a transaction. On debug
kernels, this results in an ASSERT failure like so:

XFS: Assertion failed: !(flags & XFS_BLF_CANCEL), file: fs/xfs/xfs_log_recover.c, line: 1815

xfstests generic/311 reproduces this ASSERT failure with 100%
reproducability.

Fix it by making readahead only peek at the buffer cancelled state
rather than the full accounting that xlog_check_buffer_cancelled()
does.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

  It can take a long time to run log recovery operation because it is
single threaded and is bound by read latency. We can find that it took
most of the time to wait for the read IO to occur, so if one object
readahead is introduced to log recovery, it will obviously reduce the
log recovery time.

Log recovery time stat:

          w/o this patch        w/ this patch

real:        0m15.023s             0m7.802s
user:        0m0.001s              0m0.001s
sys:         0m0.246s              0m0.107s
Signed-off-by: NZhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NZhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NZhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NZhi Yong Wu <wuzhy@linux.vnet.ibm.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

xlog_find_tail() currently leaks a bp on one error path.

There is no error target, so manually free the bp before
returning the error.

Found by Coverity.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

xlog_find_zeroed() currently leaks a bp on one error path.

Using the bp_err: target resolves this.

Found by Coverity.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

With the new xfs_trans_res structure has been introduced, the log
reservation size, log count as well as log flags are pre-initialized
at mount time.  So it's time to refine xfs_trans_reserve() interface
to be more neat.

Also, introduce a new helper M_RES() to return a pointer to the
mp->m_resv structure to simplify the input.
Signed-off-by: NJie Liu <jeff.liu@oracle.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

There are a few small helper functions in xfs_util, all related to
xfs_inode modifications. Move them all to xfs_inode.c so all
xfs_inode operations are consiolidated in the one place.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Many of the definitions within xfs_dir2_priv.h are needed in
userspace outside libxfs. Definitions within xfs_dir2_priv.h are
wholly contained within libxfs, so we need to shuffle some of the
definitions around to keep consistency across files shared between
user and kernel space.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The on disk format definitions of the on-disk dquot, log formats and
quota off log formats are all intertwined with other definitions for
quotas. Separate them out into their own header file so they can
easily be shared with userspace.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The on-disk format definitions for the log are spread randoms
through a couple of header files. Consolidate it all in a single
file that can be shared easily with userspace. This means that
xfs_log.h and xfs_log_priv.h no longer need to be shared with
userspace.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

When we made all inode updates transactional, we no longer needed
the log recovery detection for inodes being newer on disk than the
transaction being replayed - it was redundant as replay of the log
would always result in the latest version of the inode would be on
disk. It was redundant, but left in place because it wasn't
considered to be a problem.

However, with the new "don't read inodes on create" optimisation,
flushiter has come back to bite us. Essentially, the optimisation
made always initialises flushiter to zero in the create transaction,
and so if we then crash and run recovery and the inode already on
disk has a non-zero flushiter it will skip recovery of that inode.
As a result, log recovery does the wrong thing and we end up with a
corrupt filesystem.

Because we have to support old kernel to new kernel upgrades, we
can't just get rid of the flushiter support in log recovery as we
might be upgrading from a kernel that doesn't have fully transactional
inode updates.  Unfortunately, for v4 superblocks there is no way to
guarantee that log recovery knows about this fact.

We cannot add a new inode format flag to say it's a "special inode
create" because it won't be understood by older kernels and so
recovery could do the wrong thing on downgrade. We cannot specially
detect the combination of zero mode/non-zero flushiter on disk to
non-zero mode, zero flushiter in the log item during recovery
because wrapping of the flushiter can result in false detection.

Hence that makes this "don't use flushiter" optimisation limited to
a disk format that guarantees that we don't need it. And that means
the only fix here is to limit the "no read IO on create"
optimisation to version 5 superblocks....
Reported-by: NMarkus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit e60896d8)

When we made all inode updates transactional, we no longer needed
the log recovery detection for inodes being newer on disk than the
transaction being replayed - it was redundant as replay of the log
would always result in the latest version of the inode would be on
disk. It was redundant, but left in place because it wasn't
considered to be a problem.

