Darrick J. Wong [Wed, 29 May 2024 04:11:50 +0000 (21:11 -0700)]
xfs: add a realtime flag to the rmap update log redo items
Extend the rmap update (RUI) log items with a new realtime flag that
indicates that the updates apply against the realtime rmapbt. We'll
wire up the actual rmap code later.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:49 +0000 (21:11 -0700)]
xfs: prepare rmap functions to deal with rtrmapbt
Prepare the high-level rmap functions to deal with the new realtime
rmapbt and its slightly different conventions. Provide the ability
to talk to either rmapbt or rtrmapbt formats from the same high
level code.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:49 +0000 (21:11 -0700)]
xfs: add realtime rmap btree operations
Implement the generic btree operations needed to manipulate rtrmap
btree blocks. This is different from the regular rmapbt in that we
allocate space from the filesystem at large, and are neither
constrained to the free space nor any particular AG.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:48 +0000 (21:11 -0700)]
xfs: realtime rmap btree transaction reservations
Make sure that there's enough log reservation to handle mapping
and unmapping realtime extents. We have to reserve enough space
to handle a split in the rtrmapbt to add the record and a second
split in the regular rmapbt to record the rtrmapbt split.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:47 +0000 (21:11 -0700)]
xfs: define the on-disk realtime rmap btree format
Start filling out the rtrmap btree implementation. Start with the
on-disk btree format; add everything needed to read, write and
manipulate rmap btree blocks. This prepares the way for connecting the
btree operations implementation.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:47 +0000 (21:11 -0700)]
xfs: introduce realtime rmap btree definitions
Add new realtime rmap btree definitions. The realtime rmap btree will
be rooted from a hidden inode, but has its own shape and therefore
needs to have most of its own separate types.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:46 +0000 (21:11 -0700)]
xfs: simplify the xfs_rmap_{alloc,free}_extent calling conventions
Simplify the calling conventions by allowing callers to pass a fsbno
(xfs_fsblock_t) directly into these functions, since we're just going to
set it in a struct anyway.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:46 +0000 (21:11 -0700)]
xfs: prepare rmap btree cursor tracepoints for realtime
Rework the rmap btree cursor tracepoints in preparation to handle the
realtime rmap btree cursor. Mostly this involves renaming the field to
"rmapbno" and extracting the group number from the cursor when possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:33 +0000 (21:11 -0700)]
xfs: update btree keys correctly when _insrec splits an inode root block
In commit 2c813ad66a72, I partially fixed a bug wherein xfs_btree_insrec
would erroneously try to update the parent's key for a block that had
been split if we decided to insert the new record into the new block.
The solution was to detect this situation and update the in-core key
value that we pass up to the caller so that the caller will (eventually)
add the new block to the parent level of the tree with the correct key.
However, I missed a subtlety about the way inode-rooted btrees work. If
the full block was a maximally sized inode root block, we'll solve that
fullness by moving the root block's records to a new block, resizing the
root block, and updating the root to point to the new block. We don't
pass a pointer to the new block to the caller because that work has
already been done. The new record will /always/ land in the new block,
so in this case we need to use xfs_btree_update_keys to update the keys.
This bug can theoretically manifest itself in the very rare case that we
split a bmbt root block and the new record lands in the very first slot
of the new block, though I've never managed to trigger it in practice.
However, it is very easy to reproduce by running generic/522 with the
realtime rmapbt patchset if rtinherit=1.
Fixes: 2c813ad66a72 ("xfs: support btrees with overlapping intervals for keys") Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:32 +0000 (21:11 -0700)]
xfs: support storing records in the inode core root
Add the necessary flags and code so that we can support storing leaf
records in the inode root block of a btree. This hasn't been necessary
before, but the realtime rmapbt will need to be able to do this.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:32 +0000 (21:11 -0700)]
xfs: hoist the node iroot update code out of xfs_btree_kill_iroot
In preparation for allowing records in an inode btree root, hoist the
code that copies keyptrs from an existing node child into the root block
to a separate function. Remove some unnecessary conditionals and clean
up a few function calls in the new function. Note that this change
reorders the ->free_block call with respect to the change in bc_nlevels
to make it easier to support inode root leaf blocks in the next patch.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:31 +0000 (21:11 -0700)]
xfs: hoist the node iroot update code out of xfs_btree_new_iroot
In preparation for allowing records in an inode btree root, hoist the
code that copies keyptrs from an existing node root into a child block
to a separate function. Note that the new function explicitly computes
the keys of the new child block and stores that in the root block; while
the bmap btree could rely on leaving the key alone, realtime rmap needs
to set the new high key.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:30 +0000 (21:11 -0700)]
xfs: generalize the btree root reallocation function
In preparation for storing realtime rmap btree roots in an inode fork,
make xfs_iroot_realloc take an ops structure that takes care of all the
btree-specific geometry pieces.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:30 +0000 (21:11 -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>
Darrick J. Wong [Wed, 29 May 2024 04:11:28 +0000 (21:11 -0700)]
xfs: move the zero records logic into xfs_bmap_broot_space_calc
The bmap btree cannot ever have zero records in an incore btree block.
