]> www.infradead.org Git - users/hch/xfsprogs.git/log
users/hch/xfsprogs.git
11 months agoxfs_io: support vectored scrub
Darrick J. Wong [Wed, 3 Jul 2024 21:21:29 +0000 (14:21 -0700)]
xfs_io: support vectored scrub

Create a new scrubv command to xfs_io to support the vectored scrub
ioctl.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: support vectored scrub
Darrick J. Wong [Wed, 3 Jul 2024 21:21:29 +0000 (14:21 -0700)]
libfrog: support vectored scrub

Enhance libfrog to support performing vectored metadata scrub.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoman: document vectored scrub mode
Darrick J. Wong [Wed, 3 Jul 2024 21:21:29 +0000 (14:21 -0700)]
man: document vectored scrub mode

Add a manpage to document XFS_IOC_SCRUBV_METADATA.  From the kernel
patch:

Introduce a variant on XFS_SCRUB_METADATA that allows for a vectored
mode.  The caller specifies the principal metadata object that they want
to scrub (allocation group, inode, etc.) once, followed by an array of
scrub types they want called on that object.  The kernel runs the scrub
operations and writes the output flags and errno code to the
corresponding array element.

A new pseudo scrub type BARRIER is introduced to force the kernel to
return to userspace if any corruptions have been found when scrubbing
the previous scrub types in the array.  This enables userspace to
schedule, for example, the sequence:

 1. data fork
 2. barrier
 3. directory

If the data fork scrub is clean, then the kernel will perform the
directory scrub.  If not, the barrier in 2 will exit back to userspace.

The alternative would have been an interface where userspace passes a
pointer to an empty buffer, and the kernel formats that with
xfs_scrub_vecs that tell userspace what it scrubbed and what the outcome
was.  With that the kernel would have to communicate that the buffer
needed to have been at least X size, even though for our cases
XFS_SCRUB_TYPE_NR + 2 would always be enough.

Compared to that, this design keeps all the dependency policy and
ordering logic in userspace where it already resides instead of
duplicating it in the kernel. The downside of that is that it needs the
barrier logic.

When running fstests in "rebuild all metadata after each test" mode, I
observed a 10% reduction in runtime due to fewer transitions across the
system call boundary.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: defer phase5 file scans if dirloop fails
Darrick J. Wong [Wed, 3 Jul 2024 21:21:29 +0000 (14:21 -0700)]
xfs_scrub: defer phase5 file scans if dirloop fails

If we cannot fix dirloop problems during the initial phase 5 inode scan,
defer them until later.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: detect and repair directory tree corruptions
Darrick J. Wong [Wed, 3 Jul 2024 21:21:28 +0000 (14:21 -0700)]
xfs_scrub: detect and repair directory tree corruptions

Now that we have online fsck for directory tree structure problems, we
need to find a place to call it.  The scanner requires that parent
pointers are enabled, that directory link counts are correct, and that
every directory entry has a corresponding parent pointer.  Therefore, we
can only run it after phase 4 fixes every file, and phase 5 resets the
link counts.

In other words, we call it as part of the phase 5 file scan that we do
to warn about weird looking file names.  This has the added benefit that
opening the directory by handle is less likely to fail if there are
loops in the directory structure.  For now, only plumb in enough to try
to fix directory tree problems right away; the next patch will make
phase 5 retry the dirloop scanner until the problems are fixed or we
stop making forward progress.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: fix erroring out of check_inode_names
Darrick J. Wong [Wed, 3 Jul 2024 21:21:28 +0000 (14:21 -0700)]
xfs_scrub: fix erroring out of check_inode_names

The early exit logic in this function is a bit suboptimal -- we don't
need to close the @fd if we haven't even opened it, and since all errors
are fatal, we don't need to bump the progress counter.  The logic in
this function is about to get more involved due to the addition of the
directory tree structure checker, so clean up these warts.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_spaceman: report directory tree corruption in the health information
Darrick J. Wong [Wed, 3 Jul 2024 21:21:28 +0000 (14:21 -0700)]
xfs_spaceman: report directory tree corruption in the health information

Report directories that are the source of corruption in the directory
tree.  While we're at it, add the documentation updates for the new
reporting flags and scrub type.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: add directory tree structure scrubber to scrub library
Darrick J. Wong [Wed, 3 Jul 2024 21:21:28 +0000 (14:21 -0700)]
libfrog: add directory tree structure scrubber to scrub library

Make it so that scrub clients can detect corruptions within the
directory tree structure itself.  Update the documentation for the scrub
ioctl to mention this new functionality.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: wipe ondisk parent pointers when there are none
Darrick J. Wong [Wed, 3 Jul 2024 21:21:27 +0000 (14:21 -0700)]
xfs_repair: wipe ondisk parent pointers when there are none

Erase all the parent pointers when there aren't any found by the
directory entry scan.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: update ondisk parent pointer records
Darrick J. Wong [Wed, 3 Jul 2024 21:21:27 +0000 (14:21 -0700)]
xfs_repair: update ondisk parent pointer records

Update the ondisk parent pointer records as necessary.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: dump garbage parent pointer attributes
Darrick J. Wong [Wed, 3 Jul 2024 21:21:27 +0000 (14:21 -0700)]
xfs_repair: dump garbage parent pointer attributes

Delete xattrs that have ATTR_PARENT set but are so garbage that they
clearly aren't parent pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: check parent pointers
Darrick J. Wong [Wed, 3 Jul 2024 21:21:27 +0000 (14:21 -0700)]
xfs_repair: check parent pointers

Use the parent pointer index that we constructed in the previous patch
to check that each file's parent pointer records exactly match the
directory entries that we recorded while walking directory entries.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: deduplicate strings stored in string blob
Darrick J. Wong [Wed, 3 Jul 2024 21:21:27 +0000 (14:21 -0700)]
xfs_repair: deduplicate strings stored in string blob

Reduce the memory requirements of the string blob structure by
deduplicating the strings stored within.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: move the global dirent name store to a separate object
Darrick J. Wong [Wed, 3 Jul 2024 21:21:26 +0000 (14:21 -0700)]
xfs_repair: move the global dirent name store to a separate object

Abstract the main parent pointer dirent names xfblob object into a
separate data structure to hide implementation details.

