Make the necessary alterations to the extended attribute log intent item
ondisk format so that we can log parent pointer operations. This
requires the creation of new opcodes specific to parent pointers, and a
new four-argument replace operation to handle renames. At this point
this part of the patchset has changed so much from what Allison original
wrote that I no longer think her SoB applies.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
If a file is hardlinked with the same name but from multiple parents,
the parent pointers will all have the same dirent name (== attr name)
but with different parent_ino/parent_gen values. To disambiguate, we
need to be able to match on both the attr name and the attr value. This
is in contrast to regular xattrs, which are matchtg edit
d only on name.
Therefore, plumb in the ability to match shortform and local attrs on
name and value in the XFS_ATTR_PARENT namespace. Parent pointer attr
values are never large enough to be stored in a remote attr, so we need
can reject these cases as corruption.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
We need to define the parent pointer attribute format before we start
adding support for it into all the code that needs to use it. The EA
format we will use encodes the following information:
name={dirent name}
value={parent inumber, parent inode generation}
hash=xfs_dir2_hashname(dirent name) ^ (parent_inumber)
The inode/gen gives all the information we need to reliably identify the
parent without requiring child->parent lock ordering, and allows
userspace to do pathname component level reconstruction without the
kernel ever needing to verify the parent itself as part of ioctl calls.
By using the name-value lookup mode in the extended attribute code to
match parent pointers using both the xattr name and value, we can
identify the exact parent pointer EA we need to modify/remove in
rename/unlink operations without searching the entire EA space.
By storing the dirent name, we have enough information to be able to
validate and reconstruct damaged directory trees. Earlier iterations of
this patchset encoded the directory offset in the parent pointer key,
but this format required repair to keep that in sync across directory
rebuilds, which is unnecessary complexity.
Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Add the new parent attribute type. XFS_ATTR_PARENT is used only for parent pointer
entries; it uses reserved blocks like XFS_ATTR_ROOT.
Signed-off-by: Mark Tinguely <mark.tinguely@oracle.com>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Create a separate function to compute name hashvalues for extended
attributes. When we get to parent pointers we'll be altering the rules
so that metadump obfuscation doesn't turn heinous.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Move the code that adds the incore xfs_attr_item deferred work data to a
transaction live with the ATTRI log item code. This means that the
upper level extended attribute code no longer has to know about the
inner workings of the ATTRI log items.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Checking the flags match is much cheaper than a memcmp, so do it early
on in xfs_attr_match, and also add a little helper to calculate the
match mask right under the comment explaining the logic for 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>
Rearrange the parameters to this function so that they match the order
of attr listent: attr_flags -> name -> namelen -> value -> valuelen.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Create a standardized helper function to enforce one namespace bit per
extended attribute, and refactor all the open-coded hweight logic. This
function is not a static inline to avoid porting hassles in userspace.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Create helper functions to extract the xattr op from the ondisk xattri
log item and the incore attr intent item. These will get more use in
the patches that follow.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Eliminate the local variable from this function so that we can
streamline things a bit later when we add the PPTR_REPLACE op code.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The xattr scrubber doesn't check for undefined flags in shortform attr
entries. Therefore, define a mask XFS_ATTR_ONDISK_MASK that has all
possible XFS_ATTR_* flags in it, and use that to check for unknown bits
in xchk_xattr_actor.
Refactor the check in the dabtree scanner function to use the new mask
as well. The redundant checks need to be in place because the dabtree
check examines the hash mappings and therefore needs to decode the attr
leaf entries to compute the namehash. This happens before the walk of
the xattr entries themselves.
Fixes: ae0506eba7 ("xfs: check used space of shortform xattr structures")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Christoph noticed that the xfs_attr_is_leaf in xfs_attr_get_ilocked can
access the incore extent tree of the attr fork, but nothing in the
xfs_attr_get path guarantees that the incore tree is actually loaded.
Most of the time it is, but seeing as xfs_attr_is_leaf ignores the
return value of xfs_iext_get_extent I guess we've been making choices
based on random stack contents and nobody's complained?
Reported-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
A few notes about struct xfs_da_args:
The XFS_ATTR_* flags only go up as far as XFS_ATTR_INCOMPLETE, which
means that attr_filter could be a u8 field.