If the number of records drops to zero, that means we're converting the
fork to extents format and are trying to remove the tree. This logic
won't hold for the future realtime rmap btree, so move the logic into
the bmbt code.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:28 +0000 (21:11 -0700)]
xfs: hoist the code that moves the incore inode fork broot memory
Whenever we change the size of the memory buffer holding an inode fork
btree root block, we have to copy the contents over. Refactor all this
into a single function that handles both, in preparation for making
xfs_iroot_realloc more generic.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:27 +0000 (21:11 -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>
Darrick J. Wong [Wed, 29 May 2024 04:11:27 +0000 (21:11 -0700)]
xfs: refactor creation of bmap btree roots
Now that we've created inode fork helpers to allocate and free btree
roots, create a new bmap btree helper to create a new bmbt root, and
refactor the extents <-> btree conversion functions to use our new
helpers.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:26 +0000 (21:11 -0700)]
xfs: refactor the allocation and freeing of incore inode fork btree roots
Refactor the code that allocates and freese the incore inode fork btree
roots. This will help us disentangle some of the weird logic when we're
creating and tearing down inode-based btrees.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:26 +0000 (21:11 -0700)]
xfs: replace shouty XFS_BM{BT,DR} macros
Replace all the shouty bmap btree and bmap disk root macros with actual
functions, and fix a type handling error in the xattr code that the
macros previously didn't care about.
Darrick J. Wong [Wed, 29 May 2024 04:11:24 +0000 (21:11 -0700)]
xfs: scrub each rtgroup's portion of the rtbitmap separately
Create a new scrub type code so that userspace can scrub each rtgroup's
portion of the rtbitmap file separately. This reduces the long tail
latency that results from scanning the entire bitmap all at once, and
prepares us for future patchsets, wherein we'll need to be able to lock
a specific rtgroup so that we can rebuild that rtgroup's part of the
rtbitmap contents from the rtgroup's rmap btree.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 26 Jun 2024 18:19:00 +0000 (11:19 -0700)]
xfs: don't coalesce file mappings that cross allocation group boundaries
The bmbt scrubber will combine file mappings if they are mergeable to
reduce the number of cross-referencing checks. However, we shouldn't
combine mappings that cross rt group boundaries because that will cause
verifiers to trip incorrectly.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Christoph Hellwig [Mon, 8 Jul 2024 21:38:06 +0000 (14:38 -0700)]
xfs: don't merge ioends across RTGs
Unlike AGs, RTGs don't always have metadata in their first blocks, and
thus we don't get automatic protection from merging I/O completions
across RTG boundaries. Add code to set the IOMAP_F_BOUNDARY flag for
ioends that start at the first block of a RTG so that they never get
merged into the previous ioend.
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>
Darrick J. Wong [Wed, 29 May 2024 04:11:54 +0000 (21:11 -0700)]
xfs: use realtime EFI to free extents when rtgroups are enabled
When rmap is enabled, XFS expects a certain order of operations, which
is: 1) remove the file mapping, 2) remove the reverse mapping, and then
3) free the blocks. When reflink is enabled, XFS replaces (3) with a
deferred refcount decrement operation that can schedule freeing the
blocks if that was the last refcount.
For realtime files, xfs_bmap_del_extent_real tries to do 1 and 3 in the
same transaction, which will break both rmap and reflink unless we
switch it to use realtime EFIs. Both rmap and reflink depend on the
rtgroups feature, so let's turn on EFIs for all rtgroups filesystems.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:22 +0000 (21:11 -0700)]
xfs: force swapext to a realtime file to use the file content exchange ioctl
xfs_swap_extent_rmap does not use log items to track the overall
progress of an attempt to swap the extent mappings between two files.