The goals here are (a) reduce memory usage when we can by deduplicating
dirent names that exist in multiple directories; and (b) provide a
unique id for each name in the system so that sorting incore parent
pointer records can be done in a stable manner.  Fast stable sorting of
records is required for the dirent <-> pptr matching algorithm.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: build a parent pointer index
Darrick J. Wong [Wed, 3 Jul 2024 21:21:26 +0000 (14:21 -0700)]
xfs_repair: build a parent pointer index

When we're walking directories during phase 6, build an index of parent
pointers that we expect to find.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: junk duplicate hashtab entries when processing sf dirents
Darrick J. Wong [Wed, 3 Jul 2024 21:21:26 +0000 (14:21 -0700)]
xfs_repair: junk duplicate hashtab entries when processing sf dirents

dir_hash_add() adds the passed-in dirent to the directory hashtab even
if there's already a duplicate.  Therefore, if we detect a duplicate or
a garbage entry while processing the a shortform directory's entries, we
need to junk the newly added entry, just like we do when processing
directory data blocks.

This will become particularly relevant in the next patch, where we
generate a master index of parent pointers from the non-junked hashtab
entries of each directory that phase6 scans.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: add parent pointers when messing with /lost+found
Darrick J. Wong [Wed, 3 Jul 2024 21:21:26 +0000 (14:21 -0700)]
xfs_repair: add parent pointers when messing with /lost+found

Make sure that the /lost+found gets created with parent pointers, and
that lost children being put in there get new parent pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: junk parent pointer attributes when filesystem doesn't support them
Darrick J. Wong [Wed, 3 Jul 2024 21:21:26 +0000 (14:21 -0700)]
xfs_repair: junk parent pointer attributes when filesystem doesn't support them

Drop a parent pointer xattr if the filesystem doesn't support parent
pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: actually report errors from libxfs_attr_set
Darrick J. Wong [Wed, 3 Jul 2024 21:21:25 +0000 (14:21 -0700)]
xfs_db: actually report errors from libxfs_attr_set

Actually tell the user what went wrong when setting or removing xattrs.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: remove some boilerplate from xfs_attr_set
Darrick J. Wong [Wed, 3 Jul 2024 21:21:25 +0000 (14:21 -0700)]
xfs_db: remove some boilerplate from xfs_attr_set

In preparation for online/offline repair wanting to use xfs_attr_set,
move some of the boilerplate out of this function into the callers.
Repair can initialize the da_args completely, and the userspace flag
handling/twisting goes away once we move it to xfs_attr_change.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoman2: update ioctl_xfs_scrub_metadata.2 for parent pointers
Darrick J. Wong [Wed, 3 Jul 2024 21:21:25 +0000 (14:21 -0700)]
man2: update ioctl_xfs_scrub_metadata.2 for parent pointers

Update the man page for the scrub ioctl to reflect the new scrubbing
abilities when parent pointers are enabled.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs: create a blob array data structure
Darrick J. Wong [Wed, 3 Jul 2024 21:21:25 +0000 (14:21 -0700)]
xfs: create a blob array data structure

Create a simple 'blob array' data structure for storage of arbitrarily
sized metadata objects that will be used to reconstruct metadata.  For
the intended usage (temporarily storing extended attribute names and
values) we only have to support storing objects and retrieving them.
Use the xfile abstraction to store the attribute information in memory
that can be swapped out.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agomkfs: enable formatting with parent pointers
Allison Henderson [Wed, 3 Jul 2024 21:21:24 +0000 (14:21 -0700)]
mkfs: enable formatting with parent pointers

Enable parent pointer support in mkfs via the '-n parent' parameter.

Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: move the no-V4 filesystem check to join the rest]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agomkfs: Add parent pointers during protofile creation
Allison Henderson [Wed, 3 Jul 2024 21:21:24 +0000 (14:21 -0700)]
mkfs: Add parent pointers during protofile creation

Inodes created from protofile parsing will also need to add the
appropriate parent pointers.

Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: use xfs_parent_add from libxfs instead of open-coding xfs_attr_set]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibxfs: create new files with attr forks if necessary
Darrick J. Wong [Wed, 3 Jul 2024 21:21:24 +0000 (14:21 -0700)]
libxfs: create new files with attr forks if necessary

Create new files with attr forks if they're going to have parent
pointers.  In the next patch we'll fix mkfs to use the same parent
creation functions as the kernel, so we're going to need this.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: compute hashes of parent pointers
Darrick J. Wong [Wed, 3 Jul 2024 21:21:24 +0000 (14:21 -0700)]
xfs_db: compute hashes of parent pointers

Enhance the hash command to compute the hashes of parent pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: add link and unlink expert commands
Darrick J. Wong [Wed, 3 Jul 2024 21:21:24 +0000 (14:21 -0700)]
xfs_db: add link and unlink expert commands

Create a pair of commands to create and remove directory entries to
support functional testing of directory tree corruption.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: make attr_set and attr_remove handle parent pointers
Darrick J. Wong [Wed, 3 Jul 2024 21:21:23 +0000 (14:21 -0700)]
xfs_db: make attr_set and attr_remove handle parent pointers

Make it so that xfs_db can load up the filesystem (somewhat uselessly)
with parent pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: add a parents command to list the parents of a file
Darrick J. Wong [Wed, 3 Jul 2024 21:21:23 +0000 (14:21 -0700)]
xfs_db: add a parents command to list the parents of a file

Create a command to dump the parents of a file.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibxfs: export attr3_leaf_hdr_from_disk via libxfs_api_defs.h
Darrick J. Wong [Wed, 3 Jul 2024 21:21:23 +0000 (14:21 -0700)]
libxfs: export attr3_leaf_hdr_from_disk via libxfs_api_defs.h

Do the xfs -> libxfs switcheroo and cleanups separately so the next
patch doesn't become an even larger mess.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: obfuscate dirent and parent pointer names consistently
Darrick J. Wong [Wed, 3 Jul 2024 21:21:23 +0000 (14:21 -0700)]
xfs_db: obfuscate dirent and parent pointer names consistently

When someone wants to perform an obfuscated metadump of a filesystem
where parent pointers are enabled, we have to use the *exact* same
obfuscated name for both the directory entry and the parent pointer.

