]> www.infradead.org Git - users/hch/xfs.git/log
users/hch/xfs.git
7 months agoxfs: convert remaining trace points to pass pag structures
Christoph Hellwig [Sat, 31 Aug 2024 08:10:49 +0000 (11:10 +0300)]
xfs: convert remaining trace points to pass pag structures

Convert all tracepoints that take [mp,agno] tuples to take a pag argument
instead so that decoding only happens when tracepoints are enabled and to
clean up the callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass the pag to the xrep_newbt_extent_class tracepoints
Christoph Hellwig [Sun, 1 Sep 2024 05:39:19 +0000 (08:39 +0300)]
xfs: pass the pag to the xrep_newbt_extent_class tracepoints

This requires moving a few of the callsites a little bit to ensure that
we already have the reference, but allows for the decoding to only happen
when tracing is actually enabled, and cleans up the callsites a bit.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass the pag to the trace_xrep_calc_ag_resblks{,_btsize} trace points
Christoph Hellwig [Sun, 1 Sep 2024 05:34:33 +0000 (08:34 +0300)]
xfs: pass the pag to the trace_xrep_calc_ag_resblks{,_btsize} trace points

This requires holding the pag refcount a little longer, but allows for the
decoding to only happen when tracing is actually enabled, and cleans up the
callsites a bit.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass objects to the xrep_ibt_walk_rmap tracepoint
Christoph Hellwig [Sun, 1 Sep 2024 05:26:13 +0000 (08:26 +0300)]
xfs: pass objects to the xrep_ibt_walk_rmap tracepoint

Pass the perag structure and the irec so that the decoding is only done
when tracing is actually enabled and the call sites look a lot neater,
and remove the pointless class indirection.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass the iunlink item to the xfs_iunlink_update_dinode trace point
Christoph Hellwig [Sat, 31 Aug 2024 08:20:24 +0000 (11:20 +0300)]
xfs: pass the iunlink item to the xfs_iunlink_update_dinode trace point

So that decoding is only done when tracing is actually enabled and the
call site look a lot neater.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass objects to the xfs_irec_merge_{pre,post} trace points
Christoph Hellwig [Sat, 31 Aug 2024 07:39:59 +0000 (10:39 +0300)]
xfs: pass objects to the xfs_irec_merge_{pre,post} trace points

Pass the perag structure and the irec to these tracepoints so that the
decoding is only done when tracing is actually enabled and the call sites
look a lot neater.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass a perag structure to the xfs_ag_resv_init_error trace point
Christoph Hellwig [Sat, 31 Aug 2024 07:37:07 +0000 (10:37 +0300)]
xfs: pass a perag structure to the xfs_ag_resv_init_error trace point

And remove the single instance class indirection for it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: constify pag arguments to trace points
Christoph Hellwig [Thu, 19 Sep 2024 07:02:34 +0000 (09:02 +0200)]
xfs: constify pag arguments to trace points

Trace points never modify their arguments.  Mark all the pag objects
passed to trace points.  The exception is the xfs_ag_resv_class, which
uses the xfs_perag_resv helper that can't be marked const due to
other users modifying the returned structure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the unused xrep_bmap_walk_rmap trace point
Christoph Hellwig [Sat, 14 Sep 2024 06:51:29 +0000 (08:51 +0200)]
xfs: remove the unused xrep_bmap_walk_rmap trace point

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the unused trace_xfs_iwalk_ag trace point
Christoph Hellwig [Sat, 31 Aug 2024 08:22:05 +0000 (11:22 +0300)]
xfs: remove the unused trace_xfs_iwalk_ag trace point

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the mount field from struct xfs_busy_extents
Christoph Hellwig [Sat, 14 Sep 2024 06:31:53 +0000 (08:31 +0200)]
xfs: remove the mount field from struct xfs_busy_extents

The mount field is only passed to xfs_extent_busy_clear, which never uses
it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: keep a reference to the pag for busy extents
Christoph Hellwig [Sat, 31 Aug 2024 05:29:58 +0000 (08:29 +0300)]
xfs: keep a reference to the pag for busy extents

Processing of busy extents requires the perag structure, so keep the
reference while they are in flight.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass a pag to xfs_extent_busy_{search,reuse}
Christoph Hellwig [Sun, 1 Sep 2024 05:10:12 +0000 (08:10 +0300)]
xfs: pass a pag to xfs_extent_busy_{search,reuse}

Replace the [mp,agno] tuple with the perag structure, which will become
more useful later.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: add a xfs_agino_to_ino helper
Christoph Hellwig [Sat, 31 Aug 2024 07:55:11 +0000 (10:55 +0300)]
xfs: add a xfs_agino_to_ino helper

Add a helpers to convert an agino to an ino based on a pag structure.

This provides a simpler conversion and better type safety compared to the
existing code that passes the mount structure and the agno separately.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: add xfs_agbno_to_fsb and xfs_agbno_to_daddr helpers
Christoph Hellwig [Thu, 19 Sep 2024 06:25:30 +0000 (08:25 +0200)]
xfs: add xfs_agbno_to_fsb and xfs_agbno_to_daddr helpers

Add helpers to convert an agbno to a daddr or fsbno based on a pag
structure.

This provides a simpler conversion and better type safety compared to the
existing code that passes the mount structure and the agno separately.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the agno argument to xfs_free_ag_extent
Christoph Hellwig [Sat, 31 Aug 2024 08:18:02 +0000 (11:18 +0300)]
xfs: remove the agno argument to xfs_free_ag_extent

xfs_free_ag_extent already has a pointer to the pag structure through
the agf buffer.  Use that instead of passing the redundant argument,
and do the same for the tracepoint.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: pass a pag to xfs_difree_inode_chunk
Christoph Hellwig [Sat, 31 Aug 2024 18:05:55 +0000 (21:05 +0300)]
xfs: pass a pag to xfs_difree_inode_chunk

We'll want to use more than just the agno field in a bit.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the unused pag_active_wq field in struct xfs_perag
Christoph Hellwig [Sun, 1 Sep 2024 14:29:54 +0000 (17:29 +0300)]
xfs: remove the unused pag_active_wq field in struct xfs_perag

pag_active_wq is only woken, but never waited for.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove the unused pagb_count field in struct xfs_perag
Christoph Hellwig [Sun, 1 Sep 2024 14:29:37 +0000 (17:29 +0300)]
xfs: remove the unused pagb_count field in struct xfs_perag

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: fix superfluous clearing of info->low in __xfs_getfsmap_datadev
Christoph Hellwig [Sun, 15 Sep 2024 04:49:40 +0000 (06:49 +0200)]
xfs: fix superfluous clearing of info->low in __xfs_getfsmap_datadev

The for_each_perag helpers update the agno passed in for each iteration,
and thus the "if (pag->pag_agno == start_ag)" check will always be true.

Add another variable for the loop iterator so that the field is only
cleared after the first iteration.

Signed-off-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: don't use __GFP_RETRY_MAYFAIL in xfs_initialize_perag
Christoph Hellwig [Thu, 19 Sep 2024 13:15:10 +0000 (15:15 +0200)]
xfs: don't use __GFP_RETRY_MAYFAIL in xfs_initialize_perag

__GFP_RETRY_MAYFAIL increases the likelyhood of allocations to fail,
which isn't really helpful during log recovery.  Remove the flag and
stick to the default GFP_KERNEL policies.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: merge the perag freeing helpers
Christoph Hellwig [Sun, 1 Sep 2024 08:09:32 +0000 (11:09 +0300)]
xfs: merge the perag freeing helpers

There is no good reason to have two different routines for freeing perag
structures for the unmount and error cases.  Add two arguments to specify
the range of AGs to free to xfs_free_perag, and use that to replace
xfs_free_unused_perag_range.

