platform/kernel/linux-rpi.git
3 years agoxfs: shut down the filesystem if we screw up quota reservation
Darrick J. Wong [Fri, 29 Jan 2021 23:44:32 +0000 (15:44 -0800)]
xfs: shut down the filesystem if we screw up quota reservation

If we ever screw up the quota reservations enough to trip the
assertions, something's wrong with the quota code.  Shut down the
filesystem when this happens, because this is corruption.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: rename code to error in xfs_ioctl_setattr
Darrick J. Wong [Sat, 30 Jan 2021 01:14:33 +0000 (17:14 -0800)]
xfs: rename code to error in xfs_ioctl_setattr

Rename the 'code' variable to 'error' to follow the naming convention of
most other functions in xfs.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: remove xfs_qm_vop_chown_reserve
Darrick J. Wong [Mon, 1 Feb 2021 18:38:51 +0000 (10:38 -0800)]
xfs: remove xfs_qm_vop_chown_reserve

Now that the only caller of this function is xfs_trans_alloc_ichange,
just open-code the meat of _chown_reserve in that caller.  Drop the
(redundant) [ugp]id checks because xfs has a 1:1 relationship between
quota ids and incore dquots.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
3 years agoxfs: refactor inode ownership change transaction/inode/quota allocation idiom
Darrick J. Wong [Fri, 29 Jan 2021 19:32:09 +0000 (11:32 -0800)]
xfs: refactor inode ownership change transaction/inode/quota allocation idiom

For file ownership (uid, gid, prid) changes, create a new helper
xfs_trans_alloc_ichange that allocates a transaction and reserves the
appropriate amount of quota against that transction in preparation for a
change of user, group, or project id.  Replace all the open-coded idioms
with a single call to this helper so that we can contain the retry loops
in the next patchset.

This changes the locking behavior for ichange transactions slightly.
Since tr_ichange does not have a permanent reservation and cannot roll,
we pass XFS_ILOCK_EXCL to ijoin so that the inode will be unlocked
automatically at commit time.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: refactor inode creation transaction/inode/quota allocation idiom
Darrick J. Wong [Wed, 27 Jan 2021 20:07:57 +0000 (12:07 -0800)]
xfs: refactor inode creation transaction/inode/quota allocation idiom

For file creation, create a new helper xfs_trans_alloc_icreate that
allocates a transaction and reserves the appropriate amount of quota
against that transction.  Replace all the open-coded idioms with a
single call to this helper so that we can contain the retry loops in the
next patchset.

This changes the locking behavior for non-tempfile creation slightly, in
that we now make the quota reservation without holding the directory
ILOCK.  While the dquots chosen for inode creation are based on the
directory state at a given point in time, the directory ILOCK was
released as soon as the dquot references are picked up.  Hence it was
never necessary to hold the directory ILOCK for the quota reservation.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
3 years agoxfs: refactor reflink functions to use xfs_trans_alloc_inode
Darrick J. Wong [Wed, 27 Jan 2021 18:07:27 +0000 (10:07 -0800)]
xfs: refactor reflink functions to use xfs_trans_alloc_inode

The two remaining callers of xfs_trans_reserve_quota_nblks are in the
reflink code.  These conversions aren't as uniform as the previous
conversions, so call that out in a separate patch.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
3 years agoxfs: allow reservation of rtblocks with xfs_trans_alloc_inode
Darrick J. Wong [Wed, 27 Jan 2021 00:44:07 +0000 (16:44 -0800)]
xfs: allow reservation of rtblocks with xfs_trans_alloc_inode

Make it so that we can reserve rt blocks with the xfs_trans_alloc_inode
wrapper function, then convert a few more callsites.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: refactor common transaction/inode/quota allocation idiom
Darrick J. Wong [Wed, 27 Jan 2021 00:33:29 +0000 (16:33 -0800)]
xfs: refactor common transaction/inode/quota allocation idiom

Create a new helper xfs_trans_alloc_inode that allocates a transaction,
locks and joins an inode to it, and then reserves the appropriate amount
of quota against that transction.  Then replace all the open-coded
idioms with a single call to this helper.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: reserve data and rt quota at the same time
Darrick J. Wong [Wed, 27 Jan 2021 01:20:42 +0000 (17:20 -0800)]
xfs: reserve data and rt quota at the same time

Modify xfs_trans_reserve_quota_nblks so that we can reserve data and
realtime blocks from the dquot at the same time.  This change has the
theoretical side effect that for allocations to realtime files we will
reserve from the dquot both the number of rtblocks being allocated and
the number of bmbt blocks that might be needed to add the mapping.
However, since the mount code disables quota if it finds a realtime
device, this should not result in any behavior changes.