Create a name remapping table so that when we obfuscate a dirent name or
a parent pointer name, we can apply the same obfuscation when we find
the corresponding parent pointer or dirent.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: report parent pointers embedded in xattrs
Darrick J. Wong [Wed, 3 Jul 2024 21:21:22 +0000 (14:21 -0700)]
xfs_db: report parent pointers embedded in xattrs

Decode the parent pointer inode, generation, and name fields if the
parent pointer passes basic validation checks.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: report parent bit on xattrs
Darrick J. Wong [Wed, 3 Jul 2024 21:21:22 +0000 (14:21 -0700)]
xfs_db: report parent bit on xattrs

Display the parent bit on xattr keys

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_db: report parent pointers in version command
Darrick J. Wong [Wed, 3 Jul 2024 21:21:22 +0000 (14:21 -0700)]
xfs_db: report parent pointers in version command

Report the presents of PARENT pointers from the version subcommand.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: use parent pointers to report lost file data
Darrick J. Wong [Wed, 3 Jul 2024 21:21:22 +0000 (14:21 -0700)]
xfs_scrub: use parent pointers to report lost file data

If parent pointers are enabled, compute the path to the file while we're
doing the fsmap scan and report that, instead of walking the entire
directory tree to print the paths of the (hopefully few) files that lost
data.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: use parent pointers when possible to report file operations
Darrick J. Wong [Wed, 3 Jul 2024 21:21:22 +0000 (14:21 -0700)]
xfs_scrub: use parent pointers when possible to report file operations

If parent pointers are available, use them to supply file paths when
doing things to files, instead of merely printing the inode number.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_spaceman: report file paths
Darrick J. Wong [Wed, 3 Jul 2024 21:21:21 +0000 (14:21 -0700)]
xfs_spaceman: report file paths

Teach the health command to report file paths when possible.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_logprint: decode parent pointers in ATTRI items fully
Allison Henderson [Wed, 3 Jul 2024 21:21:21 +0000 (14:21 -0700)]
xfs_logprint: decode parent pointers in ATTRI items fully

This patch modifies the ATTRI print routines to look for the parent
pointer flag, and decode logged parent pointers fully when dumping log
contents.  Between the existing ATTRI: printouts and the new ones
introduced here, we can figure out what was stored in each log iovec,
as well as the higher level parent pointer that was logged.

Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: adjust to new ondisk format]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_io: Add i, n and f flags to parent command
Allison Henderson [Wed, 3 Jul 2024 21:21:21 +0000 (14:21 -0700)]
xfs_io: Add i, n and f flags to parent command

This patch adds the flags i, n, and f to the parent command. These flags add
filtering options that are used by the new parent pointer tests in xfstests, and
help to improve the test run time.  The flags are:

-i: Only show parent pointer records containing the given inode
-n: Only show parent pointer records containing the given filename
-f: Print records in short format: ino/gen/namelen/name

Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
[djwong: adapt to new getparents ioctl]
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_io: adapt parent command to new parent pointer ioctls
Darrick J. Wong [Wed, 3 Jul 2024 21:21:21 +0000 (14:21 -0700)]
xfs_io: adapt parent command to new parent pointer ioctls

For ages, xfs_io has had a totally useless 'parent' command that enabled
callers to walk the parents or print the directory tree path of an open
file.  This code used the ioctl interface presented by SGI's version of
parent pointers that was never merged.  Rework the code in here to use
the new ioctl interfaces that we've settled upon.  Get rid of the old
parent pointer checking code since xfs_repair/xfs_scrub will take care
of that.

(This originally was in the "xfsprogs: implement the upper half of
parent pointers" megapatch.)

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: add parent pointer support code
Darrick J. Wong [Wed, 3 Jul 2024 21:21:20 +0000 (14:21 -0700)]
libfrog: add parent pointer support code

Add some support code to libfrog so that client programs can walk file
descriptors and handles upwards through the directory tree; and obtain a
reasonable file path from a file descriptor/handle.  This code will be
used in xfsprogs utilities.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: report parent pointers to userspace
Darrick J. Wong [Wed, 3 Jul 2024 21:21:20 +0000 (14:21 -0700)]
libfrog: report parent pointers to userspace

Report the presence of parent pointer to userspace.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoman: document the XFS_IOC_GETPARENTS ioctl
Darrick J. Wong [Wed, 3 Jul 2024 21:21:20 +0000 (14:21 -0700)]
man: document the XFS_IOC_GETPARENTS ioctl

Document how this new ioctl works.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_logprint: dump new attr log item fields
Darrick J. Wong [Wed, 3 Jul 2024 21:21:20 +0000 (14:21 -0700)]
xfs_logprint: dump new attr log item fields

Dump the new extended attribute log item fields.  This was split out
from the previous patch to make libxfs resyncing easier.  This code
needs more cleaning, which we'll do in the next few patches before
moving on to the parent pointer code.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_{db,repair}: implement new attr hash value function
Darrick J. Wong [Wed, 3 Jul 2024 21:21:20 +0000 (14:21 -0700)]
xfs_{db,repair}: implement new attr hash value function

Port existing utilities to use libxfs_attr_hashname instead of calling
libxfs_da_hashname directly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibxfs: create attr log item opcodes and formats for parent pointers
Darrick J. Wong [Wed, 3 Jul 2024 21:21:19 +0000 (14:21 -0700)]
libxfs: create attr log item opcodes and formats for parent pointers