If the system crashes in the middle of swapping a partially written
realtime extent, the mapping will be left in an inconsistent state
wherein a file can point to multiple extents on the rt volume.
The new file range exchange functionality handles this correctly, so all
callers must upgrade to that.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:22 +0000 (21:11 -0700)]
xfs: use an incore rtgroup rotor for rtpick
During the 6.7 merge window, Linus noticed that the realtime allocator
was doing some sketchy things trying to encode a u64 sequence counter
into the rtbitmap file's atime. The sketchy casting of a struct pointer
to a u64 pointer has subtly broken several times over the past decade as
the codebase has transitioned to using the VFS i_atime field and that
field has changed in size and layout over time.
Since the goal of the rtpick code is to _suggest_ a starting place for
new rt file allocations, the repeated breakage has not resulted in
inconsistent metadata. IOWs, it's a hint.
For rtgroups, we don't need this complex code to cut the rtextents space
into fractions. Add an rtgroup rotor and use that for rtpick, similar
to AG rotoring on the data device. The new rotor does not persist,
which reduces the logging overhead slightly.
Between this and the new restrictions on open-by-handle on metadir, it's
no longer possible for userspace to control the rtpick rotor.
Darrick J. Wong [Wed, 29 May 2024 04:11:21 +0000 (21:11 -0700)]
xfs: store rtgroup information with a bmap intent
Make the bmap intent items take an active reference to the rtgroup
containing the space that is being mapped or unmapped. We will need
this functionality once we start enabling rmap and reflink on the rt
volume.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:21 +0000 (21:11 -0700)]
xfs: encode the rtsummary in big endian format
Currently, the ondisk realtime summary file counters are accessed in
units of 32-bit words. There's no endian translation of the contents of
this file, which means that the Bad Things Happen(tm) if you go from
(say) x86 to powerpc. Since we have a new feature flag, let's take the
opportunity to enforce an endianness on the file. Encode the summary
information in big endian format, like most of the rest of the
filesystem.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:19 +0000 (21:11 -0700)]
xfs: encode the rtbitmap in big endian format
Currently, the ondisk realtime bitmap file is accessed in units of
32-bit words. There's no endian translation of the contents of this
file, which means that the Bad Things Happen(tm) if you go from (say)
x86 to powerpc. Since we have a new feature flag, let's take the
opportunity to enforce an endianness on the file.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:18 +0000 (21:11 -0700)]
xfs: define locking primitives for realtime groups
Define helper functions to lock all metadata inodes related to a
realtime group. There's not much to look at now, but this will become
important when we add per-rtgroup metadata files and online fsck code
for them.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:17 +0000 (21:11 -0700)]
xfs: add frextents to the lazysbcounters when rtgroups enabled
Make the free rt extent count a part of the lazy sb counters when the
realtime groups feature is enabled. This is possible because the patch
to recompute frextents from the rtbitmap during log recovery predates
the code adding rtgroup support, hence we know that the value will
always be correct during runtime.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Christoph Hellwig [Tue, 3 Oct 2023 08:26:14 +0000 (10:26 +0200)]
xfs: add a helper to prevent bmap merges across rtgroup boundaries
Except for the rt superblock, realtime groups do not store any metadata
at the start (or end) of the group. There is nothing to prevent the
bmap code from merging allocations from multiple groups into a single
bmap record. Add a helper to check for this case.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: massage the commit message after pulling this into rtgroups] Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 26 Jun 2024 18:07:24 +0000 (11:07 -0700)]
xfs: clamp allocations to rtgroup boundaries
For sharding reasons, we want to have realtime allocation groups that
are totally self-contained. Therefore, xfs_bmap_rtalloc should never
return a free space extent that crosses an rtgroup boundary. After
we've found a candidate freespace extent, clamp the end of the candidate
to the end of a rt group.
Also, cut down on the variable reuse in xfs_rtallocate_extent_exact and
avoid an integer division in the hot path of xfs_rtallocate_clamp_len.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 26 Jun 2024 17:56:55 +0000 (10:56 -0700)]
xfs: create a separate helper to validate rt freespace extents
Realtime allocation groups are not like AGs on the data device which
have a superblock to ensure that a space extent cannot ever cross an AG
boundary. To make sharding and fsck easier, we would like for allocated
space extents in the realtime volume never to cross an rtgroup boundary.
Therefore, we need two rtblock/rtextent predicates here. In the next
patch, the xfs_verify_rtbext helper will check that the arguments do not
cross an rtgroup boundary.