Now that we've moved the inode creation callers away from using the
_nblks function, we can repurpose the (now unused) ninos argument for
realtime blocks, so make that change.  This also replaces the flags
argument with a boolean parameter to force the reservation since we
don't need to distinguish between data and rt quota reservations any
more, and the only flag being passed in was FORCE_RES.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: fix up build warnings when quotas are disabled
Darrick J. Wong [Mon, 25 Jan 2021 19:18:00 +0000 (11:18 -0800)]
xfs: fix up build warnings when quotas are disabled

Fix some build warnings on gcc 10.2 when quotas are disabled.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: clean up icreate quota reservation calls
Darrick J. Wong [Sat, 23 Jan 2021 00:48:34 +0000 (16:48 -0800)]
xfs: clean up icreate quota reservation calls

Create a proper helper so that inode creation calls can reserve quota
with a dedicated function.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
3 years agoxfs: remove xfs_trans_unreserve_quota_nblks completely
Darrick J. Wong [Wed, 27 Jan 2021 01:23:30 +0000 (17:23 -0800)]
xfs: remove xfs_trans_unreserve_quota_nblks completely

xfs_trans_cancel will release all the quota resources that were reserved
on behalf of the transaction, so get rid of the explicit unreserve step.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: create convenience wrappers for incore quota block reservations
Darrick J. Wong [Sat, 23 Jan 2021 00:48:34 +0000 (16:48 -0800)]
xfs: create convenience wrappers for incore quota block reservations

Create a couple of convenience wrappers for creating and deleting quota
block reservations against future changes.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: clean up quota reservation callsites
Darrick J. Wong [Sat, 23 Jan 2021 00:48:33 +0000 (16:48 -0800)]
xfs: clean up quota reservation callsites

Convert a few xfs_trans_*reserve* callsites that are open-coding other
convenience functions.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: reduce quota reservation when doing a dax unwritten extent conversion
Darrick J. Wong [Thu, 28 Jan 2021 18:56:38 +0000 (10:56 -0800)]
xfs: reduce quota reservation when doing a dax unwritten extent conversion

In commit 3b0fe47805802, we reduced the free space requirement to
perform a pre-write unwritten extent conversion on an S_DAX file.  Since
we're not actually allocating any space, the logic goes, we only need
enough reservation to handle shape changes in the bmbt.

The same logic should have been applied to quota -- we're not allocating
any space, so we only need to reserve enough quota to handle the bmbt
shape changes.

Fixes: 3b0fe4780580 ("xfs: Don't use reserved blocks for data blocks with DAX")
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: fix chown leaking delalloc quota blocks when fssetxattr fails
Darrick J. Wong [Sat, 30 Jan 2021 03:06:10 +0000 (19:06 -0800)]
xfs: fix chown leaking delalloc quota blocks when fssetxattr fails

While refactoring the quota code to create a function to allocate inode
change transactions, I noticed that xfs_qm_vop_chown_reserve does more
than just make reservations: it also *modifies* the incore counts
directly to handle the owner id change for the delalloc blocks.

I then observed that the fssetxattr code continues validating input
arguments after making the quota reservation but before dirtying the
transaction.  If the routine decides to error out, it fails to undo the
accounting switch!  This leads to incorrect quota reservation and
failure down the line.

We can fix this by making the reservation function do only that -- for
the new dquot, it reserves ondisk and delalloc blocks to the
transaction, and the old dquot hangs on to its incore reservation for
now.  Once we actually switch the dquots, we can then update the incore
reservations because we've dirtied the transaction and it's too late to
turn back now.

No fixes tag because this has been broken since the start of git.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: Fix 'set but not used' warning in xfs_bmap_compute_alignments()
Chandan Babu R [Wed, 27 Jan 2021 16:35:14 +0000 (08:35 -0800)]
xfs: Fix 'set but not used' warning in xfs_bmap_compute_alignments()

With both CONFIG_XFS_DEBUG and CONFIG_XFS_WARN disabled, the only reference to
local variable "error" in xfs_bmap_compute_alignments() gets eliminated during
pre-processing stage of the compilation process. This causes the compiler to
generate a "set but not used" warning.

Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: fix unused log variable in xfs_log_cover()
Brian Foster [Wed, 27 Jan 2021 03:14:55 +0000 (19:14 -0800)]
xfs: fix unused log variable in xfs_log_cover()

The log variable is only used in kernels with asserts enabled.
Remove it and open code the dereference to avoid unused variable
warnings.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: reduce ilock acquisitions in xfs_file_fsync
Christoph Hellwig [Sat, 23 Jan 2021 00:48:25 +0000 (16:48 -0800)]
xfs: reduce ilock acquisitions in xfs_file_fsync

If the inode is not pinned by the time fsync is called we don't need the
ilock to protect against concurrent clearing of ili_fsync_fields as the
inode won't need a log flush or clearing of these fields.  Not taking
the iolock allows for full concurrency of fsync and thus O_DSYNC
completions with io_uring/aio write submissions.