I've reduced the number of XFS_DA_OP_* flags down to the point where
op_flags would also fit into a u8.
filetype has 7 bytes of slack after it, which is wasteful.
namelen will never be greater than MAXNAMELEN, which is 256. This field
could be reduced to a short.
Rearrange the fields in xfs_da_args to waste less space. This reduces
the structure size from 136 bytes to 128. Later when we add extra
fields to support parent pointer replacement, this will only bloat the
structure to 144 bytes, instead of 168.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Parent pointers match attrs on name+value, unlike everything else which
matches on only the name. Therefore, we cannot keep using the heuristic
that !value means remove. Make this an explicit operation code.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
This field only ever contains XATTR_{CREATE,REPLACE}, and it only goes
as deep as xfs_attr_set. Remove the field from the structure and
replace it with an enum specifying exactly what kind of change we want
to make to the xattr structure. Upsert is the name that we'll give to
the flags==0 operation, because we're either updating an existing value
or inserting it, and the caller doesn't care.
Note: The "UPSERTR" name created here is to make userspace porting
easier. It will be removed in the next patch.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The only user of this flag sets it prior to an xfs_attr_get_ilocked
call, which doesn't update anything. Get rid of the flag.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
When xfs_bmap_del_extent_delay has to split an indirect block it tries
to steal blocks from the the part that gets unmapped to increase the
indirect block reservation that now needs to cover for two extents
instead of one.
This works perfectly fine on the data device, where the data and
indirect blocks come from the same pool. It has no chance of working
when the inode sits on the RT device. To support re-enabling delalloc
for inodes on the RT device, make this behavior conditional on not
being for rt extents.
Note that split of delalloc extents should only happen on writeback
failure, as for other kinds of hole punching we first write back all
data and thus convert the delalloc reservations covering the hole to
a real allocation.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Move the check if we have enough indirect blocks and the stealing of
the deleted extent blocks out of xfs_bmap_split_indlen and into the
caller to prepare for handling delayed allocation of RT extents that
can't easily be stolen.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
To prepare for re-enabling delalloc on RT devices, track the data blocks
(which use the RT device when the inode sits on it) and the indirect
blocks (which don't) separately to xfs_mod_delalloc, and add a new
percpu counter to also track the RT delalloc blocks.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
The code to account fdblocks and frextents in xfs_bmap_del_extent_delay
is a bit weird in that it accounts frextents before the iext tree
manipulations and fdblocks after it. Given that the iext tree
manipulations cannot fail currently that's not really a problem, but
still odd. Move the frextent manipulation to the end, and use a
fdblocks variable to account of the unconditional indirect blocks and
the data blocks only freed for !RT. This prepares for following
updates in the area and already makes the code more readable.
Also remove the !isrt assert given that this code clearly handles
rt extents correctly, and we'll soon reinstate delalloc support for
RT inodes.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Allocate data blocks for RT inodes using xfs_dec_frextents. While at
it optimize the data device case by doing only a single xfs_dec_fdblocks
call for the extent itself and the indirect blocks.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
xfs_mod_freecounter has two entirely separate code paths for adding or
subtracting from the free counters. Only the subtract case looks at the
rsvd flag and can return an error.
Split xfs_mod_freecounter into separate helpers for subtracting or
adding the freecounter, and remove all the impossible to reach error
handling for the addition case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
__xfs_bunmapi is a bit of an odd place to lock the rtbitmap and rtsummary
inodes given that it is very high level code. While this only looks ugly
right now, it will become a problem when supporting delayed allocations
for RT inodes as __xfs_bunmapi might end up deleting only delalloc extents
and thus never unlock the rt inodes.
Move the locking into xfs_bmap_del_extent_real just before the call to
xfs_rtfree_blocks instead and use a new flag in the transaction to ensure
that the locking happens only once.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Currently xfs_bmap_del_extent_real frees RT extents before updating
the bmap btree, while it frees regular blocks after performing the bmap
btree update for convoluted historic reasons. Switch to free the RT
blocks in the same place as the regular data blocks instead to simply
the code and fix a very theoretical bug.