However, with the new "don't read inodes on create" optimisation,
flushiter has come back to bite us. Essentially, the optimisation
made always initialises flushiter to zero in the create transaction,
and so if we then crash and run recovery and the inode already on
disk has a non-zero flushiter it will skip recovery of that inode.
As a result, log recovery does the wrong thing and we end up with a
corrupt filesystem.

Because we have to support old kernel to new kernel upgrades, we
can't just get rid of the flushiter support in log recovery as we
might be upgrading from a kernel that doesn't have fully transactional
inode updates.  Unfortunately, for v4 superblocks there is no way to
guarantee that log recovery knows about this fact.

We cannot add a new inode format flag to say it's a "special inode
create" because it won't be understood by older kernels and so
recovery could do the wrong thing on downgrade. We cannot specially
detect the combination of zero mode/non-zero flushiter on disk to
non-zero mode, zero flushiter in the log item during recovery
because wrapping of the flushiter can result in false detection.

Hence that makes this "don't use flushiter" optimisation limited to
a disk format that guarantees that we don't need it. And that means
the only fix here is to limit the "no read IO on create"
optimisation to version 5 superblocks....
Reported-by: NMarkus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

When we find a icreate transaction, we need to get and initialise
the buffers in the range that has been passed. Extract and verify
the information in the item record, then loop over the range
initialising and issuing the buffer writes delayed.

Support an arbitrary size range to initialise so that in
future when we allocate inodes in much larger chunks all kernels
that understand this transaction can still recover them.
Signed-off-by: NDave Chinner <david@fromorbit.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Unfortunately, we cannot guarantee that items logged multiple times
and replayed by log recovery do not take objects back in time. When
they are taken back in time, the go into an intermediate state which
is corrupt, and hence verification that occurs on this intermediate
state causes log recovery to abort with a corruption shutdown.

Instead of causing a shutdown and unmountable filesystem, don't
verify post-recovery items before they are written to disk. This is
less than optimal, but there is no way to detect this issue for
non-CRC filesystems If log recovery successfully completes, this
will be undone and the object will be consistent by subsequent
transactions that are replayed, so in most cases we don't need to
take drastic action.

For CRC enabled filesystems, leave the verifiers in place - we need
to call them to recalculate the CRCs on the objects anyway. This
recovery problem can be solved for such filesystems - we have a LSN
stamped in all metadata at writeback time that we can to determine
whether the item should be replayed or not. This is a separate piece
of work, so is not addressed by this patch.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit 9222a9cf)

Unfortunately, we cannot guarantee that items logged multiple times
and replayed by log recovery do not take objects back in time. When
they are taken back in time, the go into an intermediate state which
is corrupt, and hence verification that occurs on this intermediate
state causes log recovery to abort with a corruption shutdown.

Instead of causing a shutdown and unmountable filesystem, don't
verify post-recovery items before they are written to disk. This is
less than optimal, but there is no way to detect this issue for
non-CRC filesystems If log recovery successfully completes, this
will be undone and the object will be consistent by subsequent
transactions that are replayed, so in most cases we don't need to
take drastic action.

For CRC enabled filesystems, leave the verifiers in place - we need
to call them to recalculate the CRCs on the objects anyway. This
recovery problem can be solved for such filesystems - we have a LSN
stamped in all metadata at writeback time that we can to determine
whether the item should be replayed or not. This is a separate piece
of work, so is not addressed by this patch.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The inode unlinked list manipulations operate directly on the inode
buffer, and so bypass the inode CRC calculation mechanisms. Hence an
inode on the unlinked list has an invalid CRC. Fix this by
recalculating the CRC whenever we modify an unlinked list pointer in
an inode, ncluding during log recovery. This is trivial to do and
results in  unlinked list operations always leaving a consistent
inode in the buffer.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit 0a32c26e)

There are several constraints that inode allocation and unlink
logging impose on log recovery. These all stem from the fact that
inode alloc/unlink are logged in buffers, but all other inode
changes are logged in inode items. Hence there are ordering
constraints that recovery must follow to ensure the correct result
occurs.