Update xfs_attr_defer_add to use the pptr-specific opcodes if it's
reading or writing parent pointers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: check for unknown flags in attr entries
Darrick J. Wong [Wed, 3 Jul 2024 21:21:19 +0000 (14:21 -0700)]
xfs_repair: check for unknown flags in attr entries

Explicitly check for unknown bits being set in the shortform and leaf
attr entries.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: enforce one namespace bit per extended attribute
Darrick J. Wong [Wed, 3 Jul 2024 21:21:19 +0000 (14:21 -0700)]
xfs_repair: enforce one namespace bit per extended attribute

Enforce that all extended attributes have at most one namespace bit.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_repair: check free space requirements before allowing upgrades
Darrick J. Wong [Wed, 3 Jul 2024 21:21:19 +0000 (14:21 -0700)]
xfs_repair: check free space requirements before allowing upgrades

Currently, the V5 feature upgrades permitted by xfs_repair do not affect
filesystem space usage, so we haven't needed to verify the geometry.

However, this will change once we start to allow the sysadmin to add new
metadata indexes to existing filesystems.  Add all the infrastructure we
need to ensure that there's enough space for metadata space reservations
and per-AG reservations the next time the filesystem will be mounted.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandan.babu@oracle.com>
[david: Recompute transaction reservation values; Exit with error if upgrade fails]
Signed-off-by: Dave Chinner <david@fromorbit.com>
[djwong: Refuse to upgrade if any part of the fs has < 10% free]
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: add an optimization-only mode
Darrick J. Wong [Wed, 3 Jul 2024 21:21:18 +0000 (14:21 -0700)]
xfs_scrub: add an optimization-only mode

Add a "preen" mode in which we only optimize filesystem metadata.
Repairs will result in early exits.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: automatic downgrades to dry-run mode in service mode
Darrick J. Wong [Wed, 3 Jul 2024 21:21:18 +0000 (14:21 -0700)]
xfs_scrub: automatic downgrades to dry-run mode in service mode

When service mode is enabled, xfs_scrub is being run within the context
of a systemd service.  The service description language doesn't have any
particularly good constructs for adding in a '-n' argument if the
filesystem is readonly, which means that xfs_scrub is passed a path, and
needs to switch to dry-run mode on its own if the fs is mounted
readonly or the kernel doesn't support repairs.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: fail fast on masked units
Darrick J. Wong [Tue, 23 Jul 2024 23:27:45 +0000 (16:27 -0700)]
xfs_scrub_all: fail fast on masked units

If xfs_scrub_all tries to start a masked xfs_scrub@ unit, that's a sign
that the system administrator really didn't want us to scrub that
filesystem.  Instead of retrying pointlessly, just make a note of the
failure and move on.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: implement retry and backoff for dbus calls
Darrick J. Wong [Wed, 3 Jul 2024 21:21:18 +0000 (14:21 -0700)]
xfs_scrub_all: implement retry and backoff for dbus calls

Calls to systemd across dbus are remote procedure calls, which means
that they're subject to transitory connection failures (e.g. systemd
re-exec itself).  We don't want to fail at the *first* sign of what
could be temporary trouble, so implement a limited retry with fibonacci
backoff before we resort to invoking xfs_scrub as a subprocess.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: convert systemctl calls to dbus
Darrick J. Wong [Wed, 3 Jul 2024 21:21:18 +0000 (14:21 -0700)]
xfs_scrub_all: convert systemctl calls to dbus

Convert the systemctl invocations to direct dbus calls, which decouples
us from the CLI in favor of direct API calls.  This spares us from some
of the insanity of divining service state from program outputs.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: add CLI option for easier debugging
Darrick J. Wong [Wed, 3 Jul 2024 21:21:17 +0000 (14:21 -0700)]
xfs_scrub_all: add CLI option for easier debugging

Add a new CLI argument to make it easier to figure out what exactly the
program is doing.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: encapsulate all the systemctl code in an object
Darrick J. Wong [Wed, 3 Jul 2024 21:21:17 +0000 (14:21 -0700)]
xfs_scrub_all: encapsulate all the systemctl code in an object

Move all the systemd service handling code to an object so that we can
contain all the insanity^Wdetails in a single place.  This also makes
the killfuncs handling similar to starting background processes.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: encapsulate all the subprocess code in an object
Darrick J. Wong [Wed, 3 Jul 2024 21:21:17 +0000 (14:21 -0700)]
xfs_scrub_all: encapsulate all the subprocess code in an object

Move all the xfs_scrub subprocess handling code to an object so that we
can contain all the details in a single place.  This also simplifies the
background state management.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: failure reporting for the xfs_scrub_all job
Darrick J. Wong [Wed, 3 Jul 2024 21:21:17 +0000 (14:21 -0700)]
xfs_scrub_all: failure reporting for the xfs_scrub_all job

Create a failure reporting service for when xfs_scrub_all fails.  This
shouldn't happen often, but let's report anyways.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: trigger automatic media scans once per month
Darrick J. Wong [Wed, 3 Jul 2024 21:21:16 +0000 (14:21 -0700)]
xfs_scrub_all: trigger automatic media scans once per month

Teach the xfs_scrub_all background service to trigger an automatic scan
of all file data once per month.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: enable periodic file data scrubs automatically
Darrick J. Wong [Wed, 3 Jul 2024 21:21:16 +0000 (14:21 -0700)]
xfs_scrub_all: enable periodic file data scrubs automatically

Enhance xfs_scrub_all with the ability to initiate a file data scrub
periodically.  The user must specify the period, and they may optionally
specify the path to a file that will record the last time the file data
was scrubbed.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: support metadata+media scans of all filesystems
Darrick J. Wong [Wed, 3 Jul 2024 21:21:16 +0000 (14:21 -0700)]
xfs_scrub_all: support metadata+media scans of all filesystems

Add the necessary systemd services and control bits so that
xfs_scrub_all can kick off a metadata+media scan of a filesystem.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: remove journalctl background process
Darrick J. Wong [Wed, 3 Jul 2024 21:21:16 +0000 (14:21 -0700)]
xfs_scrub_all: remove journalctl background process

Now that we only start systemd services if we're running in service
mode, there's no need for the background journalctl process that only
ran if we had started systemd services in non-service mode.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: only use the xfs_scrub@ systemd services in service mode
Darrick J. Wong [Wed, 3 Jul 2024 21:21:16 +0000 (14:21 -0700)]
xfs_scrub_all: only use the xfs_scrub@ systemd services in service mode

Since the per-mount xfs_scrub@.service definition includes a bunch of
resource usage constraints, we no longer want to use those services if
xfs_scrub_all is being run directly by the sysadmin (aka not in service
mode) on the presumption that sysadmins want answers as quickly as
possible.