The addition RCU grace period for the error case is harmless, and the
extra check for the AG to actually exist is not required now that the
callers pass the exact known allocated range.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: pass the exact range to initialize to xfs_initialize_perag
Christoph Hellwig [Sun, 8 Sep 2024 07:53:41 +0000 (10:53 +0300)]
xfs: pass the exact range to initialize to xfs_initialize_perag

Currently only the new agcount is passed to xfs_initialize_perag, which
requires lookups of existing AGs to skip them and complicates error
handling.  Also pass the previous agcount so that the range that
xfs_initialize_perag operates on is exactly defined.  That way the
extra lookups can be avoided, and error handling can clean up the
exact range from the old count to the last added perag structure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: ensure st_blocks never goes to zero during COW writes
Christoph Hellwig [Tue, 27 Aug 2024 05:03:21 +0000 (07:03 +0200)]
xfs: ensure st_blocks never goes to zero during COW writes

COW writes remove the amount overwritten either directly for delalloc
reservations, or in earlier deferred transactions than adding the new
amount back in the bmap map transaction.  This means st_blocks on an
inode where all data is overwritten using the COW path can temporarily
show a 0 st_blocks.  This can easily be reproduced with the pending
zoned device support where all writes use this path and trips the
check in generic/615, but could also happen on a reflink file without
that.

Fix this by temporarily add the pending blocks to be mapped to
i_delayed_blks while the item is queued.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: use xas_for_each_marked in xfs_reclaim_inodes_count
Christoph Hellwig [Thu, 29 Aug 2024 04:08:41 +0000 (07:08 +0300)]
xfs: use xas_for_each_marked in xfs_reclaim_inodes_count

xfs_reclaim_inodes_count iterates over all AGs to sum up the reclaimable
inodes counts.  There is no point in grabbing a reference to the them or
unlock the RCU critical section for each iteration, so switch to the
more efficient xas_for_each_marked iterator.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: convert perag lookup to xarray
Christoph Hellwig [Thu, 29 Aug 2024 04:08:40 +0000 (07:08 +0300)]
xfs: convert perag lookup to xarray

Convert the perag lookup from the legacy radix tree to the xarray,
which allows for much nicer iteration and bulk lookup semantics.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: simplify tagged perag iteration
Christoph Hellwig [Thu, 29 Aug 2024 04:08:39 +0000 (07:08 +0300)]
xfs: simplify tagged perag iteration

Pass the old perag structure to the tagged loop helpers so that they can
grab the old agno before releasing the reference.  This removes the need
to separately track the agno and the iterator macro, and thus also
obsoletes the for_each_perag_tag syntactic sugar.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: move the tagged perag lookup helpers to xfs_icache.c
Christoph Hellwig [Thu, 29 Aug 2024 04:08:38 +0000 (07:08 +0300)]
xfs: move the tagged perag lookup helpers to xfs_icache.c

The tagged perag helpers are only used in xfs_icache.c in the kernel code
and not at all in xfsprogs.  Move them to xfs_icache.c in preparation for
switching to an xarray, for which I have no plan to implement the tagged
lookup functions for userspace.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: use kfree_rcu_mightsleep to free the perag structures
Christoph Hellwig [Thu, 29 Aug 2024 04:08:37 +0000 (07:08 +0300)]
xfs: use kfree_rcu_mightsleep to free the perag structures

Using the kfree_rcu_mightsleep is simpler and removes the need for a
rcu_head in the perag structure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: use LIST_HEAD() to simplify code
Hongbo Li [Wed, 21 Aug 2024 06:43:55 +0000 (14:43 +0800)]
xfs: use LIST_HEAD() to simplify code

list_head can be initialized automatically with LIST_HEAD()
instead of calling INIT_LIST_HEAD().

Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: Remove duplicate xfs_trans_priv.h header
Jiapeng Chong [Wed, 10 Jul 2024 02:45:14 +0000 (10:45 +0800)]
xfs: Remove duplicate xfs_trans_priv.h header

./fs/xfs/libxfs/xfs_defer.c: xfs_trans_priv.h is included more than once.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Closes: https://bugzilla.openanolis.cn/show_bug.cgi?id=9491
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: remove unnecessary check
Dan Carpenter [Fri, 12 Jul 2024 14:07:36 +0000 (09:07 -0500)]
xfs: remove unnecessary check

We checked that "pip" is non-NULL at the start of the if else statement
so there is no need to check again here.  Delete the check.

Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: Use xfs set and clear mp state helpers
John Garry [Wed, 10 Jul 2024 10:31:19 +0000 (10:31 +0000)]
xfs: Use xfs set and clear mp state helpers

Use the set and clear mp state helpers instead of open-coding.

It is noted that in some instances calls to atomic operation set_bit() and
clear_bit() are being replaced with test_and_set_bit() and
test_and_clear_bit(), respectively, as there is no specific helpers for
set_bit() and clear_bit() only. However should be ok, as we are just
ignoring the returned value from those "test" variants.

Signed-off-by: John Garry <john.g.garry@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: reclaim speculative preallocations for append only files
Christoph Hellwig [Tue, 13 Aug 2024 07:39:42 +0000 (09:39 +0200)]
xfs: reclaim speculative preallocations for append only files

The XFS XFS_DIFLAG_APPEND maps to the VFS S_APPEND flag, which forbids
writes that don't append at the current EOF.

But the commit originally adding XFS_DIFLAG_APPEND support (commit
a23321e766d in xfs xfs-import repository) also checked it to skip
releasing speculative preallocations, which doesn't make any sense.

Another commit (dd9f438e3290 in the xfs-import repository) later extended
that flag to also report these speculation preallocations which should
not exist in getbmap.

Remove these checks as nothing XFS_DIFLAG_APPEND implies that
preallocations beyond EOF should exist, but explicitly check for
XFS_DIFLAG_APPEND in xfs_file_release to bypass the algorithm that
discard preallocations on the first close as append only files aren't
expected to be written to only once.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: simplify extent lookup in xfs_can_free_eofblocks
Christoph Hellwig [Tue, 13 Aug 2024 07:39:41 +0000 (09:39 +0200)]
xfs: simplify extent lookup in xfs_can_free_eofblocks

xfs_can_free_eofblocks just cares if there is an extent beyond EOF.
Replace the call to xfs_bmapi_read with a xfs_iext_lookup_extent
as we've already checked that extents are read in earlier.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: check XFS_EOFBLOCKS_RELEASED earlier in xfs_release_eofblocks
Christoph Hellwig [Tue, 13 Aug 2024 07:39:40 +0000 (09:39 +0200)]
xfs: check XFS_EOFBLOCKS_RELEASED earlier in xfs_release_eofblocks

If the XFS_EOFBLOCKS_RELEASED flag is set, we are not going to free the
eofblocks, so don't bother locking the inode or performing the checks in
xfs_can_free_eofblocks.  Also switch to a test_and_set operation once
the iolock has been acquire so that only the caller that sets it actually
frees the post-EOF blocks.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: only free posteof blocks on first close
Darrick J. Wong [Tue, 13 Aug 2024 07:39:39 +0000 (09:39 +0200)]
xfs: only free posteof blocks on first close

Certain workloads fragment files on XFS very badly, such as a software
package that creates a number of threads, each of which repeatedly run
the sequence: open a file, perform a synchronous write, and close the
file, which defeats the speculative preallocation mechanism.  We work
around this problem by only deleting posteof blocks the /first/ time a
file is closed to preserve the behavior that unpacking a tarball lays
out files one after the other with no gaps.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
[hch: rebased, updated comment, renamed the flag]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: don't free post-EOF blocks on read close
Dave Chinner [Tue, 13 Aug 2024 07:39:38 +0000 (09:39 +0200)]
xfs: don't free post-EOF blocks on read close

When we have a workload that does open/read/close in parallel with other
allocation, the file becomes rapidly fragmented. This is due to close()
calling xfs_file_release() and removing the speculative preallocation
beyond EOF.