As it turns out, this ordering has been working mostly by chance
than good management. The existing code moves all buffers except
cancelled buffers to the head of the list, and everything else to
the tail of the list. The problem with this is that is interleaves
inode items with the buffer cancellation items, and hence whether
the inode item in an cancelled buffer gets replayed is essentially
left to chance.

Further, this ordering causes problems for log recovery when inode
CRCs are enabled. It typically replays the inode unlink buffer long before
it replays the inode core changes, and so the CRC recorded in an
unlink buffer is going to be invalid and hence any attempt to
validate the inode in the buffer is going to fail. Hence we really
need to enforce the ordering that the inode alloc/unlink code has
expected log recovery to have since inode chunk de-allocation was
introduced back in 2003...
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit a775ad77)

Calculating dquot CRCs when the backing buffer is written back just
doesn't work reliably. There are several places which manipulate
dquots directly in the buffers, and they don't calculate CRCs
appropriately, nor do they always set the buffer up to calculate
CRCs appropriately.

Firstly, if we log a dquot buffer (e.g. during allocation) it gets
logged without valid CRC, and so on recovery we end up with a dquot
that is not valid.

Secondly, if we recover/repair a dquot, we don't have a verifier
attached to the buffer and hence CRCs are not calculated on the way
down to disk.

Thirdly, calculating the CRC after we've changed the contents means
that if we re-read the dquot from the buffer, we cannot verify the
contents of the dquot are valid, as the CRC is invalid.

So, to avoid all the dquot CRC errors that are being detected by the
read verifier, change to using the same model as for inodes. That
is, dquot CRCs are calculated and written to the backing buffer at
the time the dquot is flushed to the backing buffer. If we modify
the dquot directly in the backing buffer, calculate the CRC
immediately after the modification is complete. Hence the dquot in
the on-disk buffer should always have a valid CRC.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit 6fcdc59d)

The inode unlinked list manipulations operate directly on the inode
buffer, and so bypass the inode CRC calculation mechanisms. Hence an
inode on the unlinked list has an invalid CRC. Fix this by
recalculating the CRC whenever we modify an unlinked list pointer in
an inode, ncluding during log recovery. This is trivial to do and
results in  unlinked list operations always leaving a consistent
inode in the buffer.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

There are several constraints that inode allocation and unlink
logging impose on log recovery. These all stem from the fact that
inode alloc/unlink are logged in buffers, but all other inode
changes are logged in inode items. Hence there are ordering
constraints that recovery must follow to ensure the correct result
occurs.

As it turns out, this ordering has been working mostly by chance
than good management. The existing code moves all buffers except
cancelled buffers to the head of the list, and everything else to
the tail of the list. The problem with this is that is interleaves
inode items with the buffer cancellation items, and hence whether
the inode item in an cancelled buffer gets replayed is essentially
left to chance.

Further, this ordering causes problems for log recovery when inode
CRCs are enabled. It typically replays the inode unlink buffer long before
it replays the inode core changes, and so the CRC recorded in an
unlink buffer is going to be invalid and hence any attempt to
validate the inode in the buffer is going to fail. Hence we really
need to enforce the ordering that the inode alloc/unlink code has
expected log recovery to have since inode chunk de-allocation was
introduced back in 2003...
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Calculating dquot CRCs when the backing buffer is written back just
doesn't work reliably. There are several places which manipulate
dquots directly in the buffers, and they don't calculate CRCs
appropriately, nor do they always set the buffer up to calculate
CRCs appropriately.

Firstly, if we log a dquot buffer (e.g. during allocation) it gets
logged without valid CRC, and so on recovery we end up with a dquot
that is not valid.

Secondly, if we recover/repair a dquot, we don't have a verifier
attached to the buffer and hence CRCs are not calculated on the way
down to disk.

Thirdly, calculating the CRC after we've changed the contents means
that if we re-read the dquot from the buffer, we cannot verify the
contents of the dquot are valid, as the CRC is invalid.