Therefore, only try to call the systemd service from xfs_scrub_all if
SERVICE_MODE is set in the environment.  If reaching out to systemd
fails and we're in service mode, we still want to run xfs_scrub
directly.  Split the makefile variables as necessary so that we only
pass -b to xfs_scrub in service mode.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_all: tighten up the security on the background systemd service
Darrick J. Wong [Wed, 3 Jul 2024 21:21:15 +0000 (14:21 -0700)]
xfs_scrub_all: tighten up the security on the background systemd service

Currently, xfs_scrub_all has to run with enough privileges to find
mounted XFS filesystems and the device associated with that mount and to
start xfs_scrub@<mountpoint> sub-services.  Minimize the risk of
xfs_scrub_all escaping its service container or contaminating the rest
of the system by using systemd's sandboxing controls to prohibit as much
access as possible.

The directives added by this patch were recommended by the command
'systemd-analyze security xfs_scrub_all.service' in systemd 249.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub_fail: tighten up the security on the background systemd service
Darrick J. Wong [Wed, 3 Jul 2024 21:21:15 +0000 (14:21 -0700)]
xfs_scrub_fail: tighten up the security on the background systemd service

Currently, xfs_scrub_fail has to run with enough privileges to access
the journal contents for a given scrub run and to send a report via
email.  Minimize the risk of xfs_scrub_fail escaping its service
container or contaminating the rest of the system by using systemd's
sandboxing controls to prohibit as much access as possible.

The directives added by this patch were recommended by the command
'systemd-analyze security xfs_scrub_fail@.service' in systemd 249.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: tighten up the security on the background systemd service
Darrick J. Wong [Wed, 3 Jul 2024 21:21:15 +0000 (14:21 -0700)]
xfs_scrub: tighten up the security on the background systemd service

Currently, xfs_scrub has to run with some elevated privileges.  Minimize
the risk of xfs_scrub escaping its service container or contaminating
the rest of the system by using systemd's sandboxing controls to
prohibit as much access as possible.

The directives added by this patch were recommended by the command
'systemd-analyze security xfs_scrub@.service' in systemd 249.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: use dynamic users when running as a systemd service
Darrick J. Wong [Wed, 3 Jul 2024 21:21:15 +0000 (14:21 -0700)]
xfs_scrub: use dynamic users when running as a systemd service

Five years ago, systemd introduced the DynamicUser directive that
allocates a new unique user/group id, runs a service with those ids, and
deletes them after the service exits.  This is a good replacement for
User=nobody, since it eliminates the threat of nobody-services messing
with each other.

Make this transition ahead of all the other security tightenings that
will land in the next few patches, and add credits for the people who
suggested the change and reviewed it.

Link: https://0pointer.net/blog/dynamic-users-with-systemd.html
Suggested-by: Helle Vaanzinn <glitsj16@riseup.net>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
11 months agoxfs_scrub.service: reduce background CPU usage to less than one core if possible
Darrick J. Wong [Wed, 3 Jul 2024 21:21:14 +0000 (14:21 -0700)]
xfs_scrub.service: reduce background CPU usage to less than one core if possible

Currently, the xfs_scrub background service is configured to use -b,
which means that the program runs completely serially.  However, even
using all of one CPU core with idle priority may be enough to cause
thermal throttling and unwanted fan noise on smaller systems (e.g.
laptops) with fast IO systems.

Let's try to avoid this (at least on systemd) by using cgroups to limit
the program's usage to slghtly more than half of one CPU and lowering
the nice priority in the scheduler.  What we /really/ want is to run
steadily on an efficiency core, but there doesn't seem to be a means to
ask the scheduler not to ramp up the CPU frequency for a particular
task.

While we're at it, group the resource limit directives together.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: allow auxiliary pathnames for sandboxing
Darrick J. Wong [Wed, 3 Jul 2024 21:21:14 +0000 (14:21 -0700)]
xfs_scrub: allow auxiliary pathnames for sandboxing

In the next patch, we'll tighten up the security on the xfs_scrub
service so that it can't escape.  However, sandboxing the service
involves making the host filesystem as inaccessible as possible, with
the filesystem to scrub bind mounted onto a known location within the
sandbox.  Hence we need one path for reporting and a new -M argument to
tell scrub what it should actually be trying to open.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: tune fstrim minlen parameter based on free space histograms
Darrick J. Wong [Wed, 3 Jul 2024 21:21:14 +0000 (14:21 -0700)]
xfs_scrub: tune fstrim minlen parameter based on free space histograms

Currently, phase 8 runs very slowly on filesystems with a lot of small
free space extents.  To reduce the amount of time spent on fstrim
activities during phase 8, we want to balance estimated runtime against
completeness of the trim.  In short, the goal is to reduce runtime by
avoiding small trim requests.

At the start of phase 8, a CDF is computed in decreasing order of extent
length from the histogram buckets created during the fsmap scan in phase
7.  A point corresponding to the fstrim percentage target is chosen from
the CDF and mapped back to a histogram bucket, and free space extents
smaller than that amount are ommitted from fstrim.