However, the rtbitmap/summary files can describe free space extents that
/do/ cross rtgroup boundaries. For them, create a xfs_verify_rt_freesp
helper that doesn't care about rtgroup boundaries.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:13 +0000 (21:11 -0700)]
xfs: update realtime super every time we update the primary fs super
Every time we update parts of the primary filesystem superblock that are
echoed in the rt superblock, we must update the rt super. Avoid
changing the log to support logging to the rt device by using ordered
buffers.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:13 +0000 (21:11 -0700)]
xfs: check the realtime superblock at mount time
Check the realtime superblock at mount time, to ensure that the label
and uuids actually match the primary superblock on the data device. If
the rt superblock is good, attach it to the xfs_mount so that the log
can use ordered buffers to keep this primary in sync with the primary
super on the data device.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:12 +0000 (21:11 -0700)]
xfs: define the format of rt groups
Define the ondisk format of realtime group metadata, and a superblock
for realtime volumes. rt supers are protected by a separate rocompat
bit so that we can leave them off if the rt device is zoned.
Add a xfs_sb_version_hasrtgroups so that xfs_repair knows how to zero
the tail of superblocks.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:11 +0000 (21:11 -0700)]
xfs: create incore realtime group structures
Create an incore object that will contain information about a realtime
allocation group. This will eventually enable us to shard the realtime
section in a similar manner to how we shard the data section.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Christoph Hellwig [Mon, 8 Jul 2024 21:36:01 +0000 (14:36 -0700)]
iomap: add a merge boundary flag
File systems might have boundaries over which merges aren't possible.
In fact these are very common, although most of the time some kind of
header at the beginning of this region (e.g. XFS alloation groups, ext4
block groups) automatically create a merge barrier. But if that is
not present, say for a device purely used for data we need to manually
communicate that to iomap.
Add a IOMAP_F_BOUNDARY flag to never merge I/O into a previous mapping.
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>
Darrick J. Wong [Wed, 29 May 2024 04:11:40 +0000 (21:11 -0700)]
xfs: support error injection when freeing rt extents
A handful of fstests expect to be able to test what happens when extent
free intents fail to actually free the extent. Now that we're
supporting EFIs for realtime extents, add to xfs_rtfree_extent the same
injection point that exists in the regular extent freeing code.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:39 +0000 (21:11 -0700)]
xfs: support logging EFIs for realtime extents
Teach the EFI mechanism how to free realtime extents. We're going to
need this to enforce proper ordering of operations when we enable
realtime rmap.
Declare a new log intent item type (XFS_LI_EFI_RT) and a separate defer
ops for rt extents. This keeps the ondisk artifacts and processing code
completely separate between the rt and non-rt cases. Hopefully this
will make it easier to debug filesystem problems.
Previous versions of this patch accomplished this by setting the high
bit in each rt EFI extent. This was found to be less transparent by
reviewers.
[Contains a bug fix and cleanups from hch]
Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Christoph Hellwig [Mon, 1 Jul 2024 18:57:44 +0000 (11:57 -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>
Christoph Hellwig [Mon, 1 Jul 2024 18:55:19 +0000 (11:55 -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>
Darrick J. Wong [Sat, 29 Jun 2024 16:14:53 +0000 (09:14 -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>
Darrick J. Wong [Sat, 29 Jun 2024 16:01:22 +0000 (09:01 -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>
Darrick J. Wong [Fri, 28 Jun 2024 16:40:47 +0000 (09:40 -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>
Darrick J. Wong [Fri, 28 Jun 2024 16:24:31 +0000 (09:24 -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>
Darrick J. Wong [Fri, 28 Jun 2024 16:08:07 +0000 (09:08 -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>
Darrick J. Wong [Fri, 28 Jun 2024 16:02:36 +0000 (09:02 -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>
Darrick J. Wong [Wed, 29 May 2024 04:11:10 +0000 (21:11 -0700)]
xfs: use separate lock classes for realtime metadata inode ILOCKs
Realtime metadata files are not quite regular files because userspace
can't access the realtime bitmap directly, and because we take the ILOCK
of the rt bitmap file while holding the ILOCK of a realtime file. The
double nature of inodes confuses lockdep, so up until now we've created
lockdep subclasses to help lockdep keep things straight.
We've gotten away with using lockdep subclasses because there's only two
rt metadata files, but with the coming addition of realtime rmap and
refcounting, we'd need two more subclasses, which is a lot of class bits
to burn on a side feature.