So, to avoid all the dquot CRC errors that are being detected by the
read verifier, change to using the same model as for inodes. That
is, dquot CRCs are calculated and written to the backing buffer at
the time the dquot is flushed to the backing buffer. If we modify
the dquot directly in the backing buffer, calculate the CRC
immediately after the modification is complete. Hence the dquot in
the on-disk buffer should always have a valid CRC.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

A long time ago in a galaxy far away....

.. the was a commit made to fix some ilinux specific "fragmented
buffer" log recovery problem:

http://oss.sgi.com/cgi-bin/gitweb.cgi?p=archive/xfs-import.git;a=commitdiff;h=b29c0bece51da72fb3ff3b61391a391ea54e1603

That problem occurred when a contiguous dirty region of a buffer was
split across across two pages of an unmapped buffer. It's been a
long time since that has been done in XFS, and the changes to log
the entire inode buffers for CRC enabled filesystems has
re-introduced that corner case.

And, of course, it turns out that the above commit didn't actually
fix anything - it just ensured that log recovery is guaranteed to
fail when this situation occurs. And now for the gory details.

xfstest xfs/085 is failing with this assert:

XFS (vdb): bad number of regions (0) in inode log format
XFS: Assertion failed: 0, file: fs/xfs/xfs_log_recover.c, line: 1583

Largely undocumented factoid #1: Log recovery depends on all log
buffer format items starting with this format:

struct foo_log_format {
	__uint16_t	type;
	__uint16_t	size;
	....

As recoery uses the size field and assumptions about 32 bit
alignment in decoding format items.  So don't pay much attention to
the fact log recovery thinks that it decoding an inode log format
item - it just uses them to determine what the size of the item is.

But why would it see a log format item with a zero size? Well,
luckily enough xfs_logprint uses the same code and gives the same
error, so with a bit of gdb magic, it turns out that it isn't a log
format that is being decoded. What logprint tells us is this:

Oper (130): tid: a0375e1a  len: 28  clientid: TRANS  flags: none
BUF:  #regs: 2   start blkno: 144 (0x90)  len: 16  bmap size: 2  flags: 0x4000
Oper (131): tid: a0375e1a  len: 4096  clientid: TRANS  flags: none
BUF DATA
----------------------------------------------------------------------------
Oper (132): tid: a0375e1a  len: 4096  clientid: TRANS  flags: none
xfs_logprint: unknown log operation type (4e49)
**********************************************************************
* ERROR: data block=2                                                 *
**********************************************************************

That we've got a buffer format item (oper 130) that has two regions;
the format item itself and one dirty region. The subsequent region
after the buffer format item and it's data is them what we are
tripping over, and the first bytes of it at an inode magic number.
Not a log opheader like there is supposed to be.

That means there's a problem with the buffer format item. It's dirty
data region is 4096 bytes, and it contains - you guessed it -
initialised inodes. But inode buffers are 8k, not 4k, and we log
them in their entirety. So something is wrong here. The buffer
format item contains:

(gdb) p /x *(struct xfs_buf_log_format *)in_f
$22 = {blf_type = 0x123c, blf_size = 0x2, blf_flags = 0x4000,
       blf_len = 0x10, blf_blkno = 0x90, blf_map_size = 0x2,
       blf_data_map = {0xffffffff, 0xffffffff, .... }}

Two regions, and a signle dirty contiguous region of 64 bits.  64 *
128 = 8k, so this should be followed by a single 8k region of data.
And the blf_flags tell us that the type of buffer is a
XFS_BLFT_DINO_BUF. It contains inodes. And because it doesn't have
the XFS_BLF_INODE_BUF flag set, that means it's an inode allocation
buffer. So, it should be followed by 8k of inode data.

But we know that the next region has a header of:

(gdb) p /x *ohead
$25 = {oh_tid = 0x1a5e37a0, oh_len = 0x100000, oh_clientid = 0x69,
       oh_flags = 0x0, oh_res2 = 0x0}

and so be32_to_cpu(oh_len) = 0x1000 = 4096 bytes. It's simply not
long enough to hold all the logged data. There must be another
region. There is - there's a following opheader for another 4k of
data that contains the other half of the inode cluster data - the
one we assert fail on because it's not a log format header.