A short history of this code researched by Dave Chiner below:
The truncate for data device extents was originally a two-phase
operation. First it removed the bmapbt record, but because this can
free BMBT extents, it can use up all the free space tree reservation
space. So the transaction gets rolled to commit the BMBT change and
the xfs_bmap_finish() call that frees the data extent runs with a
new transaction reservation that allows different free space btrees
to be logged without overrun.
However, on crash, this could lose the free space because there was
nothing to tell recovery about the extents removed from the BMBT,
hence EFIs were introduced. They tie the extent free operation to the
bmapbt record removal commit for recovery of the second phase of the
extent removal process.
Then RT extents came along. RT extent freeing does not require a
free space btree reservation because the free space metadata is
static and transaction size is bound. Hence we don't need to care if
the BMBT record removal modifies the per-ag free space trees and we
don't need a two-phase extent remove transaction. The only thing we
have to care about is not losing space on crash.
Hence instead of recording the extent for freeing in the bmap list
for xfs_bmap_finish() to process in a new transaction, it simply
freed the rtextent directly. So the original code (from 1994) simply
replaced the "free AG extent later" queueing with a direct free.
This code was originally at the start of xfs_dmap_del_extent(), but
the xfs_bmap_add_free() got moved to the end of the function via the
"do_fx" flag (the current code logic) in 1997 (commit c4fac74eaa58
in the historic xfs-import tree) because there was a shutdown occurring
because of a case where splitting the extent record failed because the
BMBT split and the filesystem didn't have enough space for the split to
be done. (FWIW, I'm not sure this can happen anymore.)
The commit backed out the BMBT change on ENOSPC error, and in doing
so I think this actually breaks RT free space tracking. However, it
then returns an ENOSPC error, and we have a dirty transaction in the
RT case so this will shut down the filesysetm when the transaction
is cancelled. Hence the corrupted "bmbt now points at freed rt dev
space" condition never make it to disk, but it's still the wrong way
to handle the issue.
IOWs, this proposed change fixes that "shutdown at ENOSPC on rt
devices" situation that was introduced by the above commit back in
1997.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Create helper functions to deal with locking realtime metadata inodes.
This enables us to maintain correct locking order once we start adding
the realtime rmap and refcount btree inodes.
Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Commit bb7b1c9c5d ("xfs: tag transactions that contain intent done
items") switched the XFS_TRANS_ definitions to be bit based, and using
comments above the definitions. As XFS_TRANS_LOWMODE was last and has
a big fat comment it was missed. Switch it to the same style.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>
Renames that generate parent pointer updates can join up to 5
inodes locked in sorted order. So we need to increase the
number of defer ops inodes and relock them in the same way.
Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Catherine Hoang <catherine.hoang@oracle.com>
[djwong: have one sorting function]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The VFS inc_nlink function does not explicitly check for integer
overflows in the i_nlink field. Instead, it checks the link count
against s_max_links in the vfs_{link,create,rename} functions. XFS
sets the maximum link count to 2.1 billion, so integer overflows should
not be a problem.
However. It's possible that online repair could find that a file has
more than four billion links, particularly if the link count got
corrupted while creating hardlinks to the file. The di_nlinkv2 field is
not large enough to store a value larger than 2^32, so we ought to
define a magic pin value of ~0U which means that the inode never gets
deleted. This will prevent a UAF error if the repair finds this
situation and users begin deleting links to the file.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
I noticed that xfs/413 and xfs/375 occasionally failed while fuzzing
core.mode of an inode. The root cause of these problems is that the
field we fuzzed (core.mode or core.magic, typically) causes the entire
inode cluster buffer verification to fail, which affects several inodes
at once. The repair process tries to create either a /lost+found or a
temporary repair file, but regrettably it picks the same inode cluster
that we just corrupted, with the result that repair triggers the demise
of the filesystem.
Try avoid this by making the inode allocation path detect when the perag
health status indicates that someone has found bad inode cluster
buffers, and try to read the inode cluster buffer. If the cluster
buffer fails the verifiers, try another AG. This isn't foolproof and
can result in premature ENOSPC, but that might be better than shutting
down.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
v2/v3 inodes use di_nlink and not di_onlink; and v1 inodes use di_onlink
and not di_nlink. Whichever field is not in use, make sure its contents
are zero, and teach xfs_scrub to fix that if it is.