Add a check for a writable context to xfs_file_release to skip the
post-EOF block freeing (an the similarly pointless flushing on truncate
down).

Before:

Test 1: sync write fragmentation counts

/mnt/scratch/file.0: 919
/mnt/scratch/file.1: 916
/mnt/scratch/file.2: 919
/mnt/scratch/file.3: 920
/mnt/scratch/file.4: 920
/mnt/scratch/file.5: 921
/mnt/scratch/file.6: 916
/mnt/scratch/file.7: 918

After:

Test 1: sync write fragmentation counts

/mnt/scratch/file.0: 24
/mnt/scratch/file.1: 24
/mnt/scratch/file.2: 11
/mnt/scratch/file.3: 24
/mnt/scratch/file.4: 3
/mnt/scratch/file.5: 24
/mnt/scratch/file.6: 24
/mnt/scratch/file.7: 23

Signed-off-by: Dave Chinner <dchinner@redhat.com>
[darrick: wordsmithing, fix commit message]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
[hch: ported to the new ->release code structure]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: skip all of xfs_file_release when shut down
Christoph Hellwig [Tue, 13 Aug 2024 07:39:37 +0000 (09:39 +0200)]
xfs: skip all of xfs_file_release when shut down

There is no point in trying to free post-EOF blocks when the file system
is shutdown, as it will just error out ASAP.  Instead return instantly
when xfs_file_release is called on a shut down file system.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: don't bother returning errors from xfs_file_release
Christoph Hellwig [Tue, 13 Aug 2024 07:39:36 +0000 (09:39 +0200)]
xfs: don't bother returning errors from xfs_file_release

While ->release returns int, the only caller ignores the return value.
As we're only doing cleanup work there isn't much of a point in
return a value to start with, so just document the situation instead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: refactor f_op->release handling
Christoph Hellwig [Tue, 13 Aug 2024 07:39:35 +0000 (09:39 +0200)]
xfs: refactor f_op->release handling

Currently f_op->release is split in not very obvious ways.  Fix that by
folding xfs_release into xfs_file_release.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoxfs: remove the i_mode check in xfs_release
Christoph Hellwig [Tue, 13 Aug 2024 07:39:34 +0000 (09:39 +0200)]
xfs: remove the i_mode check in xfs_release

xfs_release is only called from xfs_file_release, which is wired up as
the f_op->release handler for regular files only.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
7 months agoMerge tag 'btree-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 04:33:07 +0000 (10:03 +0530)]
Merge tag 'btree-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: cleanups for inode rooted btree code [v4.2 8/8]

This series prepares the btree code to support realtime reverse mapping btrees
by refactoring xfs_ifork_realloc to be fed a per-btree ops structure so that it
can handle multiple types of inode-rooted btrees.  It moves on to refactoring
the btree code to use the new realloc routines.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'btree-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: standardize the btree maxrecs function parameters
  xfs: replace shouty XFS_BM{BT,DR} macros

7 months agoMerge tag 'xfs-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel...
Chandan Babu R [Tue, 3 Sep 2024 03:45:32 +0000 (09:15 +0530)]
Merge tag 'xfs-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: various bug fixes for 6.12 [7/8]

Various bug fixes for 6.12.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'xfs-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: fix a sloppy memory handling bug in xfs_iroot_realloc
  xfs: fix FITRIM reporting again
  xfs: fix C++ compilation errors in xfs_fs.h

7 months agoMerge tag 'quota-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 03:44:59 +0000 (09:14 +0530)]
Merge tag 'quota-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: cleanups for quota mount [v4.2 6/8]

Refactor the quota file loading code in preparation for adding metadata
directory trees.  Did you know that quotarm works even when quota isn't active?

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'quota-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: refactor loading quota inodes in the regular case

7 months agoMerge tag 'rtalloc-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 03:44:28 +0000 (09:14 +0530)]
Merge tag 'rtalloc-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: cleanups for the realtime allocator [v4.2 5/8]

This third series cleans up the realtime allocator code so that it'll be
somewhat less difficult to figure out what on earth it's doing.  We also
rearrange the fsmap code a bit.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'rtalloc-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: move xfs_ioc_getfsmap out of xfs_ioctl.c
  xfs: rearrange xfs_fsmap.c a little bit
  xfs: replace m_rsumsize with m_rsumblocks
  xfs: remove xfs_{rtbitmap,rtsummary}_wordcount
  xfs: add xchk_setup_nothing and xchk_nothing helpers
  xfs: make the rtalloc start hint a xfs_rtblock_t
  xfs: factor out a xfs_rtallocate_align helper
  xfs: rework the rtalloc fallback handling
  xfs: factor out a xfs_rtallocate helper
  xfs: clean up the ISVALID macro in xfs_bmap_adjacent

7 months agoMerge tag 'rtalloc-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 03:43:51 +0000 (09:13 +0530)]
Merge tag 'rtalloc-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: fixes for the realtime allocator [v4.2 4/8]

While I was reviewing how to integrate realtime allocation groups with
the rt allocator, I noticed several bugs in the existing allocation code
with regards to calculating the maximum range of rtx to scan for free
space.  This series fixes those range bugs and cleans up a few things
too.

I also added a few cleanups from Christoph.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'rtalloc-fixes-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: simplify xfs_rtalloc_query_range
  xfs: remove xfs_rtb_to_rtxrem
  xfs: fix broken variable-sized allocation detection in xfs_rtallocate_extent_block
  xfs: reduce excessive clamping of maxlen in xfs_rtallocate_extent_near
  xfs: clean up xfs_rtallocate_extent_exact a bit
  xfs: refactor aligning bestlen to prod
  xfs: don't scan off the end of the rt volume in xfs_rtallocate_extent_block
  xfs: don't return too-short extents from xfs_rtallocate_extent_block
  xfs: ensure rtx mask/shift are correct after growfs
  xfs: use the recalculated transaction reservation in xfs_growfs_rt_bmblock

7 months agoMerge tag 'rtbitmap-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 03:43:18 +0000 (09:13 +0530)]
Merge tag 'rtbitmap-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: clean up the rtbitmap code [v4.2 3/8]

Here are some cleanups and reorganization of the realtime bitmap code to share
more of that code between userspace and the kernel.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'rtbitmap-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: push transaction join out of xfs_rtbitmap_lock and xfs_rtgroup_lock
  xfs: factor out rtbitmap/summary initialization helpers
  xfs: factor out a xfs_last_rt_bmblock helper
  xfs: factor out a xfs_growfs_rt_bmblock helper
  xfs: push the calls to xfs_rtallocate_range out to xfs_bmap_rtalloc
  xfs: cleanup the calling convention for xfs_rtpick_extent
  xfs: add bounds checking to xfs_rt{bitmap,summary}_read_buf
  xfs: assert a valid limit in xfs_rtfind_forw
  xfs: remove the limit argument to xfs_rtfind_back
  xfs: make the RT rsum_cache mandatory
  xfs: factor out a xfs_validate_rt_geometry helper
  xfs: remove xfs_validate_rtextents

7 months agoMerge tag 'metadir-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux...
Chandan Babu R [Tue, 3 Sep 2024 03:42:47 +0000 (09:12 +0530)]
Merge tag 'metadir-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: cleanups before adding metadata directories [v4.2 2/8]

Before we start adding code for metadata directory trees, let's clean up
some warts in the realtime bitmap code and the inode allocator code.