So why is the second part of the data not being accounted to the
correct buffer log format structure? It took a little more work with
gdb to work out that the buffer log format structure was both
expecting it to be there but hadn't accounted for it. It was at that
point I went to the kernel code, as clearly this wasn't a bug in
xfs_logprint and the kernel was writing bad stuff to the log.

First port of call was the buffer item formatting code, and the
discontiguous memory/contiguous dirty region handling code
immediately stood out. I've wondered for a long time why the code
had this comment in it:

                        vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
                        vecp->i_len = nbits * XFS_BLF_CHUNK;
                        vecp->i_type = XLOG_REG_TYPE_BCHUNK;
/*
 * You would think we need to bump the nvecs here too, but we do not
 * this number is used by recovery, and it gets confused by the boundary
 * split here
 *                      nvecs++;
 */
                        vecp++;

And it didn't account for the extra vector pointer. The case being
handled here is that a contiguous dirty region lies across a
boundary that cannot be memcpy()d across, and so has to be split
into two separate operations for xlog_write() to perform.

What this code assumes is that what is written to the log is two
consecutive blocks of data that are accounted in the buf log format
item as the same contiguous dirty region and so will get decoded as
such by the log recovery code.

The thing is, xlog_write() knows nothing about this, and so just
does it's normal thing of adding an opheader for each vector. That
means the 8k region gets written to the log as two separate regions
of 4k each, but because nvecs has not been incremented, the buf log
format item accounts for only one of them.

Hence when we come to log recovery, we process the first 4k region
and then expect to come across a new item that starts with a log
format structure of some kind that tells us whenteh next data is
going to be. Instead, we hit raw buffer data and things go bad real
quick.

So, the commit from 2002 that commented out nvecs++ is just plain
wrong. It breaks log recovery completely, and it would seem the only
reason this hasn't been since then is that we don't log large
contigous regions of multi-page unmapped buffers very often. Never
would be a closer estimate, at least until the CRC code came along....

So, lets fix that by restoring the nvecs accounting for the extra
region when we hit this case.....

.... and there's the problemin log recovery it is apparently working
around:

XFS: Assertion failed: i == item->ri_total, file: fs/xfs/xfs_log_recover.c, line: 2135

Yup, xlog_recover_do_reg_buffer() doesn't handle contigous dirty
regions being broken up into multiple regions by the log formatting
code. That's an easy fix, though - if the number of contiguous dirty
bits exceeds the length of the region being copied out of the log,
only account for the number of dirty bits that region covers, and
then loop again and copy more from the next region. It's a 2 line
fix.

Now xfstests xfs/085 passes, we have one less piece of mystery
code, and one more important piece of knowledge about how to
structure new log format items..
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

(cherry picked from commit 709da6a6)

A long time ago in a galaxy far away....

.. the was a commit made to fix some ilinux specific "fragmented
buffer" log recovery problem:

http://oss.sgi.com/cgi-bin/gitweb.cgi?p=archive/xfs-import.git;a=commitdiff;h=b29c0bece51da72fb3ff3b61391a391ea54e1603

That problem occurred when a contiguous dirty region of a buffer was
split across across two pages of an unmapped buffer. It's been a
long time since that has been done in XFS, and the changes to log
the entire inode buffers for CRC enabled filesystems has
re-introduced that corner case.

And, of course, it turns out that the above commit didn't actually
fix anything - it just ensured that log recovery is guaranteed to
fail when this situation occurs. And now for the gory details.

xfstest xfs/085 is failing with this assert:

XFS (vdb): bad number of regions (0) in inode log format
XFS: Assertion failed: 0, file: fs/xfs/xfs_log_recover.c, line: 1583

Largely undocumented factoid #1: Log recovery depends on all log
buffer format items starting with this format:

struct foo_log_format {
	__uint16_t	type;
	__uint16_t	size;
	....