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>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
3 years agoxfs: refactor xfs_file_fsync
Christoph Hellwig [Sat, 23 Jan 2021 00:48:25 +0000 (16:48 -0800)]
xfs: refactor xfs_file_fsync

Factor out the log syncing logic into two helpers to make the code easier
to read and more maintainable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
3 years agoxfs: cover the log on freeze instead of cleaning it
Brian Foster [Sat, 23 Jan 2021 00:48:24 +0000 (16:48 -0800)]
xfs: cover the log on freeze instead of cleaning it

Filesystem freeze cleans the log and immediately redirties it so log
recovery runs if a crash occurs after the filesystem is frozen. Now
that log quiesce covers the log, there is no need to clean the log and
redirty it to trigger log recovery because covering has the same
effect. Update xfs_fs_freeze() to quiesce (and thus cover) the log.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: remove xfs_quiesce_attr()
Brian Foster [Sat, 23 Jan 2021 00:48:24 +0000 (16:48 -0800)]
xfs: remove xfs_quiesce_attr()

xfs_quiesce_attr() is now a wrapper for xfs_log_clean(). Remove it
and call xfs_log_clean() directly.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: remove duplicate wq cancel and log force from attr quiesce
Brian Foster [Sat, 23 Jan 2021 00:48:23 +0000 (16:48 -0800)]
xfs: remove duplicate wq cancel and log force from attr quiesce

These two calls are repeated at the beginning of xfs_log_quiesce().
Drop them from xfs_quiesce_attr().

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: fold sbcount quiesce logging into log covering
Brian Foster [Sat, 23 Jan 2021 00:48:23 +0000 (16:48 -0800)]
xfs: fold sbcount quiesce logging into log covering

xfs_log_sbcount() calls xfs_sync_sb() to sync superblock counters to
disk when lazy superblock accounting is enabled. This occurs on
unmount, freeze, and read-only (re)mount and ensures the final
values are calculated and persisted to disk before each form of
quiesce completes.

Now that log covering occurs in all of these contexts and uses the
same xfs_sync_sb() mechanism to update log state, there is no need
to log the superblock separately for any reason. Update the log
quiesce path to sync the superblock at least once for any mount
where lazy superblock accounting is enabled. If the log is already
covered, it will remain in the covered state. Otherwise, the next
sync as part of the normal covering sequence will carry the
associated superblock update with it. Remove xfs_log_sbcount() now
that it is no longer needed.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: don't reset log idle state on covering checkpoints
Brian Foster [Sat, 23 Jan 2021 00:48:22 +0000 (16:48 -0800)]
xfs: don't reset log idle state on covering checkpoints

Now that log covering occurs on quiesce, we'd like to reuse the
underlying superblock sync for final superblock updates. This
includes things like lazy superblock counter updates, log feature
incompat bits in the future, etc. One quirk to this approach is that
once the log is in the IDLE (i.e. already covered) state, any
subsequent log write resets the state back to NEED. This means that
a final superblock sync to an already covered log requires two more
sb syncs to return the log back to IDLE again.

For example, if a lazy superblock enabled filesystem is mount cycled
without any modifications, the unmount path syncs the superblock
once and writes an unmount record. With the desired log quiesce
covering behavior, we sync the superblock three times at unmount
time: once for the lazy superblock counter update and twice more to
cover the log. By contrast, if the log is active or only partially
covered at unmount time, a final superblock sync would doubly serve
as the one or two remaining syncs required to cover the log.

This duplicate covering sequence is unnecessary because the
filesystem remains consistent if a crash occurs at any point. The
superblock will either be recovered in the event of a crash or
written back before the log is quiesced and potentially cleaned with
an unmount record.

Update the log covering state machine to remain in the IDLE state if
additional covering checkpoints pass through the log. This
facilitates final superblock updates (such as lazy superblock
counters) via a single sb sync without losing covered status. This
provides some consistency with the active and partially covered
cases and also avoids harmless, but spurious checkpoints when
quiescing the log.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: cover the log during log quiesce
Brian Foster [Sat, 23 Jan 2021 00:48:22 +0000 (16:48 -0800)]
xfs: cover the log during log quiesce

The log quiesce mechanism historically terminates by marking the log
clean with an unmount record. The primary objective is to indicate
that log recovery is no longer required after the quiesce has
flushed all in-core changes and written back filesystem metadata.
While this is perfectly fine, it is somewhat hacky as currently used
in certain contexts. For example, filesystem freeze quiesces (i.e.
cleans) the log and immediately redirties it with a dummy superblock
transaction to ensure that log recovery runs in the event of a
crash.

While this functions correctly, cleaning the log from freeze context
is clearly superfluous given the current redirtying behavior.
Instead, the desired behavior can be achieved by simply covering the
log. This effectively retires all on-disk log items from the active
range of the log by issuing two synchronous and sequential dummy
superblock update transactions that serve to update the on-disk log
head and tail. The subtle difference is that the log technically
remains dirty due to the lack of an unmount record, though recovery
is effectively a no-op due to the content of the checkpoints being
clean (i.e. the unmodified on-disk superblock).