On my aging /home filesystem, the free space histogram reported by
xfs_spaceman looks like this:

   from      to extents    blocks    pct blkcdf extcdf
      1       1  121953    121953   0.04 100.00 100.00
      2       3  124741    299694   0.09  99.96  81.16
      4       7  113492    593763   0.18  99.87  61.89
      8      15  109215   1179524   0.36  99.69  44.36
     16      31   76972   1695455   0.52  99.33  27.48
     32      63   48655   2219667   0.68  98.82  15.59
     64     127   31398   2876898   0.88  98.14   8.08
    128     255    8014   1447920   0.44  97.27   3.23
    256     511    4142   1501758   0.46  96.82   1.99
    512    1023    2433   1768732   0.54  96.37   1.35
   1024    2047    1795   2648460   0.81  95.83   0.97
   2048    4095    1429   4206103   1.28  95.02   0.69
   4096    8191    1045   6162111   1.88  93.74   0.47
   8192   16383     791   9242745   2.81  91.87   0.31
  16384   32767     473  10883977   3.31  89.06   0.19
  32768   65535     272  12385566   3.77  85.74   0.12
  65536  131071     192  18098739   5.51  81.98   0.07
 131072  262143     108  20675199   6.29  76.47   0.04
 262144  524287      80  29061285   8.84  70.18   0.03
 524288 1048575      39  29002829   8.83  61.33   0.02
1048576 2097151      25  36824985  11.21  52.51   0.01
2097152 4194303      32 101727192  30.95  41.30   0.01
4194304 8388607       7  34007410  10.35  10.35   0.00

From this table, we see that free space extents that are 16 blocks or
longer constitute 99.3% of the free space in the filesystem but only
27.5% of the extents.  If we set the fstrim minlen parameter to 16
blocks, that means that we can trim over 99% of the space in one third
of the time it would take to trim everything.

Add a new -o fstrim_pct= option to xfs_scrub just in case there are
users out there who want a different percentage.  For example, accepting
a 95% trim would net us a speed increase of nearly two orders of
magnitude, ignoring system call overhead.  Setting it to 100% will trim
everything, just like fstrim(8).

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: collect free space histograms during phase 7
Darrick J. Wong [Wed, 3 Jul 2024 21:21:14 +0000 (14:21 -0700)]
xfs_scrub: collect free space histograms during phase 7

Collect a histogram of free space observed during phase 7.  We'll put
this information to use in the next patch.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: remove pointless spacemap.c arguments
Darrick J. Wong [Wed, 3 Jul 2024 21:21:14 +0000 (14:21 -0700)]
xfs_scrub: remove pointless spacemap.c arguments

Remove unused parameters from the full-device spacemap scan functions.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: don't close stdout when closing the progress bar
Darrick J. Wong [Wed, 3 Jul 2024 21:21:13 +0000 (14:21 -0700)]
xfs_scrub: don't close stdout when closing the progress bar

When we're tearing down the progress bar file stream, check that it's
not an alias of stdout before closing it.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: print cdf of free space buckets
Darrick J. Wong [Wed, 3 Jul 2024 21:21:13 +0000 (14:21 -0700)]
libfrog: print cdf of free space buckets

Print the cumulative distribution function of the free space buckets in
reverse order.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: print wider columns for free space histogram
Darrick J. Wong [Wed, 3 Jul 2024 21:21:13 +0000 (14:21 -0700)]
libfrog: print wider columns for free space histogram

The values reported here can reach very large values, so compute the
column width dynamically.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: hoist free space histogram code
Darrick J. Wong [Wed, 3 Jul 2024 21:21:13 +0000 (14:21 -0700)]
libfrog: hoist free space histogram code

Combine the two free space histograms in xfs_db and xfs_spaceman into a
single implementation.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: improve responsiveness while trimming the filesystem
Darrick J. Wong [Wed, 3 Jul 2024 21:21:12 +0000 (14:21 -0700)]
xfs_scrub: improve responsiveness while trimming the filesystem

On a 10TB filesystem where the free space in each AG is heavily
fragmented, I noticed some very high runtimes on a FITRIM call for the
entire filesystem.  xfs_scrub likes to report progress information on
each phase of the scrub, which means that a strace for the entire
filesystem:

ioctl(3, FITRIM, {start=0x0, len=10995116277760, minlen=0}) = 0 <686.209839>

shows that scrub is uncommunicative for the entire duration.  We can't
report any progress for the duration of the call, and the program is not
responsive to signals.  Reducing the size of the FITRIM requests to a
single AG at a time produces lower times for each individual call, but
even this isn't quite acceptable, because the time between progress
reports are still very high:

Strace for the first 4x 1TB AGs looks like (2):
ioctl(3, FITRIM, {start=0x0, len=1099511627776, minlen=0}) = 0 <68.352033>
ioctl(3, FITRIM, {start=0x10000000000, len=1099511627776, minlen=0}) = 0 <68.760323>
ioctl(3, FITRIM, {start=0x20000000000, len=1099511627776, minlen=0}) = 0 <67.235226>
ioctl(3, FITRIM, {start=0x30000000000, len=1099511627776, minlen=0}) = 0 <69.465744>

I then had the idea to limit the length parameter of each call to a
smallish amount (~11GB) so that we could report progress relatively
quickly, but much to my surprise, each FITRIM call still took ~68
seconds!

Unfortunately, the by-length fstrim implementation handles this poorly
because it walks the entire free space by length index (cntbt), which is
a very inefficient way to walk a subset of an AG when the free space is
fragmented.

To fix that, I created a second implementation in the kernel that will
walk the bnobt and perform the trims in block number order.  This
algorithm constrains the amount of btree scanning to something
resembling the range passed in, which reduces the amount of time it
takes to respond to a signal.