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'metadir-cleanups-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: pass the icreate args object to xfs_dialloc
  xfs: match on the global RT inode numbers in xfs_is_metadata_inode
  xfs: validate inumber in xfs_iget

7 months agoMerge tag 'atomic-file-commits-6.12_2024-09-02' of https://git.kernel.org/pub/scm...
Chandan Babu R [Tue, 3 Sep 2024 03:41:43 +0000 (09:11 +0530)]
Merge tag 'atomic-file-commits-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux into xfs-6.12-mergeA

xfs: atomic file content commits [v31.1 1/8]

This series creates XFS_IOC_START_COMMIT and XFS_IOC_COMMIT_RANGE ioctls
to perform the exchange only if the target file has not been changed
since a given sampling point.

This new functionality uses the mechanism underlying EXCHANGE_RANGE to
stage and commit file updates such that reader programs will see either
the old contents or the new contents in their entirety, with no chance
of torn writes.  A successful call completion guarantees that the new
contents will be seen even if the system fails.  The pair of ioctls
allows userspace to perform what amounts to a compare and exchange
operation on entire file contents.

Note that there are ongoing arguments in the community about how best to
implement some sort of file data write counter that nfsd could also use
to signal invalidations to clients.  Until such a thing is implemented,
this patch will rely on ctime/mtime updates.

Here are the proposed manual pages:

IOCTL-XFS-COMMIT-RANGE(2) System Calls ManualIOCTL-XFS-COMMIT-RANGE(2)

NAME
       ioctl_xfs_start_commit  -  prepare  to exchange the contents of
       two files ioctl_xfs_commit_range - conditionally  exchange  the
       contents of parts of two files

SYNOPSIS
       #include <sys/ioctl.h>
       #include <xfs/xfs_fs.h>

       int  ioctl(int  file2_fd, XFS_IOC_START_COMMIT, struct xfs_com‐
       mit_range *arg);

       int ioctl(int file2_fd, XFS_IOC_COMMIT_RANGE,  struct  xfs_com‐
       mit_range *arg);

DESCRIPTION
       Given  a  range  of bytes in a first file file1_fd and a second
       range of bytes in a second file  file2_fd,  this  ioctl(2)  ex‐
       changes  the contents of the two ranges if file2_fd passes cer‐
       tain freshness criteria.

       Before exchanging the  contents,  the  program  must  call  the
       XFS_IOC_START_COMMIT   ioctl   to  sample  freshness  data  for
       file2_fd.  If the sampled metadata  does  not  match  the  file
       metadata  at  commit  time,  XFS_IOC_COMMIT_RANGE  will  return
       EBUSY.

       Exchanges are atomic with regards  to  concurrent  file  opera‐
       tions.   Implementations must guarantee that readers see either
       the old contents or the new contents in their entirety, even if
       the system fails.

       The  system  call  parameters are conveyed in structures of the
       following form:

           struct xfs_commit_range {
               __s32    file1_fd;
               __u32    pad;
               __u64    file1_offset;
               __u64    file2_offset;
               __u64    length;
               __u64    flags;
               __u64    file2_freshness[5];
           };

       The field pad must be zero.

       The fields file1_fd, file1_offset, and length define the  first
       range of bytes to be exchanged.

       The fields file2_fd, file2_offset, and length define the second
       range of bytes to be exchanged.

       The field file2_freshness is an opaque field whose contents are
       determined  by  the  kernel.  These file attributes are used to
       confirm that file2_fd has not changed by another  thread  since
       the current thread began staging its own update.

       Both  files must be from the same filesystem mount.  If the two
       file descriptors represent the same file, the byte ranges  must
       not  overlap.   Most  disk-based  filesystems  require that the
       starts of both ranges must be aligned to the file  block  size.
       If  this  is  the  case, the ends of the ranges must also be so
       aligned unless the XFS_EXCHANGE_RANGE_TO_EOF flag is set.

       The field flags control the behavior of the exchange operation.

           XFS_EXCHANGE_RANGE_TO_EOF
                  Ignore the length parameter.  All bytes in  file1_fd
                  from  file1_offset to EOF are moved to file2_fd, and
                  file2's size is set to  (file2_offset+(file1_length-
                  file1_offset)).   Meanwhile, all bytes in file2 from
                  file2_offset to EOF are moved to file1  and  file1's
                  size    is   set   to   (file1_offset+(file2_length-
                  file2_offset)).

           XFS_EXCHANGE_RANGE_DSYNC
                  Ensure that all modified in-core data in  both  file
                  ranges  and  all  metadata updates pertaining to the
                  exchange operation are flushed to persistent storage
                  before  the  call  returns.  Opening either file de‐
                  scriptor with O_SYNC or O_DSYNC will have  the  same
                  effect.

           XFS_EXCHANGE_RANGE_FILE1_WRITTEN
                  Only  exchange sub-ranges of file1_fd that are known
                  to contain data  written  by  application  software.
                  Each  sub-range  may  be  expanded (both upwards and
                  downwards) to align with the file  allocation  unit.
                  For files on the data device, this is one filesystem
                  block.  For files on the realtime  device,  this  is
                  the realtime extent size.  This facility can be used
                  to implement fast atomic  scatter-gather  writes  of
                  any  complexity for software-defined storage targets
                  if all writes are aligned  to  the  file  allocation
                  unit.

           XFS_EXCHANGE_RANGE_DRY_RUN
                  Check  the parameters and the feasibility of the op‐
                  eration, but do not change anything.

RETURN VALUE
       On error, -1 is returned, and errno is set to indicate the  er‐
       ror.

ERRORS
       Error  codes can be one of, but are not limited to, the follow‐
       ing:

       EBADF  file1_fd is not open for reading and writing or is  open
              for  append-only  writes;  or  file2_fd  is not open for
              reading and writing or is open for append-only writes.

       EBUSY  The file2 inode number and timestamps  supplied  do  not
              match file2_fd.

       EINVAL The  parameters  are  not correct for these files.  This
              error can also appear if either file  descriptor  repre‐
              sents  a device, FIFO, or socket.  Disk filesystems gen‐
              erally require the offset and  length  arguments  to  be
              aligned to the fundamental block sizes of both files.

       EIO    An I/O error occurred.

       EISDIR One of the files is a directory.

       ENOMEM The  kernel  was unable to allocate sufficient memory to
              perform the operation.

       ENOSPC There is not enough free space  in  the  filesystem  ex‐
              change the contents safely.

       EOPNOTSUPP
              The filesystem does not support exchanging bytes between
              the two files.

       EPERM  file1_fd or file2_fd are immutable.

       ETXTBSY
              One of the files is a swap file.

       EUCLEAN
              The filesystem is corrupt.

       EXDEV  file1_fd and  file2_fd  are  not  on  the  same  mounted
              filesystem.

CONFORMING TO
       This API is XFS-specific.

USE CASES
       Several use cases are imagined for this system call.  Coordina‐
       tion between multiple threads is performed by the kernel.

       The first is a filesystem defragmenter, which copies  the  con‐
       tents  of  a  file into another file and wishes to exchange the
       space mappings of the two files,  provided  that  the  original
       file has not changed.