Log covering currently runs in the background and only triggers once
the filesystem and log has idled. The purpose of the background
mechanism is to prevent log recovery from replaying the most
recently logged items long after those items may have been written
back. In the quiesce path, the log has been deliberately idled by
forcing the log and pushing the AIL until empty in a context where
no further mutable filesystem operations are allowed. Therefore, we
can cover the log as the final step in the log quiesce codepath to
reflect that all previously active items have been successfully
written back.

This facilitates selective log covering from certain contexts (i.e.
freeze) that only seek to quiesce, but not necessarily clean the
log. Note that as a side effect of this change, log covering now
occurs when cleaning the log as well. This is harmless, facilitates
subsequent cleanups, and is mostly temporary as various operations
switch to use explicit log covering.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
3 years agoxfs: separate log cleaning from log quiesce
Brian Foster [Sat, 23 Jan 2021 00:48:21 +0000 (16:48 -0800)]
xfs: separate log cleaning from log quiesce

Log quiesce is currently associated with cleaning the log, which is
accomplished by writing an unmount record as the last step of the
quiesce sequence. The quiesce codepath is a bit convoluted in this
regard due to how it is reused from various contexts. In preparation
to create separate log cleaning and log covering interfaces, lift
the write of the unmount record into a new cleaning helper and call
that wherever xfs_log_quiesce() is currently invoked. No functional
changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: lift writable fs check up into log worker task
Brian Foster [Sat, 23 Jan 2021 00:48:21 +0000 (16:48 -0800)]
xfs: lift writable fs check up into log worker task

The log covering helper checks whether the filesystem is writable to
determine whether to cover the log. The helper is currently only
called from the background log worker. In preparation to reuse the
helper from freezing contexts, lift the check into xfs_log_worker().

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: sync lazy sb accounting on quiesce of read-only mounts
Brian Foster [Sat, 23 Jan 2021 00:48:20 +0000 (16:48 -0800)]
xfs: sync lazy sb accounting on quiesce of read-only mounts

xfs_log_sbcount() syncs the superblock specifically to accumulate
the in-core percpu superblock counters and commit them to disk. This
is required to maintain filesystem consistency across quiesce
(freeze, read-only mount/remount) or unmount when lazy superblock
accounting is enabled because individual transactions do not update
the superblock directly.

This mechanism works as expected for writable mounts, but
xfs_log_sbcount() skips the update for read-only mounts. Read-only
mounts otherwise still allow log recovery and write out an unmount
record during log quiesce. If a read-only mount performs log
recovery, it can modify the in-core superblock counters and write an
unmount record when the filesystem unmounts without ever syncing the
in-core counters. This leaves the filesystem with a clean log but in
an inconsistent state with regard to lazy sb counters.

Update xfs_log_sbcount() to use the same logic
xfs_log_unmount_write() uses to determine when to write an unmount
record. This ensures that lazy accounting is always synced before
the log is cleaned. Refactor this logic into a new helper to
distinguish between a writable filesystem and a writable log.
Specifically, the log is writable unless the filesystem is mounted
with the norecovery mount option, the underlying log device is
read-only, or the filesystem is shutdown. Drop the freeze state
check because the update is already allowed during the freezing
process and no context calls this function on an already frozen fs.
Also, retain the shutdown check in xfs_log_unmount_write() to catch
the case where the preceding log force might have triggered a
shutdown.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Gao Xiang <hsiangkao@redhat.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: set inode size after creating symlink
Jeffrey Mitchell [Sat, 23 Jan 2021 00:48:20 +0000 (16:48 -0800)]
xfs: set inode size after creating symlink

When XFS creates a new symlink, it writes its size to disk but not to the
VFS inode. This causes i_size_read() to return 0 for that symlink until
it is re-read from disk, for example when the system is rebooted.

I found this inconsistency while protecting directories with eCryptFS.
The command "stat path/to/symlink/in/ecryptfs" will report "Size: 0" if
the symlink was created after the last reboot on an XFS root.

Call i_size_write() in xfs_symlink()

Signed-off-by: Jeffrey Mitchell <jeffrey.mitchell@starlab.io>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoxfs: don't drain buffer lru on freeze and read-only remount
Brian Foster [Sat, 23 Jan 2021 00:48:20 +0000 (16:48 -0800)]
xfs: don't drain buffer lru on freeze and read-only remount

xfs_buftarg_drain() is called from xfs_log_quiesce() to ensure the
buffer cache is reclaimed during unmount. xfs_log_quiesce() is also
called from xfs_quiesce_attr(), however, which means that cache
state is completely drained for filesystem freeze and read-only
remount. While technically harmless, this is unnecessarily
heavyweight. Both freeze and read-only mounts allow reads and thus
allow population of the buffer cache. Therefore, the transitional
sequence in either case really only needs to quiesce outstanding
writes to return the filesystem in a generally read-only state.