Therefore, break up the FITRIM calls so they don't scan more than 11GB
of space at a time.  Break the calls up by AG so that each call only has
to take one AGF per call, because each AG that we traverse causes a log
force.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: don't call FITRIM after runtime errors
Darrick J. Wong [Wed, 3 Jul 2024 21:21:12 +0000 (14:21 -0700)]
xfs_scrub: don't call FITRIM after runtime errors

Don't call FITRIM if there have been runtime errors -- we don't want to
touch anything after any kind of unfixable problem.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: report FITRIM errors properly
Darrick J. Wong [Wed, 3 Jul 2024 21:21:12 +0000 (14:21 -0700)]
xfs_scrub: report FITRIM errors properly

Move the error reporting for the FITRIM ioctl out of vfs.c and into
phase8.c.  This makes it so that IO errors encountered during trim are
counted as runtime errors instead of being dropped silently.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: fix the work estimation for phase 8
Darrick J. Wong [Wed, 3 Jul 2024 21:21:12 +0000 (14:21 -0700)]
xfs_scrub: fix the work estimation for phase 8

If there are latent errors on the filesystem, we aren't going to do any
work during phase 8 and it makes no sense to add that into the work
estimate for the progress bar.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: collapse trim_filesystem
Darrick J. Wong [Wed, 3 Jul 2024 21:21:12 +0000 (14:21 -0700)]
xfs_scrub: collapse trim_filesystem

Collapse this two-line helper into the main function since it's trivial.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: ignore phase 8 if the user disabled fstrim
Darrick J. Wong [Wed, 3 Jul 2024 21:21:11 +0000 (14:21 -0700)]
xfs_scrub: ignore phase 8 if the user disabled fstrim

If the user told us to skip trimming the filesystem, don't run the phase
at all.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: move FITRIM to phase 8
Darrick J. Wong [Wed, 3 Jul 2024 21:21:11 +0000 (14:21 -0700)]
xfs_scrub: move FITRIM to phase 8

Issuing discards against the filesystem should be the *last* thing that
xfs_scrub does, after everything else has been checked, repaired, and
found to be clean.  If we can't satisfy all those conditions, we have no
business telling the storage to discard itself.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: dump unicode points
Darrick J. Wong [Wed, 3 Jul 2024 21:21:11 +0000 (14:21 -0700)]
xfs_scrub: dump unicode points

Add some debug functions to make it easier to query unicode character
properties.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: report deceptive file extensions
Darrick J. Wong [Wed, 3 Jul 2024 21:21:11 +0000 (14:21 -0700)]
xfs_scrub: report deceptive file extensions

Earlier this year, ESET revealed that Linux users had been tricked into
opening executables containing malware payloads.  The trickery came in
the form of a malicious zip file containing a filename with the string
"job offer․pdf".  Note that the filename does *not* denote a real pdf
file, since the last four codepoints in the file name are "ONE DOT
LEADER", p, d, and f.  Not period (ok, FULL STOP), p, d, f like you'd
normally expect.

Teach xfs_scrub to look for codepoints that could be confused with a
period followed by alphanumerics.

Link: https://www.welivesecurity.com/2023/04/20/linux-malware-strengthens-links-lazarus-3cx-supply-chain-attack/
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: rename struct unicrash.normalizer
Darrick J. Wong [Wed, 3 Jul 2024 21:21:10 +0000 (14:21 -0700)]
xfs_scrub: rename struct unicrash.normalizer

We're about to introduce a second normalizer, so change the name of the
existing one to reflect the algorithm that you'll get if you use it.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: reduce size of struct name_entry
Darrick J. Wong [Wed, 3 Jul 2024 21:21:10 +0000 (14:21 -0700)]
xfs_scrub: reduce size of struct name_entry

libicu doesn't support processing strings longer than 2GB in length, and
we never feed the unicrash code a name longer than about 300 bytes.
Rearrange the structure to reduce the head structure size from 56 bytes
to 44 bytes.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: type-coerce the UNICRASH_* flags
Darrick J. Wong [Wed, 3 Jul 2024 21:21:10 +0000 (14:21 -0700)]
xfs_scrub: type-coerce the UNICRASH_* flags

Promote this type to something that we can type-check.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: rename UNICRASH_ZERO_WIDTH to UNICRASH_INVISIBLE
Darrick J. Wong [Wed, 3 Jul 2024 21:21:10 +0000 (14:21 -0700)]
xfs_scrub: rename UNICRASH_ZERO_WIDTH to UNICRASH_INVISIBLE

"Zero width" doesn't fully describe what the flag represents -- it gets
set for any codepoint that doesn't render.  Rename it accordingly.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: store bad flags with the name entry
Darrick J. Wong [Wed, 3 Jul 2024 21:21:10 +0000 (14:21 -0700)]
xfs_scrub: store bad flags with the name entry

When scrub is checking unicode names, there are certain properties of
the directory/attribute/label name itself that it can complain about.
Store these in struct name_entry so that the confusable names detector
can pick this up later.

This restructuring enables a subsequent patch to detect suspicious
sequences in the NFC normalized form of the name without needing to hang
on to that NFC form until the end of processing.  IOWs, it's a memory
usage optimization.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: hoist non-rendering character predicate
Darrick J. Wong [Wed, 3 Jul 2024 21:21:09 +0000 (14:21 -0700)]
xfs_scrub: hoist non-rendering character predicate

Hoist this predicate code into its own function; we're going to use it
elsewhere later on.  While we're at it, document how we generated this
list in the first place.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: guard against libicu returning negative buffer lengths
Darrick J. Wong [Wed, 3 Jul 2024 21:21:09 +0000 (14:21 -0700)]
xfs_scrub: guard against libicu returning negative buffer lengths

The libicu functions u_strFromUTF8, unorm2_normalize, and
uspoof_getSkeleton return int32_t values.  Guard against negative return
values, even though the library itself never does this.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: avoid potential UAF after freeing a duplicate name entry
Darrick J. Wong [Wed, 3 Jul 2024 21:21:09 +0000 (14:21 -0700)]
xfs_scrub: avoid potential UAF after freeing a duplicate name entry

Change the function declaration of unicrash_add to set the caller's
@new_entry to NULL if we detect an updated name entry and do not wish to
continue processing.  This avoids a theoretical UAF if the unicrash_add
caller were to accidentally continue using the pointer.