       An example program might look like this:

           int fd = open("/some/file", O_RDWR);
           int temp_fd = open("/some", O_TMPFILE | O_RDWR);
           struct stat sb;
           struct xfs_commit_range args = {
               .flags = XFS_EXCHANGE_RANGE_TO_EOF,
           };

           /* gather file2's freshness information */
           ioctl(fd, XFS_IOC_START_COMMIT, &args);
           fstat(fd, &sb);

           /* make a fresh copy of the file with terrible alignment to avoid reflink */
           clone_file_range(fd, NULL, temp_fd, NULL, 1, 0);
           clone_file_range(fd, NULL, temp_fd, NULL, sb.st_size - 1, 0);

           /* commit the entire update */
           args.file1_fd = temp_fd;
           ret = ioctl(fd, XFS_IOC_COMMIT_RANGE, &args);
           if (ret && errno == EBUSY)
               printf("file changed while defrag was underway
");

       The  second is a data storage program that wants to commit non-
       contiguous updates to a file atomically.  This  program  cannot
       coordinate updates to the file and therefore relies on the ker‐
       nel to reject the COMMIT_RANGE command if the file has been up‐
       dated  by  someone else.  This can be done by creating a tempo‐
       rary file, calling FICLONE(2) to share the contents, and  stag‐
       ing  the  updates into the temporary file.  The FULL_FILES flag
       is recommended for this purpose.  The  temporary  file  can  be
       deleted or punched out afterwards.

       An example program might look like this:

           int fd = open("/some/file", O_RDWR);
           int temp_fd = open("/some", O_TMPFILE | O_RDWR);
           struct xfs_commit_range args = {
               .flags = XFS_EXCHANGE_RANGE_TO_EOF,
           };

           /* gather file2's freshness information */
           ioctl(fd, XFS_IOC_START_COMMIT, &args);

           ioctl(temp_fd, FICLONE, fd);

           /* append 1MB of records */
           lseek(temp_fd, 0, SEEK_END);
           write(temp_fd, data1, 1000000);

           /* update record index */
           pwrite(temp_fd, data1, 600, 98765);
           pwrite(temp_fd, data2, 320, 54321);
           pwrite(temp_fd, data2, 15, 0);

           /* commit the entire update */
           args.file1_fd = temp_fd;
           ret = ioctl(fd, XFS_IOC_COMMIT_RANGE, &args);
           if (ret && errno == EBUSY)
               printf("file changed before commit; will roll back
");

NOTES
       Some  filesystems may limit the amount of data or the number of
       extents that can be exchanged in a single call.

SEE ALSO
       ioctl(2)

XFS                           2024-02-18     IOCTL-XFS-COMMIT-RANGE(2)

With a bit of luck, this should all go splendidly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
* tag 'atomic-file-commits-6.12_2024-09-02' of https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux:
  xfs: introduce new file range commit ioctls

7 months agoxfs: standardize the btree maxrecs function parameters
Darrick J. Wong [Fri, 30 Aug 2024 22:37:21 +0000 (15:37 -0700)]
xfs: standardize the btree maxrecs function parameters

Standardize the parameters in xfs_{alloc,bm,ino,rmap,refcount}bt_maxrecs
so that we have consistent calling conventions.  This doesn't affect the
kernel that much, but enables us to clean up userspace a bit.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: fix a sloppy memory handling bug in xfs_iroot_realloc
Darrick J. Wong [Fri, 30 Aug 2024 22:37:20 +0000 (15:37 -0700)]
xfs: fix a sloppy memory handling bug in xfs_iroot_realloc

While refactoring code, I noticed that when xfs_iroot_realloc tries to
shrink a bmbt root block, it allocates a smaller new block and then
copies "records" and pointers to the new block.  However, bmbt root
blocks cannot ever be leaves, which means that it's not technically
correct to copy records.  We /should/ be copying keys.

Note that this has never resulted in actual memory corruption because
sizeof(bmbt_rec) == (sizeof(bmbt_key) + sizeof(bmbt_ptr)).  However,
this will no longer be true when we start adding realtime rmap stuff,
so fix this now.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: refactor loading quota inodes in the regular case
Darrick J. Wong [Fri, 30 Aug 2024 22:37:17 +0000 (15:37 -0700)]
xfs: refactor loading quota inodes in the regular case

Create a helper function to load quota inodes in the case where the
dqtype and the sb quota inode fields correspond.  This is true for
nearly all the iget callsites in the quota code, except for when we're
switching the group and project quota inodes.  We'll need this in
subsequent patches to make the metadir handling less convoluted.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: replace shouty XFS_BM{BT,DR} macros
Darrick J. Wong [Fri, 30 Aug 2024 22:37:20 +0000 (15:37 -0700)]
xfs: replace shouty XFS_BM{BT,DR} macros

Replace all the shouty bmap btree and bmap disk root macros with actual
functions.

sed \
 -e 's/XFS_BMBT_BLOCK_LEN/xfs_bmbt_block_len/g' \
 -e 's/XFS_BMBT_REC_ADDR/xfs_bmbt_rec_addr/g' \
 -e 's/XFS_BMBT_KEY_ADDR/xfs_bmbt_key_addr/g' \
 -e 's/XFS_BMBT_PTR_ADDR/xfs_bmbt_ptr_addr/g' \
 -e 's/XFS_BMDR_REC_ADDR/xfs_bmdr_rec_addr/g' \
 -e 's/XFS_BMDR_KEY_ADDR/xfs_bmdr_key_addr/g' \
 -e 's/XFS_BMDR_PTR_ADDR/xfs_bmdr_ptr_addr/g' \
 -e 's/XFS_BMAP_BROOT_PTR_ADDR/xfs_bmap_broot_ptr_addr/g' \
 -e 's/XFS_BMAP_BROOT_SPACE_CALC/xfs_bmap_broot_space_calc/g' \
 -e 's/XFS_BMAP_BROOT_SPACE/xfs_bmap_broot_space/g' \
 -e 's/XFS_BMDR_SPACE_CALC/xfs_bmdr_space_calc/g' \
 -e 's/XFS_BMAP_BMDR_SPACE/xfs_bmap_bmdr_space/g' \
 -i $(git ls-files fs/xfs/*.[ch] fs/xfs/libxfs/*.[ch] fs/xfs/scrub/*.[ch])

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: fix FITRIM reporting again
Darrick J. Wong [Fri, 30 Aug 2024 22:37:19 +0000 (15:37 -0700)]
xfs: fix FITRIM reporting again

Don't report FITRIMming more bytes than possibly exist in the
filesystem.

Fixes: 410e8a18f8e93 ("xfs: don't bother reporting blocks trimmed via FITRIM")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: fix C++ compilation errors in xfs_fs.h
Darrick J. Wong [Fri, 30 Aug 2024 22:37:18 +0000 (15:37 -0700)]
xfs: fix C++ compilation errors in xfs_fs.h

Several people reported C++ compilation errors due to things that C
compilers allow but C++ compilers do not.  Fix both of these problems,
and hope there aren't more of these brown paper bags in 2 months when we
finally get these fixes through the process into a released xfsprogs.

NOTE: I am submitting this bugfix over the objections of a former
maintainer, who insists that we should remove this function from the
published userspace ABI instead of fixing the C++ compilation errors.
No deprecation period, no discussion, just a hard drop of an already
provided and correct C function, which would be in contravention of
Linus' rules.  IOWs, removing ABI that have already shipped in a
released kernel requires a careful deprecation period, so I will let
that maintainer run that process.