Therefore, switch to manually setting the lockdep class of the rt
metadata ILOCKs. In the next patch we'll remove the rt-related ILOCK
subclasses.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Create a pair of helpers to deal with setting up the necessary incore
context to check metadata records against the realtime metadata. Right
now this is limited to locking the realtime bitmap and summary inodes,
but as we add rmap and reflink to the realtime device this will grow to
include btree cursors.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Fri, 14 Jun 2024 21:07:48 +0000 (14:07 -0700)]
xfs: confirm dotdot target before replacing it during a repair
xfs_dir_replace trips an assertion if you tell it to change a dirent to
point to an inumber that it already points at. Look up the dotdot entry
directly to confirm that we need to make a change.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Create a new scrubber type that checks that well known metadata
directory paths are connected to the metadata inode that the incore
structures think is in use. IOWs, check that "/quota/user" in the
metadata directory tree actually points to
mp->m_quotainfo->qi_uquotaip->i_ino.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:06 +0000 (21:11 -0700)]
xfs: move repair temporary files to the metadata directory tree
Due to resource acquisition rules, we have to create the ondisk
temporary files used to stage a filesystem repair before we can acquire
a reference to the inode that we actually want to repair. Therefore,
we do not know at tempfile creation time whether the tempfile will
belong to the regular directory tree or the metadata directory tree.
This distinction becomes important when the swapext code tries to figure
out the quota accounting of the two files whose mappings are being
swapped. The swapext code assumes that accounting updates are required
for a file if dqattach attaches dquots. Metadir files are never
accounted in quota, which means that swapext must not update the quota
accounting when swapping in a repaired directory/xattr/rtbitmap structure.
Prior to the swapext call, therefore, both files must be marked as
METADIR for dqattach so that dqattach will ignore them. Add support for
a repair tempfile to be switched to the metadir tree and switched back
before being released so that ifree will just free the file.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:06 +0000 (21:11 -0700)]
xfs: check the metadata directory inumber in superblocks
When metadata directories are enabled, make sure that the secondary
superblocks point to the metadata directory. This isn't strictly
required because the secondaries are only used to recover damaged
filesystems.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Thu, 27 Jun 2024 20:59:09 +0000 (13:59 -0700)]
xfs: adjust parent pointer scrubber for sb-rooted metadata files
Starting with the metadata directory feature, we're allowed to call the
directory and parent pointer scrubbers for every metadata file,
including the ones that are children of the superblock.
For these children, checking the link count against the number of parent
pointers is a bit funny -- there's no such thing as a parent pointer for
a child of the superblock since there's no corresponding dirent. For
purposes of validating nlink, we pretend that there is a parent pointer.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:04 +0000 (21:11 -0700)]
xfs: metadata files can have xattrs if metadir is enabled
If metadata directory trees are enabled, it's possible that some future
metadata file might want to store information in extended attributes.
Or, if parent pointers are enabled, then children of the metadir tree
need parent pointers. Either way, we start allowing xattr data when
metadir is enabled, so we now need check and repair to examine attr
forks for metadata files on metadir filesystems.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:04 +0000 (21:11 -0700)]
xfs: do not count metadata directory files when doing online quotacheck
Previously, we stated that files in the metadata directory tree are not
counted in the dquot information. Fix the online quotacheck code to
reflect this.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Tue, 25 Jun 2024 18:50:11 +0000 (11:50 -0700)]
xfs: refactor directory tree root predicates
Metadata directory trees make reasoning about the parent of a file more
difficult. Traditionally, user files are children of sb_rootino, and
metadata files are "children" of the superblock. Now, we add a third
possibility -- some metadata files can be children of sb_metadirino, but
the classic ones (rt free space data and quotas) are left alone.
Let's add some helper functions (instead of open-coding the logic
everywhere) to make scrub logic easier to understand.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:02 +0000 (21:11 -0700)]
xfs: adjust xfs_bmap_add_attrfork for metadir
Online repair might use the xfs_bmap_add_attrfork to repair a file in
the metadata directory tree if (say) the metadata file lacks the correct
parent pointers. In that case, it is not correct to check that the file
is dqattached -- metadata files must be not have /any/ dquot attached at
all.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:02 +0000 (21:11 -0700)]
xfs: don't count metadata directory files to quota
Files in the metadata directory tree are internal to the filesystem.