This isn't an /actual/ UAF because the function formerly set @badflags
to zero, but let's be a little defensive.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: add a couple of omitted invisible code points
Darrick J. Wong [Wed, 3 Jul 2024 21:21:09 +0000 (14:21 -0700)]
xfs_scrub: add a couple of omitted invisible code points

I missed a few non-rendering code points in the "zero width"
classification code.  Add them now, and sort the list.  Finding them is
an annoyingly manual process because there are various code points that
are not supposed to affect the rendering of a string of text but are not
explicitly named as such.  There are other code points that, when
surrounded by code points from the same chart, actually /do/ affect the
rendering.

IOWs, the only way to figure this out is to grep the likely code points
and then go figure out how each of them render by reading the Unicode
spec or trying it.

$ wget https://www.unicode.org/Public/UCD/latest/ucd/UnicodeData.txt
$ grep -E '(separator|zero width|invisible|joiner|application)' -i UnicodeData.txt

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: hoist code that removes ignorable characters
Darrick J. Wong [Wed, 3 Jul 2024 21:21:08 +0000 (14:21 -0700)]
xfs_scrub: hoist code that removes ignorable characters

Hoist the loop that removes "ignorable" code points from the skeleton
string into a separate function and give the UChar cursors names that
are easier to understand.  Convert the code to use the safe versions of
the U16_ accessor functions.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: use proper UChar string iterators
Darrick J. Wong [Wed, 3 Jul 2024 21:21:08 +0000 (14:21 -0700)]
xfs_scrub: use proper UChar string iterators

For code that wants to examine a UChar string, use libicu's string
iterators to walk UChar strings, instead of the open-coded U16_NEXT*
macros that perform no typechecking.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: try to repair space metadata before file metadata
Darrick J. Wong [Wed, 3 Jul 2024 21:21:08 +0000 (14:21 -0700)]
xfs_scrub: try to repair space metadata before file metadata

Phase 4 (metadata repairs) of xfs_scrub has suffered a mild race
condition since the beginning of its existence.  Repair functions for
higher level metadata such as directories build the new directory blocks
in an unlinked temporary file and use atomic extent swapping to commit
the corrected directory contents into the existing directory.  Atomic
extent swapping requires consistent filesystem space metadata, but phase
4 has never enforced correctness dependencies between space and file
metadata repairs.

Before the previous patch eliminated the per-AG repair lists, this error
was not often hit in testing scenarios because the allocator generally
succeeds in placing file data blocks in the same AG as the inode.  With
pool threads now able to pop file repairs from the repair list before
space repairs complete, this error became much more obvious.

Fortunately, the new phase 4 design makes it easy to try to enforce the
consistency requirements of higher level file metadata repairs.  Split
the repair list into one for space metadata and another for file
metadata.  Phase 4 will now try to fix the space metadata until it stops
making progress on that, and only then will it try to fix file metadata.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: recheck entire metadata objects after corruption repairs
Darrick J. Wong [Wed, 3 Jul 2024 21:21:08 +0000 (14:21 -0700)]
xfs_scrub: recheck entire metadata objects after corruption repairs

When we've finished making repairs to some domain of filesystem metadata
(file, AG, etc.) to correct an inconsistency, we should recheck all the
other metadata types within that domain to make sure that we neither
made things worse nor introduced more cross-referencing problems.  If we
did, requeue the item to make the repairs.  If the only changes we made
were optimizations, don't bother.

The XFS_SCRUB_TYPE_ values are getting close to the max for a u32, so
I chose u64 for sri_selected.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agoxfs_scrub: improve thread scheduling repair items during phase 4
Darrick J. Wong [Wed, 3 Jul 2024 21:21:08 +0000 (14:21 -0700)]
xfs_scrub: improve thread scheduling repair items during phase 4

As it stands, xfs_scrub doesn't do a good job of scheduling repair items
during phase 4.  The repair lists are sharded by AG, and one repair
worker is started for each per-AG repair list.  Consequently, if one AG
requires considerably more work than the others (e.g. inodes are not
spread evenly among the AGs) then phase 4 can stall waiting for that one
worker thread when there's still plenty of CPU power available.

While our initial assumptions were that repairs would be vanishingly
scarce, the reality is that "repairs" can be triggered for optimizations
like gaps in the xattr structures, or clearing the inode reflink flag on
inodes that no longer share data.  In real world testing scenarios, the
lack of balance leads to complaints about excessive runtime of
xfs_scrub.

To fix these balance problems, we replace the per-AG repair item lists
in the scrub context with a single repair item list.  Phase 4 will be
redesigned as follows:

The repair worker will grab a repair item from the main list, try to
repair it, record whether the repair attempt made any progress, and
requeue the item if it was not fully fixed.  A separate repair scheduler
function starts the repair workers, and waits for them all to complete.
Requeued repairs are merged back into the main repair list.  If we made
any forward progress, we'll start another round of repairs with the
repair workers.  Phase 4 retains the behavior that if the pool stops
making forward progress, it will try all the repairs one last time,
serially.

To facilitate this new design, phase 2 will queue repairs of space
metadata items directly to the main list.  Phase 3's worker threads will
queue repair items to per-thread lists and splice those lists into the
main list at the end.

On a filesystem crafted to put all the inodes in a single AG, this
restores xfs_scrub's ability to parallelize repairs.  There seems to be
a slight performance hit for the evenly-spread case, but avoiding a
performance cliff due to an unbalanced fs is more important here.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
11 months agolibfrog: enhance ptvar to support initializer functions
Darrick J. Wong [Wed, 3 Jul 2024 21:21:07 +0000 (14:21 -0700)]
libfrog: enhance ptvar to support initializer functions

Modify the per-thread variable code to support passing in an initializer
function that will set up each thread's variable space when it is
claimed.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>