Additionally, some users have reported that attempts to freeze a
filesystem concurrent with a read-heavy workload causes the freeze
process to stall for a significant amount of time. This occurs
because, as mentioned above, the read workload repopulates the
buffer LRU while the freeze task attempts to drain it.

To improve this situation, replace the drain in xfs_log_quiesce()
with a buffer I/O quiesce and lift the drain into the unmount path.
This removes buffer LRU reclaim from freeze and read-only [re]mount,
but ensures the LRU is still drained before the filesystem unmounts.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: rename xfs_wait_buftarg() to xfs_buftarg_drain()
Brian Foster [Sat, 23 Jan 2021 00:48:19 +0000 (16:48 -0800)]
xfs: rename xfs_wait_buftarg() to xfs_buftarg_drain()

xfs_wait_buftarg() is vaguely named and somewhat overloaded. Its
primary purpose is to reclaim all buffers from the provided buffer
target LRU. In preparation to refactor xfs_wait_buftarg() into
serialization and LRU draining components, rename the function and
associated helpers to something more descriptive. This patch has no
functional changes with the minor exception of renaming a
tracepoint.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: Fix assert failure in xfs_setattr_size()
Yumei Huang [Sat, 23 Jan 2021 00:48:19 +0000 (16:48 -0800)]
xfs: Fix assert failure in xfs_setattr_size()

An assert failure is triggered by syzkaller test due to
ATTR_KILL_PRIV is not cleared before xfs_setattr_size.
As ATTR_KILL_PRIV is not checked/used by xfs_setattr_size,
just remove it from the assert.

Signed-off-by: Yumei Huang <yuhuang@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
3 years agoxfs: fix up non-directory creation in SGID directories
Christoph Hellwig [Sat, 23 Jan 2021 00:48:18 +0000 (16:48 -0800)]
xfs: fix up non-directory creation in SGID directories

XFS always inherits the SGID bit if it is set on the parent inode, while
the generic inode_init_owner does not do this in a few cases where it can
create a possible security problem, see commit 0fa3ecd87848
("Fix up non-directory creation in SGID directories") for details.

Switch XFS to use the generic helper for the normal path to fix this,
just keeping the simple field inheritance open coded for the case of the
non-sgid case with the bsdgrpid mount option.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Reported-by: Christian Brauner <christian.brauner@ubuntu.com>
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>
3 years agoxfs: remove a stale comment from xfs_file_aio_write_checks()
Eric Biggers [Sat, 23 Jan 2021 00:48:18 +0000 (16:48 -0800)]
xfs: remove a stale comment from xfs_file_aio_write_checks()

The comment in xfs_file_aio_write_checks() about calling file_modified()
after dropping the ilock doesn't make sense, because the code that
unconditionally acquires and drops the ilock was removed by
commit 467f78992a07 ("xfs: reduce ilock hold times in
xfs_file_aio_write_checks").

Remove this outdated comment.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Introduce error injection to allocate only minlen size extents for files
Chandan Babu R [Sat, 23 Jan 2021 00:48:17 +0000 (16:48 -0800)]
xfs: Introduce error injection to allocate only minlen size extents for files

This commit adds XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag which
helps userspace test programs to get xfs_bmap_btalloc() to always
allocate minlen sized extents.

This is required for test programs which need a guarantee that minlen
extents allocated for a file do not get merged with their existing
neighbours in the inode's BMBT. "Inode fork extent overflow check" for
Directories, Xattrs and extension of realtime inodes need this since the
file offset at which the extents are being allocated cannot be
explicitly controlled from userspace.

One way to use this error tag is to,
1. Consume all of the free space by sequentially writing to a file.
2. Punch alternate blocks of the file. This causes CNTBT to contain
   sufficient number of one block sized extent records.
3. Inject XFS_ERRTAG_BMAP_ALLOC_MINLEN_EXTENT error tag.
After step 3, xfs_bmap_btalloc() will issue space allocation
requests for minlen sized extents only.

ENOSPC error code is returned to userspace when there aren't any "one
block sized" extents left in any of the AGs.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Process allocated extent in a separate function
Chandan Babu R [Sat, 23 Jan 2021 00:48:17 +0000 (16:48 -0800)]
xfs: Process allocated extent in a separate function

This commit moves over the code in xfs_bmap_btalloc() which is
responsible for processing an allocated extent to a new function. Apart
from xfs_bmap_btalloc(), the new function will be invoked by another
function introduced in a future commit.

Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Compute bmap extent alignments in a separate function
Chandan Babu R [Sat, 23 Jan 2021 00:48:16 +0000 (16:48 -0800)]
xfs: Compute bmap extent alignments in a separate function

This commit moves over the code which computes stripe alignment and
extent size hint alignment into a separate function. Apart from
xfs_bmap_btalloc(), the new function will be used by another function
introduced in a future commit.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Remove duplicate assert statement in xfs_bmap_btalloc()
Chandan Babu R [Sat, 23 Jan 2021 00:48:16 +0000 (16:48 -0800)]
xfs: Remove duplicate assert statement in xfs_bmap_btalloc()

The check for verifying if the allocated extent is from an AG whose
index is greater than or equal to that of tp->t_firstblock is already
done a couple of statements earlier in the same function. Hence this
commit removes the redundant assert statement.

Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Introduce error injection to reduce maximum inode fork extent count
Chandan Babu R [Sat, 23 Jan 2021 00:48:15 +0000 (16:48 -0800)]
xfs: Introduce error injection to reduce maximum inode fork extent count

This commit adds XFS_ERRTAG_REDUCE_MAX_IEXTENTS error tag which enables
userspace programs to test "Inode fork extent count overflow detection"
by reducing maximum possible inode fork extent count to 10.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when swapping extents
Chandan Babu R [Sat, 23 Jan 2021 00:48:15 +0000 (16:48 -0800)]
xfs: Check for extent overflow when swapping extents

Removing an initial range of source/donor file's extent and adding a new
extent (from donor/source file) in its place will cause extent count to
increase by 1.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when remapping an extent
Chandan Babu R [Sat, 23 Jan 2021 00:48:14 +0000 (16:48 -0800)]
xfs: Check for extent overflow when remapping an extent

Remapping an extent involves unmapping the existing extent and mapping
in the new extent. When unmapping, an extent containing the entire unmap
range can be split into two extents,
i.e. | Old extent | hole | Old extent |
Hence extent count increases by 1.

Mapping in the new extent into the destination file can increase the
extent count by 1.

Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when moving extent from cow to data fork
Chandan Babu R [Sat, 23 Jan 2021 00:48:14 +0000 (16:48 -0800)]
xfs: Check for extent overflow when moving extent from cow to data fork

Moving an extent to data fork can cause a sub-interval of an existing
extent to be unmapped. This will increase extent count by 1. Mapping in
the new extent can increase the extent count by 1 again i.e.
 | Old extent | New extent | Old extent |
Hence number of extents increases by 2.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when writing to unwritten extent
Chandan Babu R [Sat, 23 Jan 2021 00:48:14 +0000 (16:48 -0800)]
xfs: Check for extent overflow when writing to unwritten extent

A write to a sub-interval of an existing unwritten extent causes
the original extent to be split into 3 extents
i.e. | Unwritten | Real | Unwritten |
Hence extent count can increase by 2.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when adding/removing xattrs
Chandan Babu R [Sat, 23 Jan 2021 00:48:13 +0000 (16:48 -0800)]
xfs: Check for extent overflow when adding/removing xattrs

Adding/removing an xattr can cause XFS_DA_NODE_MAXDEPTH extents to be
added. One extra extent for dabtree in case a local attr is large enough
to cause a double split.  It can also cause extent count to increase
proportional to the size of a remote xattr's value.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when renaming dir entries
Chandan Babu R [Sat, 23 Jan 2021 00:48:13 +0000 (16:48 -0800)]
xfs: Check for extent overflow when renaming dir entries

A rename operation is essentially a directory entry remove operation
from the perspective of parent directory (i.e. src_dp) of rename's
source. Hence the only place where we check for extent count overflow
for src_dp is in xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real()
returns -ENOSPC when it detects a possible extent count overflow and in
response, the higher layers of directory handling code do the following:
1. Data/Free blocks: XFS lets these blocks linger until a future remove
   operation removes them.
2. Dabtree blocks: XFS swaps the blocks with the last block in the Leaf
   space and unmaps the last block.

For target_dp, there are two cases depending on whether the destination
directory entry exists or not.

When destination directory entry does not exist (i.e. target_ip ==
NULL), extent count overflow check is performed only when transaction
has a non-zero sized space reservation associated with it.  With a
zero-sized space reservation, XFS allows a rename operation to continue
only when the directory has sufficient free space in its data/leaf/free
space blocks to hold the new entry.

When destination directory entry exists (i.e. target_ip != NULL), all
we need to do is change the inode number associated with the already
existing entry. Hence there is no need to perform an extent count
overflow check.

Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when removing dir entries
Chandan Babu R [Sat, 23 Jan 2021 00:48:12 +0000 (16:48 -0800)]
xfs: Check for extent overflow when removing dir entries

Directory entry removal must always succeed; Hence XFS does the
following during low disk space scenario:
1. Data/Free blocks linger until a future remove operation.
2. Dabtree blocks would be swapped with the last block in the leaf space
   and then the new last block will be unmapped.

This facility is reused during low inode extent count scenario i.e. this
commit causes xfs_bmap_del_extent_real() to return -ENOSPC error code so
that the above mentioned behaviour is exercised causing no change to the
directory's extent count.

Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when adding dir entries
Chandan Babu R [Sat, 23 Jan 2021 00:48:12 +0000 (16:48 -0800)]
xfs: Check for extent overflow when adding dir entries

Directory entry addition can cause the following,
1. Data block can be added/removed.
   A new extent can cause extent count to increase by 1.
2. Free disk block can be added/removed.
   Same behaviour as described above for Data block.
3. Dabtree blocks.
   XFS_DA_NODE_MAXDEPTH blocks can be added. Each of these
   can be new extents. Hence extent count can increase by
   XFS_DA_NODE_MAXDEPTH.

Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when punching a hole
Chandan Babu R [Sat, 23 Jan 2021 00:48:11 +0000 (16:48 -0800)]
xfs: Check for extent overflow when punching a hole

The extent mapping the file offset at which a hole has to be
inserted will be split into two extents causing extent count to
increase by 1.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Check for extent overflow when trivally adding a new extent
Chandan Babu R [Sat, 23 Jan 2021 00:48:11 +0000 (16:48 -0800)]
xfs: Check for extent overflow when trivally adding a new extent

When adding a new data extent (without modifying an inode's existing
extents) the extent count increases only by 1. This commit checks for
extent count overflow in such cases.

Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: Add helper for checking per-inode extent count overflow
Chandan Babu R [Sat, 23 Jan 2021 00:48:10 +0000 (16:48 -0800)]
xfs: Add helper for checking per-inode extent count overflow

XFS does not check for possible overflow of per-inode extent counter
fields when adding extents to either data or attr fork.

For e.g.
1. Insert 5 million xattrs (each having a value size of 255 bytes) and
   then delete 50% of them in an alternating manner.

2. On a 4k block sized XFS filesystem instance, the above causes 98511
   extents to be created in the attr fork of the inode.

   xfsaild/loop0  2008 [003]  1475.127209: probe:xfs_inode_to_disk: (ffffffffa43fb6b0) if_nextents=98511 i_ino=131

3. The incore inode fork extent counter is a signed 32-bit
   quantity. However the on-disk extent counter is an unsigned 16-bit
   quantity and hence cannot hold 98511 extents.

4. The following incorrect value is stored in the attr extent counter,
   # xfs_db -f -c 'inode 131' -c 'print core.naextents' /dev/loop0
   core.naextents = -32561

This commit adds a new helper function (i.e.
xfs_iext_count_may_overflow()) to check for overflow of the per-inode
data and xattr extent counters. Future patches will use this function to
make sure that an FS operation won't cause the extent counter to
overflow.

Suggested-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
3 years agoxfs: fix an ABBA deadlock in xfs_rename
Darrick J. Wong [Sat, 23 Jan 2021 00:48:32 +0000 (16:48 -0800)]
xfs: fix an ABBA deadlock in xfs_rename

When overlayfs is running on top of xfs and the user unlinks a file in
the overlay, overlayfs will create a whiteout inode and ask xfs to
"rename" the whiteout file atop the one being unlinked.  If the file
being unlinked loses its one nlink, we then have to put the inode on the
unlinked list.

This requires us to grab the AGI buffer of the whiteout inode to take it
off the unlinked list (which is where whiteouts are created) and to grab
the AGI buffer of the file being deleted.  If the whiteout was created
in a higher numbered AG than the file being deleted, we'll lock the AGIs
in the wrong order and deadlock.

Therefore, grab all the AGI locks we think we'll need ahead of time, and
in order of increasing AG number per the locking rules.

Reported-by: wenli xie <wlxie7296@gmail.com>
Fixes: 93597ae8dac0 ("xfs: Fix deadlock between AGI and AGF when target_ip exists in xfs_rename()")
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
3 years agoLinux 5.11-rc4
Linus Torvalds [Mon, 18 Jan 2021 00:37:05 +0000 (16:37 -0800)]
Linux 5.11-rc4

3 years agoMerge tag 'perf-tools-fixes-2021-01-17' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 17 Jan 2021 21:14:46 +0000 (13:14 -0800)]
Merge tag 'perf-tools-fixes-2021-01-17' of git://git./linux/kernel/git/acme/linux

Pull perf tools fixes from Arnaldo Carvalho de Melo:

 - Fix 'CPU too large' error in Intel PT

 - Correct event attribute sizes in 'perf inject'

 - Sync build_bug.h and kvm.h kernel copies

 - Fix bpf.h header include directive in 5sec.c 'perf trace' bpf example

 - libbpf tests fixes

 - Fix shadow stat 'perf test' for non-bash shells

 - Take cgroups into account for shadow stats in 'perf stat'

* tag 'perf-tools-fixes-2021-01-17' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
  perf inject: Correct event attribute sizes
  perf intel-pt: Fix 'CPU too large' error
  perf stat: Take cgroups into account for shadow stats
  perf stat: Introduce struct runtime_stat_data
  libperf tests: Fail when failing to get a tracepoint id
  libperf tests: If a test fails return non-zero
  libperf tests: Avoid uninitialized variable warning
  perf test: Fix shadow stat test for non-bash shells
  tools headers: Syncronize linux/build_bug.h with the kernel sources
  tools headers UAPI: Sync kvm.h headers with the kernel sources
  perf bpf examples: Fix bpf.h header include directive in 5sec.c example

3 years agoMerge tag 'powerpc-5.11-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sun, 17 Jan 2021 20:28:58 +0000 (12:28 -0800)]
Merge tag 'powerpc-5.11-4' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "One fix for a lack of alignment in our linker script, that can lead to
  crashes depending on configuration etc.