As recoery uses the size field and assumptions about 32 bit
alignment in decoding format items.  So don't pay much attention to
the fact log recovery thinks that it decoding an inode log format
item - it just uses them to determine what the size of the item is.

But why would it see a log format item with a zero size? Well,
luckily enough xfs_logprint uses the same code and gives the same
error, so with a bit of gdb magic, it turns out that it isn't a log
format that is being decoded. What logprint tells us is this:

Oper (130): tid: a0375e1a  len: 28  clientid: TRANS  flags: none
BUF:  #regs: 2   start blkno: 144 (0x90)  len: 16  bmap size: 2  flags: 0x4000
Oper (131): tid: a0375e1a  len: 4096  clientid: TRANS  flags: none
BUF DATA
----------------------------------------------------------------------------
Oper (132): tid: a0375e1a  len: 4096  clientid: TRANS  flags: none
xfs_logprint: unknown log operation type (4e49)
**********************************************************************
* ERROR: data block=2                                                 *
**********************************************************************

That we've got a buffer format item (oper 130) that has two regions;
the format item itself and one dirty region. The subsequent region
after the buffer format item and it's data is them what we are
tripping over, and the first bytes of it at an inode magic number.
Not a log opheader like there is supposed to be.

That means there's a problem with the buffer format item. It's dirty
data region is 4096 bytes, and it contains - you guessed it -
initialised inodes. But inode buffers are 8k, not 4k, and we log
them in their entirety. So something is wrong here. The buffer
format item contains:

(gdb) p /x *(struct xfs_buf_log_format *)in_f
$22 = {blf_type = 0x123c, blf_size = 0x2, blf_flags = 0x4000,
       blf_len = 0x10, blf_blkno = 0x90, blf_map_size = 0x2,
       blf_data_map = {0xffffffff, 0xffffffff, .... }}

Two regions, and a signle dirty contiguous region of 64 bits.  64 *
128 = 8k, so this should be followed by a single 8k region of data.
And the blf_flags tell us that the type of buffer is a
XFS_BLFT_DINO_BUF. It contains inodes. And because it doesn't have
the XFS_BLF_INODE_BUF flag set, that means it's an inode allocation
buffer. So, it should be followed by 8k of inode data.

But we know that the next region has a header of:

(gdb) p /x *ohead
$25 = {oh_tid = 0x1a5e37a0, oh_len = 0x100000, oh_clientid = 0x69,
       oh_flags = 0x0, oh_res2 = 0x0}

and so be32_to_cpu(oh_len) = 0x1000 = 4096 bytes. It's simply not
long enough to hold all the logged data. There must be another
region. There is - there's a following opheader for another 4k of
data that contains the other half of the inode cluster data - the
one we assert fail on because it's not a log format header.

So why is the second part of the data not being accounted to the
correct buffer log format structure? It took a little more work with
gdb to work out that the buffer log format structure was both
expecting it to be there but hadn't accounted for it. It was at that
point I went to the kernel code, as clearly this wasn't a bug in
xfs_logprint and the kernel was writing bad stuff to the log.

First port of call was the buffer item formatting code, and the
discontiguous memory/contiguous dirty region handling code
immediately stood out. I've wondered for a long time why the code
had this comment in it:

                        vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
                        vecp->i_len = nbits * XFS_BLF_CHUNK;
                        vecp->i_type = XLOG_REG_TYPE_BCHUNK;
/*
 * You would think we need to bump the nvecs here too, but we do not
 * this number is used by recovery, and it gets confused by the boundary
 * split here
 *                      nvecs++;
 */
                        vecp++;

And it didn't account for the extra vector pointer. The case being
handled here is that a contiguous dirty region lies across a
boundary that cannot be memcpy()d across, and so has to be split
into two separate operations for xlog_write() to perform.

What this code assumes is that what is written to the log is two
consecutive blocks of data that are accounted in the buf log format
item as the same contiguous dirty region and so will get decoded as
such by the log recovery code.

The thing is, xlog_write() knows nothing about this, and so just
does it's normal thing of adding an opheader for each vector. That
means the 8k region gets written to the log as two separate regions
of 4k each, but because nvecs has not been incremented, the buf log
format item accounts for only one of them.