This clears a bunch of missing scrub failure errors in xfs/385 for
core.onlink.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Require callers of xfs_symlink_write_target to pass the owner number
explicitly. This sets us up for online repair to be able to write a
remote symlink target to sc->tempip with sc->ip's inumber in the block
heaader.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Allow online repair to call xfs_bmap_local_to_extents and add a void *
argument at the end so that online repair can pass its own context.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
If the extended attributes look bad, try to sift through the rubble to
find whatever keys/values we can, stage a new attribute structure in a
temporary file and use the atomic extent swapping mechanism to commit
the results in bulk.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Build on the code that was recently added to the temporary repair file
code so that we can atomically switch the contents of any file fork,
even if the fork is in local format. The upcoming functions to repair
xattrs, directories, and symlinks will need that capability.
Repair can lock out access to these user files by holding IOLOCK_EXCL on
these user files. Therefore, it is safe to drop the ILOCK of both the
file being repaired and the tempfile being used for staging, and cancel
the scrub transaction. We do this so that we can reuse the resource
estimation and transaction allocation functions used by a regular file
exchange operation.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Port the existing directory freespace block header checking function to
accept an owner number instead of an xfs_inode, then update the
callsites to use xfs_da_args.owner when possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Port the existing directory block header checking function to accept an
owner number instead of an xfs_inode, then update the callsites to use
xfs_da_args.owner when possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Port the existing directory data header checking function to accept an
owner number instead of an xfs_inode, then update the callsites to use
xfs_da_args.owner when possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Create a leaf block header checking function to validate the owner field
of xattr leaf blocks.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
When we're creating leaf, data, freespace, or dabtree blocks for
directories and xattrs, use the explicit owner field (instead of the
xfs_inode) to set the owner field. This will enable online repair to
construct replacement data structures in a temporary file without having
to change the owner fields prior to swapping the new and old structures.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Add an explicit owner field to xfs_da_args, which will make it easier
for online fsck to set the owner field of the temporary directory and
xattr structures that it builds to repair damaged metadata.
Note: I hopefully found all the xfs_da_args definitions by looking for
automatic stack variable declarations and xfs_da_args.dp assignments:
git grep -E '(args.*dp =|struct xfs_da_args[[:space:]]*[a-z0-9][a-z0-9]*)'
Note that callers of xfs_attr_{get,set,change} can set the owner to zero
(or leave it unset) to have the default set to args->dp.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Add the XFS_SB_FEAT_INCOMPAT_EXCHRANGE feature to the set of features
that we will permit when mounting a filesystem. This turns on support
for the file range exchange feature.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Per some very late review comments, capture the generation numbers of
both inodes involved in a file content exchange operation so that we
don't accidentally target files with have been reallocated.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Now that bmap items support the realtime device, we can add the
necessary pieces to the file range exchange code to support exchanging
mappings. All we really need to do here is adjust the blockcount
upwards to the end of the rt extent and remove the inode checks.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The previous commit added a new file mapping exchange flag that enables
us to perform post-exchange processing on file2 once we're done
exchanging the extent mappings. Now add this ability for symlinks.
This isn't used anywhere right now, but we need to have the basic ondisk
flags in place so that a future online symlink repair feature can
salvage the remote target in a temporary link and exchange the data fork
mappings when ready. If one file is in extents format and the other is
inline, we will have to promote both to extents format to perform the
exchange. After the exchange, we can try to condense the fixed symlink
down to inline format if possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The previous commit added a new file mapping exchange flag that enables
us to perform post-swap processing on file2 once we're done exchanging
extent mappings. Now add this ability for directories.
This isn't used anywhere right now, but we need to have the basic ondisk
flags in place so that a future online directory repair feature can
create salvaged dirents in a temporary directory and exchange the data
fork mappings when ready. If one file is in extents format and the
other is inline, we will have to promote both to extents format to
perform the exchange. After the exchange, we can try to condense the
fixed directory down to inline format if possible.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