Reported-by: kernel@mattwhitlock.name
Reported-by: sam@gentoo.org
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=219203
Fixes: 233f4e12bbb2c ("xfs: add parent pointer ioctls")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: move xfs_ioc_getfsmap out of xfs_ioctl.c
Darrick J. Wong [Fri, 30 Aug 2024 22:37:16 +0000 (15:37 -0700)]
xfs: move xfs_ioc_getfsmap out of xfs_ioctl.c

Move this function out of xfs_ioctl.c to reduce the clutter in there,
and make the entire getfsmap implementation self-contained in a single
file.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: simplify xfs_rtalloc_query_range
Christoph Hellwig [Fri, 30 Aug 2024 22:37:08 +0000 (15:37 -0700)]
xfs: simplify xfs_rtalloc_query_range

There isn't much of a good reason to pass the xfs_rtalloc_rec structures
that describe extents to xfs_rtalloc_query_range as we really just want
a lower and upper bound xfs_rtxnum_t.  Pass the rtxnum directly and
simply the interface.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: push transaction join out of xfs_rtbitmap_lock and xfs_rtgroup_lock
Christoph Hellwig [Fri, 30 Aug 2024 22:36:59 +0000 (15:36 -0700)]
xfs: push transaction join out of xfs_rtbitmap_lock and xfs_rtgroup_lock

To prepare for being able to join an already locked rtbitmap inode to a
transaction split out separate helpers for joining the transaction from
the locking helpers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: pass the icreate args object to xfs_dialloc
Darrick J. Wong [Fri, 30 Aug 2024 22:36:49 +0000 (15:36 -0700)]
xfs: pass the icreate args object to xfs_dialloc

Pass the xfs_icreate_args object to xfs_dialloc since we can extract the
relevant mode (really just the file type) and parent inumber from there.
This simplifies the calling convention in preparation for the next
patch.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: introduce new file range commit ioctls
Darrick J. Wong [Fri, 30 Aug 2024 22:36:47 +0000 (15:36 -0700)]
xfs: introduce new file range commit ioctls

This patch introduces two more new ioctls to manage atomic updates to
file contents -- XFS_IOC_START_COMMIT and XFS_IOC_COMMIT_RANGE.  The
commit mechanism here is exactly the same as what XFS_IOC_EXCHANGE_RANGE
does, but with the additional requirement that file2 cannot have changed
since some sampling point.  The start-commit ioctl performs the sampling
of file attributes.

Note: This patch currently samples i_ctime during START_COMMIT and
checks that it hasn't changed during COMMIT_RANGE.  This isn't entirely
safe in kernels prior to 6.12 because ctime only had coarse grained
granularity and very fast updates could collide with a COMMIT_RANGE.
With the multi-granularity ctime introduced by Jeff Layton, it's now
possible to update ctime such that this does not happen.

It is critical, then, that this patch must not be backported to any
kernel that does not support fine-grained file change timestamps.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Acked-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: rearrange xfs_fsmap.c a little bit
Darrick J. Wong [Fri, 30 Aug 2024 22:37:15 +0000 (15:37 -0700)]
xfs: rearrange xfs_fsmap.c a little bit

The order of the functions in this file has gotten a little confusing
over the years.  Specifically, the two data device implementations
(bnobt and rmapbt) could be adjacent in the source code instead of split
in two by the logdev and rtdev fsmap implementations.  We're about to
add more functionality to this file, so rearrange things now.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: remove xfs_rtb_to_rtxrem
Christoph Hellwig [Fri, 30 Aug 2024 22:37:07 +0000 (15:37 -0700)]
xfs: remove xfs_rtb_to_rtxrem

Simplify the number of block number conversion helpers by removing
xfs_rtb_to_rtxrem.  Any recent compiler is smart enough to eliminate
the double divisions if using separate xfs_rtb_to_rtx and
xfs_rtb_to_rtxoff calls.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: factor out rtbitmap/summary initialization helpers
Christoph Hellwig [Fri, 30 Aug 2024 22:36:59 +0000 (15:36 -0700)]
xfs: factor out rtbitmap/summary initialization helpers

Add helpers to libxfs that can be shared by growfs and mkfs for
initializing the rtbitmap and summary, and by passing the optional data
pointer also by repair for rebuilding them.  This will become even more
useful when the rtgroups feature adds a metadata header to each block,
which means even more shared code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: minor documentation and data advance tweaks]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: match on the global RT inode numbers in xfs_is_metadata_inode
Christoph Hellwig [Fri, 30 Aug 2024 22:36:49 +0000 (15:36 -0700)]
xfs: match on the global RT inode numbers in xfs_is_metadata_inode

Match the inode number instead of the inode pointers, as the inode
pointers in the superblock will go away soon.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: port to my tree, make the parameter a const pointer]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: replace m_rsumsize with m_rsumblocks
Christoph Hellwig [Fri, 30 Aug 2024 22:37:15 +0000 (15:37 -0700)]
xfs: replace m_rsumsize with m_rsumblocks

Track the RT summary file size in blocks, just like the RT bitmap
file.  While we have users of both units, blocks are used slightly
more often and this matches the bitmap file for consistency.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: fix broken variable-sized allocation detection in xfs_rtallocate_extent_block
Darrick J. Wong [Fri, 30 Aug 2024 22:37:06 +0000 (15:37 -0700)]
xfs: fix broken variable-sized allocation detection in xfs_rtallocate_extent_block

This function tries to find a suitable free space extent starting from
a particular rtbitmap block.  Some time ago, I added a clamping function
to prevent the free space scans from running off the end of the bitmap,
but I didn't quite get the logic right.

Let's say there's an allocation request with a minlen of 5 and a maxlen
of 32 and we're scanning the last rtbitmap block.  If we come within 4
rtx of the end of the rt volume, maxlen will get clamped to 4.  If the
next 3 rtx are free, we could have satisfied the allocation, but the
code setting partial besti/bestlen for "minlen < maxlen" will think that
we're doing a non-variable allocation and ignore it.

The root of this problem is overwriting maxlen; I should have stuffed
the results in a different variable, which would not have introduced
this bug.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: factor out a xfs_last_rt_bmblock helper
Christoph Hellwig [Fri, 30 Aug 2024 22:36:58 +0000 (15:36 -0700)]
xfs: factor out a xfs_last_rt_bmblock helper

Add helper to calculate the last currently used rt bitmap block to
better structure the growfs code and prepare for future changes to it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: validate inumber in xfs_iget
Darrick J. Wong [Fri, 30 Aug 2024 22:36:48 +0000 (15:36 -0700)]
xfs: validate inumber in xfs_iget

Actually use the inumber validator to check the argument passed in here.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
7 months agoxfs: remove xfs_{rtbitmap,rtsummary}_wordcount
Christoph Hellwig [Fri, 30 Aug 2024 22:37:14 +0000 (15:37 -0700)]
xfs: remove xfs_{rtbitmap,rtsummary}_wordcount

xfs_rtbitmap_wordcount and xfs_rtsummary_wordcount are currently unused,
so remove them to simplify refactoring other rtbitmap helpers.  They
can be added back or simply open coded when actually needed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: reduce excessive clamping of maxlen in xfs_rtallocate_extent_near
Darrick J. Wong [Fri, 30 Aug 2024 22:37:05 +0000 (15:37 -0700)]
xfs: reduce excessive clamping of maxlen in xfs_rtallocate_extent_near

The near rt allocator employs two allocation strategies -- first it
tries to allocate at exactly @start.  If that fails, it will pivot back
and forth around that starting point looking for an appropriately sized
free space.

However, I clamped maxlen ages ago to prevent the exact allocation scan
from running off the end of the rt volume.  This, I realize, was
excessive.  If the allocation request is (say) for 32 rtx but the start
position is 5 rtx from the end of the volume, we clamp maxlen to 5.  If
the exact allocation fails, we then pivot back and forth looking for 5
rtx, even though the original intent was to try to get 32 rtx.

If we then find 5 rtx when we could have gotten 32 rtx, we've not done
as well as we could have.  This may be moot if the caller immediately
comes back for more space, but it might not be.  Either way, we can do
better here.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: factor out a xfs_growfs_rt_bmblock helper
Christoph Hellwig [Fri, 30 Aug 2024 22:36:57 +0000 (15:36 -0700)]
xfs: factor out a xfs_growfs_rt_bmblock helper

Add a helper to contain the per-rtbitmap block logic in xfs_growfs_rt.