Hence when we come to log recovery, we process the first 4k region
and then expect to come across a new item that starts with a log
format structure of some kind that tells us whenteh next data is
going to be. Instead, we hit raw buffer data and things go bad real
quick.

So, the commit from 2002 that commented out nvecs++ is just plain
wrong. It breaks log recovery completely, and it would seem the only
reason this hasn't been since then is that we don't log large
contigous regions of multi-page unmapped buffers very often. Never
would be a closer estimate, at least until the CRC code came along....

So, lets fix that by restoring the nvecs accounting for the extra
region when we hit this case.....

.... and there's the problemin log recovery it is apparently working
around:

XFS: Assertion failed: i == item->ri_total, file: fs/xfs/xfs_log_recover.c, line: 2135

Yup, xlog_recover_do_reg_buffer() doesn't handle contigous dirty
regions being broken up into multiple regions by the log formatting
code. That's an easy fix, though - if the number of contiguous dirty
bits exceeds the length of the region being copied out of the log,
only account for the number of dirty bits that region covers, and
then loop again and copy more from the next region. It's a 2 line
fix.

Now xfstests xfs/085 passes, we have one less piece of mystery
code, and one more important piece of knowledge about how to
structure new log format items..
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NMark Tinguely <tinguely@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Fix a build error when CONFIG_XFS_QUOTA=n:

fs/built-in.o: In function `xlog_recovery_validate_buf_type':
/home/dave/src/build/x86-64/xfsdev/fs/xfs/xfs_log_recover.c:1948: undefined
reference to `xfs_dquot_buf_ops'
Reported-by: NMichael L. Semon <mlsemon35@gmail.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The version 5 superblock has extended feature masks for compatible,
incompatible and read-only compatible feature sets. Implement the
masking and mount-time checking for these feature masks.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

With the addition of CRCs, there is such a wide and varied change to
the on disk format that it makes sense to bump the superblock
version number rather than try to use feature bits for all the new
functionality.

This commit introduces all the new superblock fields needed for all
the new functionality: feature masks similar to ext4, separate
project quota inodes, a LSN field for recovery and the CRC field.

This commit does not bump the superblock version number, however.
That will be done as a separate commit at the end of the series
after all the new functionality is present so we switch it all on in
one commit. This means that we can slowly introduce the changes
without them being active and hence maintain bisectability of the
tree.

This patch is based on a patch originally written by myself back
from SGI days, which was subsequently modified by Christoph Hellwig.
There is relatively little of that patch remaining, but the history
of the patch still should be acknowledged here.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

The buffer type passed to log recvoery in the buffer log item
overruns the blf_flags field. I had assumed that flags field was a
32 bit value, and it turns out it is a unisgned short. Therefore
having 19 flags doesn't really work.

Convert the buffer type field to numeric value, and use the top 5
bits of the flags field for it. We currently have 17 types of
buffers, so using 5 bits gives us plenty of room for expansion in
future....
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Add buffer types to the buffer log items so that log recovery can
validate the buffers and calculate CRCs correctly after the buffers
are recovered.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Add a header to the remote symlink block, containing location and
owner information, as well as CRCs and LSN fields. This requires
verifiers to be added to the remote symlink buffers for CRC enabled
filesystems.

This also fixes a bug reading multiple block symlinks, where the second
block overwrites the first block when copying out the link name.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>

Add a new inode version with a larger core.  The primary objective is
to allow for a crc of the inode, and location information (uuid and ino)
to verify it was written in the right place.  We also extend it by:

	a creation time (for Samba);
	a changecount (for NFSv4);
	a flush sequence (in LSN format for recovery);
	an additional inode flags field; and
	some additional padding.

These additional fields are not implemented yet, but already laid
out in the structure.

[dchinner@redhat.com] Added LSN and flags field, some factoring and rework to
capture all the necessary information in the crc calculation.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBen Myers <bpm@sgi.com>
Signed-off-by: NBen Myers <bpm@sgi.com>