Don't count the inodes or the blocks they use in the root dquot because
users do not need to know about their resource usage. This will also
quiet down complaints about dquot usage not matching du output.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:01 +0000 (21:11 -0700)]
xfs: allow bulkstat to return metadata directories
Allow the V5 bulkstat ioctl to return information about metadata
directory files so that xfs_scrub can find and scrub them, since they
are otherwise ordinary directories.
(Metadata files of course require per-file scrub code and hence do not
need exposure.)
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:11:00 +0000 (21:11 -0700)]
xfs: hide metadata inodes from everyone because they are special
Metadata inodes are private files and therefore cannot be exposed to
userspace. This means no bulkstat, no open-by-handle, no linking them
into the directory tree, and no feeding them to LSMs. As such, we mark
them S_PRIVATE, which stops all that.
While we're at it, put them in a separate lockdep class so that it won't
get confused by "recursive" i_rwsem locking such as what happens when we
write to a rt file and need to allocate from the rt bitmap file. The
static function that we use to do this will be exported in the rtgroups
patchset.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:59 +0000 (21:10 -0700)]
xfs: disable the agi rotor for metadata inodes
Ideally, we'd put all the metadata inodes in one place if we could, so
that the metadata all stay reasonably close together instead of
spreading out over the disk. Furthermore, if the log is internal we'd
probably prefer to keep the metadata near the log. Therefore, disable
AGI rotoring for metadata inode allocations.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:58 +0000 (21:10 -0700)]
xfs: read and write metadata inode directory tree
Plumb in the bits we need to load metadata inodes from a named entry in
a metadir directory, create (or hardlink) inodes into a metadir
directory, create metadir directories, and flag inodes as being metadata
files.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:57 +0000 (21:10 -0700)]
xfs: enforce metadata inode flag
Add checks for the metadata inode flag so that we don't ever leak
metadata inodes out to userspace, and we don't ever try to read a
regular inode as metadata.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:55 +0000 (21:10 -0700)]
xfs: define the on-disk format for the metadir feature
Define the on-disk layout and feature flags for the metadata inode
directory feature. Add a xfs_sb_version_hasmetadir for benefit of
xfs_repair, which needs to know where the new end of the superblock
lies.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:54 +0000 (21:10 -0700)]
xfs: iget for metadata inodes
Create a xfs_imeta_iget function for metadata inodes to ensure that when
we try to iget a metadata file, the inobt thinks a metadata inode is in
use and that the file type matches what we are expecting.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Mon, 1 Jul 2024 16:54:51 +0000 (09:54 -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>
Christoph Hellwig [Mon, 1 Jul 2024 18:39:34 +0000 (11:39 -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>
Darrick J. Wong [Mon, 1 Jul 2024 23:30:24 +0000 (16:30 -0700)]
xfs: attr forks require attr, not attr2
It turns out that I misunderstood the difference between the attr and
attr2 feature bits. "attr" means that at some point an attr fork was
created somewhere in the filesystem. "attr2" means that inodes have
variable-sized forks, but says nothing about whether or not there
actually /are/ attr forks in the system.
If we have an attr fork, we only need to check that attr is set.
Fixes: 99d9d8d05da26 ("xfs: scrub inode block mappings") Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Wed, 29 May 2024 04:10:39 +0000 (21:10 -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.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Darrick J. Wong [Tue, 9 Jul 2024 14:43:42 +0000 (07:43 -0700)]
xfs: fix file_path handling in tracepoints
Since file_path() takes the output buffer as one of its arguments, we
might as well have it format directly into the tracepoint's char array
instead of wasting stack space.
Fixes: 3934e8ebb7cc6 ("xfs: create a big array data structure") Fixes: 5076a6040ca16 ("xfs: support in-memory buffer cache targets") Reported-by: kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202403290419.HPcyvqZu-lkp@intel.com/ Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Christoph Hellwig [Tue, 16 Jul 2024 21:50:25 +0000 (14:50 -0700)]
xfs: fix rtalloc rotoring when delalloc is in use
If we're trying to allocate real space for a delalloc reservation at
offset 0, we should use the rotor to spread files across the rt volume.
Switch the rtalloc to use the XFS_ALLOC_INITIAL_USER_DATA flag that
is set for any write at startoff to make it match the behavior for
the main data device.
Based on a patch from Darrick J. Wong.
Fixes: 6a94b1acda7e ("xfs: reinstate delalloc for RT inodes (if sb_rextsize == 1)") Reported-by: Darrick J. Wong <djwong@kernel.org> 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>