Note that this helper now allocates a new fake mount structure for
each rtbitmap block iteration instead of reusing the memory for an
entire growfs call.  Compared to all the other work done when freeing
the blocks the overhead for this is in the noise and it keeps the code
nicely modular.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: add xchk_setup_nothing and xchk_nothing helpers
Darrick J. Wong [Fri, 30 Aug 2024 22:37:13 +0000 (15:37 -0700)]
xfs: add xchk_setup_nothing and xchk_nothing helpers

Add common helpers for no-op scrubbing methods.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
[hch: split from a larger patch]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: clean up xfs_rtallocate_extent_exact a bit
Darrick J. Wong [Fri, 30 Aug 2024 22:37:05 +0000 (15:37 -0700)]
xfs: clean up xfs_rtallocate_extent_exact a bit

Before we start doing more surgery on the rt allocator, let's clean up
the exact allocator so that it doesn't change its arguments and uses the
helper introduced in the previous patch.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: push the calls to xfs_rtallocate_range out to xfs_bmap_rtalloc
Christoph Hellwig [Fri, 30 Aug 2024 22:36:56 +0000 (15:36 -0700)]
xfs: push the calls to xfs_rtallocate_range out to xfs_bmap_rtalloc

Currently the various low-level RT allocator functions call into
xfs_rtallocate_range directly, which ties them into the locking protocol
for the RT bitmap.  As these helpers already return the allocated range,
lift the call to xfs_rtallocate_range into xfs_bmap_rtalloc so that it
happens as high as possible in the stack, which will simplify future
changes to the locking protocol.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: make the rtalloc start hint a xfs_rtblock_t
Christoph Hellwig [Fri, 30 Aug 2024 22:37:12 +0000 (15:37 -0700)]
xfs: make the rtalloc start hint a xfs_rtblock_t

0 is a valid start RT extent, and with pending changes it will become
both more common and non-unique.  Switch to pass a xfs_rtblock_t instead
so that we can use NULLRTBLOCK to determine if a hint was set or not.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: refactor aligning bestlen to prod
Darrick J. Wong [Fri, 30 Aug 2024 22:37:04 +0000 (15:37 -0700)]
xfs: refactor aligning bestlen to prod

There are two places in xfs_rtalloc.c where we want to make sure that a
count of rt extents is aligned with a particular prod(uct) factor.  In
one spot, we actually use rounddown(), albeit unnecessarily if prod < 2.
In the other case, we open-code this rounding inefficiently by promoting
the 32-bit length value to a 64-bit value and then performing a 64-bit
division to figure out the subtraction.

Refactor this into a single helper that uses the correct types and
division method for the type, and skips the division entirely unless
prod is large enough to make a difference.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: cleanup the calling convention for xfs_rtpick_extent
Christoph Hellwig [Fri, 30 Aug 2024 22:36:55 +0000 (15:36 -0700)]
xfs: cleanup the calling convention for xfs_rtpick_extent

xfs_rtpick_extent never returns an error.  Do away with the error return
and directly return the picked extent instead of doing that through a
call by reference argument.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: factor out a xfs_rtallocate_align helper
Christoph Hellwig [Fri, 30 Aug 2024 22:37:11 +0000 (15:37 -0700)]
xfs: factor out a xfs_rtallocate_align helper

Split the code to calculate the aligned allocation request from
xfs_bmap_rtalloc into a separate self-contained helper.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: don't scan off the end of the rt volume in xfs_rtallocate_extent_block
Darrick J. Wong [Fri, 30 Aug 2024 22:37:03 +0000 (15:37 -0700)]
xfs: don't scan off the end of the rt volume in xfs_rtallocate_extent_block

The loop conditional here is not quite correct because an rtbitmap block
can represent rtextents beyond the end of the rt volume.  There's no way
that it makes sense to scan for free space beyond EOFS, so don't do it.
This overrun has been present since v2.6.0.

Also fix the type of bestlen, which was incorrectly converted.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: add bounds checking to xfs_rt{bitmap,summary}_read_buf
Christoph Hellwig [Fri, 30 Aug 2024 22:36:54 +0000 (15:36 -0700)]
xfs: add bounds checking to xfs_rt{bitmap,summary}_read_buf

Add a corruption check for passing an invalid block number, which is a
lot easier to understand than the xfs_bmapi_read failure later on.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: rework the rtalloc fallback handling
Christoph Hellwig [Fri, 30 Aug 2024 22:37:10 +0000 (15:37 -0700)]
xfs: rework the rtalloc fallback handling

xfs_rtallocate currently has two fallbacks, when an allocation fails:

 1) drop the requested extent size alignment, if any, and retry
 2) ignore the locality hint

Oddly enough it does those in order, as trying a different location
is more in line with what the user asked for, and does it in a very
unstructured way.

Lift the fallback to try to allocate without the locality hint into
xfs_rtallocate to both perform them in a more sensible order and to
clean up the code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: don't return too-short extents from xfs_rtallocate_extent_block
Darrick J. Wong [Fri, 30 Aug 2024 22:37:02 +0000 (15:37 -0700)]
xfs: don't return too-short extents from xfs_rtallocate_extent_block

If xfs_rtallocate_extent_block is asked for a variable-sized allocation,
it will try to return the best-sized free extent, which is apparently
the largest one that it finds starting in this rtbitmap block.  It will
then trim the size of the extent as needed to align it with prod.

However, it misses one thing -- rounding down the best-fit candidate to
the required alignment could make the extent shorter than minlen.  In
the case where minlen > 1, we'd rather the caller relaxed its alignment
requirements and tried again, as the allocator already supports that.

Returning a too-short extent that causes xfs_bmapi_write to return
ENOSR if there aren't enough nmaps to handle multiple new allocations,
which can then cause filesystem shutdowns.

I haven't seen this happen on any production systems, but then I don't
think it's very common to set a per-file extent size hint on realtime
files.  I tripped it while working on the rtgroups feature and pounding
on the realtime allocator enthusiastically.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
7 months agoxfs: assert a valid limit in xfs_rtfind_forw
Christoph Hellwig [Fri, 30 Aug 2024 22:36:54 +0000 (15:36 -0700)]
xfs: assert a valid limit in xfs_rtfind_forw

Protect against developers passing stupid limits when refactoring the
RT code once again.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: factor out a xfs_rtallocate helper
Christoph Hellwig [Fri, 30 Aug 2024 22:37:10 +0000 (15:37 -0700)]
xfs: factor out a xfs_rtallocate helper

Split out a helper from xfs_rtallocate that performs the actual
allocation.  This keeps the scope of the xfs_rtalloc_args structure
contained, and prepares for rtgroups support.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: ensure rtx mask/shift are correct after growfs
Christoph Hellwig [Fri, 30 Aug 2024 22:37:01 +0000 (15:37 -0700)]
xfs: ensure rtx mask/shift are correct after growfs

When growfs sets an extent size, it doesn't updated the m_rtxblklog and
m_rtxblkmask values, which could lead to incorrect usage of them if they
were set before and can't be used for the new extent size.

Add a xfs_mount_sb_set_rextsize helper that updates the two fields, and
also use it when calculating the new RT geometry instead of disabling
the optimization there.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: remove the limit argument to xfs_rtfind_back
Christoph Hellwig [Fri, 30 Aug 2024 22:36:53 +0000 (15:36 -0700)]
xfs: remove the limit argument to xfs_rtfind_back

All callers pass a 0 limit to xfs_rtfind_back, so remove the argument
and hard code it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: clean up the ISVALID macro in xfs_bmap_adjacent
Christoph Hellwig [Fri, 30 Aug 2024 22:37:09 +0000 (15:37 -0700)]
xfs: clean up the ISVALID macro in xfs_bmap_adjacent

Turn the  ISVALID macro defined and used inside in xfs_bmap_adjacent
that relies on implict context into a proper inline function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: use the recalculated transaction reservation in xfs_growfs_rt_bmblock
Christoph Hellwig [Fri, 30 Aug 2024 22:37:00 +0000 (15:37 -0700)]
xfs: use the recalculated transaction reservation in xfs_growfs_rt_bmblock

After going great length to calculate the transaction reservation for
the new geometry, we should also use it to allocate the transaction it
was calculated for.

Fixes: 578bd4ce7100 ("xfs: recompute growfsrtfree transaction reservation while growing rt volume")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: make the RT rsum_cache mandatory
Christoph Hellwig [Fri, 30 Aug 2024 22:36:52 +0000 (15:36 -0700)]
xfs: make the RT rsum_cache mandatory

Currently the RT mount code simply ignores an allocation failure for the
rsum_cache.  The code mostly works fine with it, but not having it leads
to nasty corner cases in the growfs code that we don't really handle
well.  Switch to failing the mount if we can't allocate the memory, the
file system would not exactly be useful in such a constrained environment
to start with.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: factor out a xfs_validate_rt_geometry helper
Christoph Hellwig [Fri, 30 Aug 2024 22:36:51 +0000 (15:36 -0700)]
xfs: factor out a xfs_validate_rt_geometry helper

Split the RT geometry validation in the early mount code into a
helper than can be reused by repair (from which this code was
apparently originally stolen anyway).

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: u64 return value for calc_rbmblocks]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoxfs: remove xfs_validate_rtextents
Christoph Hellwig [Fri, 30 Aug 2024 22:36:50 +0000 (15:36 -0700)]
xfs: remove xfs_validate_rtextents

Replace xfs_validate_rtextents with an open coded check for 0
rtextents.  The name for the function implies it does a lot more
than a zero check, which is more obvious when open coded.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
7 months agoLinux 6.11-rc6
Linus Torvalds [Sun, 1 Sep 2024 07:46:02 +0000 (19:46 +1200)]
Linux 6.11-rc6

7 months agoMerge tag 'v6.11-rc5-smb-client-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sun, 1 Sep 2024 03:49:26 +0000 (15:49 +1200)]
Merge tag 'v6.11-rc5-smb-client-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull smb client fixes from Steve French:

 - copy_file_range fix

 - two read fixes including read past end of file rc fix and read retry
   crediting fix

 - falloc zero range fix

* tag 'v6.11-rc5-smb-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Fix FALLOC_FL_ZERO_RANGE to preflush buffered part of target region
  cifs: Fix copy offload to flush destination region
  netfs, cifs: Fix handling of short DIO read
  cifs: Fix lack of credit renegotiation on read retry

7 months agoMerge tag 'bcachefs-2024-08-21' of https://github.com/koverstreet/bcachefs
Linus Torvalds [Sun, 1 Sep 2024 03:23:20 +0000 (15:23 +1200)]
Merge tag 'bcachefs-2024-08-21' of https://github.com/koverstreet/bcachefs

Push bcachefs fixes from Kent Overstreet:
 "The data corruption in the buffered write path is troubling; inode
  lock should not have been able to cause that...

   - Fix a rare data corruption in the rebalance path, caught as a nonce
     inconsistency on encrypted filesystems

   - Revert lockless buffered write path

   - Mark more errors as autofix"

* tag 'bcachefs-2024-08-21' of https://github.com/koverstreet/bcachefs:
  bcachefs: Mark more errors as autofix
  bcachefs: Revert lockless buffered IO path
  bcachefs: Fix bch2_extents_match() false positive
  bcachefs: Fix failure to return error in data_update_index_update()

7 months agobcachefs: Mark more errors as autofix
Kent Overstreet [Thu, 22 Aug 2024 15:47:32 +0000 (11:47 -0400)]
bcachefs: Mark more errors as autofix

errors that are known to always be safe to fix should be autofix: this
should be most errors even at this point, but that will need some
thorough review.

note that errors are still logged in the superblock, so we'll still know
that they happened.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
7 months agobcachefs: Revert lockless buffered IO path
Kent Overstreet [Sat, 31 Aug 2024 21:44:51 +0000 (17:44 -0400)]
bcachefs: Revert lockless buffered IO path

We had a report of data corruption on nixos when building installer
images.

https://github.com/NixOS/nixpkgs/pull/321055#issuecomment-2184131334

It seems that writes are being dropped, but only when issued by QEMU,
and possibly only in snapshot mode. It's undetermined if it's write
calls are being dropped or dirty folios.

Further testing, via minimizing the original patch to just the change
that skips the inode lock on non appends/truncates, reveals that it
really is just not taking the inode lock that causes the corruption: it
has nothing to do with the other logic changes for preserving write
atomicity in corner cases.

It's also kernel config dependent: it doesn't reproduce with the minimal
kernel config that ktest uses, but it does reproduce with nixos's distro
config. Bisection the kernel config initially pointer the finger at page
migration or compaction, but it appears that was erroneous; we haven't
yet determined what kernel config option actually triggers it.

Sadly it appears this will have to be reverted since we're getting too
close to release and my plate is full, but we'd _really_ like to fully
debug it.

My suspicion is that this patch is exposing a preexisting bug - the
inode lock actually covers very little in IO paths, and we have a
different lock (the pagecache add lock) that guards against races with
truncate here.

Fixes: 7e64c86cdc6c ("bcachefs: Buffered write path now can avoid the inode lock")
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
7 months agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux...
Linus Torvalds [Sat, 31 Aug 2024 21:18:48 +0000 (09:18 +1200)]
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging

Pull misc fixes from Guenter Roeck.

These are fixes for regressions that Guenther has been reporting, and
the maintainers haven't picked up and sent in. With rc6 fairly imminent,
I'm taking them directly from Guenter.

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  apparmor: fix policy_unpack_test on big endian systems
  Revert "MIPS: csrc-r4k: Apply verification clocksource flags"
  microblaze: don't treat zero reserved memory regions as error

7 months agoMerge tag 'pwrseq-fixes-for-v6.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 31 Aug 2024 21:07:44 +0000 (09:07 +1200)]
Merge tag 'pwrseq-fixes-for-v6.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux

Pull power sequencing fix from Bartosz Golaszewski:
 "A follow-up fix for the power sequencing subsystem. It turned out the
  previous fix for this driver was incomplete and broke the WLAN support
  on some platforms. This addresses the issue.

   - set the direction of the wlan-enable GPIO to output after
     requesting it as-is"

* tag 'pwrseq-fixes-for-v6.11-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux:
  power: sequencing: qcom-wcn: set the wlan-enable GPIO to output

7 months agopower: sequencing: qcom-wcn: set the wlan-enable GPIO to output
Bartosz Golaszewski [Fri, 23 Aug 2024 11:55:00 +0000 (13:55 +0200)]
power: sequencing: qcom-wcn: set the wlan-enable GPIO to output

Commit a9aaf1ff88a8 ("power: sequencing: request the WLAN enable GPIO
as-is") broke WLAN on boards on which the wlan-enable GPIO enabling the
wifi module isn't in output mode by default. We need to set direction to
output while retaining the value that was already set to keep the ath
module on if it's already started.

Fixes: a9aaf1ff88a8 ("power: sequencing: request the WLAN enable GPIO as-is")
Link: https://lore.kernel.org/r/20240823115500.37280-1-brgl@bgdev.pl
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>