  One fix for the 32-bit VDSO after the C VDSO conversion.

  Thanks to Andreas Schwab, Ariel Marcovitch, and Christophe Leroy"

* tag 'powerpc-5.11-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/vdso: Fix clock_gettime_fallback for vdso32
  powerpc: Fix alignment bug within the init sections

3 years agoMerge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sun, 17 Jan 2021 20:16:47 +0000 (12:16 -0800)]
Merge branch 'fixes' of git://git./linux/kernel/git/viro/vfs

Pull misc vfs fixes from Al Viro:
 "Several assorted fixes.

  I still think that audit ->d_name race is better fixed this way for
  the benefit of backports, with any possibly fancier variants done on
  top of it"

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  dump_common_audit_data(): fix racy accesses to ->d_name
  iov_iter: fix the uaccess area in copy_compat_iovec_from_user
  umount(2): move the flag validity checks first

3 years agomm: don't put pinned pages into the swap cache
Linus Torvalds [Sat, 16 Jan 2021 23:34:57 +0000 (15:34 -0800)]
mm: don't put pinned pages into the swap cache

So technically there is nothing wrong with adding a pinned page to the
swap cache, but the pinning obviously means that the page can't actually
be free'd right now anyway, so it's a bit pointless.

However, the real problem is not with it being a bit pointless: the real
issue is that after we've added it to the swap cache, we'll try to unmap
the page.  That will succeed, because the code in mm/rmap.c doesn't know
or care about pinned pages.

Even the unmapping isn't fatal per se, since the page will stay around
in memory due to the pinning, and we do hold the connection to it using
the swap cache.  But when we then touch it next and take a page fault,
the logic in do_swap_page() will map it back into the process as a
possibly read-only page, and we'll then break the page association on
the next COW fault.

Honestly, this issue could have been fixed in any of those other places:
(a) we could refuse to unmap a pinned page (which makes conceptual
sense), or (b) we could make sure to re-map a pinned page writably in
do_swap_page(), or (c) we could just make do_wp_page() not COW the
pinned page (which was what we historically did before that "mm:
do_wp_page() simplification" commit).

But while all of them are equally valid models for breaking this chain,
not putting pinned pages into the swap cache in the first place is the
simplest one by far.

It's also the safest one: the reason why do_wp_page() was changed in the
first place was that getting the "can I re-use this page" wrong is so
fraught with errors.  If you do it wrong, you end up with an incorrectly
shared page.

As a result, using "page_maybe_dma_pinned()" in either do_wp_page() or
do_swap_page() would be a serious bug since it is only a (very good)
heuristic.  Re-using the page requires a hard black-and-white rule with
no room for ambiguity.

In contrast, saying "this page is very likely dma pinned, so let's not
add it to the swap cache and try to unmap it" is an obviously safe thing
to do, and if the heuristic might very rarely be a false positive, no
harm is done.

Fixes: 09854ba94c6a ("mm: do_wp_page() simplification")
Reported-and-tested-by: Martin Raiber <martin@urbackup.org>
Cc: Pavel Begunkov <asml.silence@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 16 Jan 2021 20:25:40 +0000 (12:25 -0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Nine minor fixes, seven in drivers and two in the core SCSI disk
  driver (sd) which should be harmless involving removing an unused
  variable and quietening a spurious warning"

Signed-off-by: James E.J. Bottomley <jejb@linux.ibm.com>
* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: sd: Remove obsolete variable in sd_remove()
  scsi: sd: Suppress spurious errors when WRITE SAME is being disabled
  scsi: scsi_debug: Fix memleak in scsi_debug_init()
  scsi: mpt3sas: Fix spelling mistake in Kconfig "compatiblity" -> "compatibility"
  scsi: qedi: Correct max length of CHAP secret
  scsi: ufs: Correct the LUN used in eh_device_reset_handler() callback
  scsi: ufs: Relocate flush of exceptional event
  scsi: ufs: Relax the condition of UFSHCI_QUIRK_SKIP_MANUAL_WB_FLUSH_CTRL
  scsi: ufs: Fix possible power drain during system suspend

3 years agodump_common_audit_data(): fix racy accesses to ->d_name
Al Viro [Tue, 5 Jan 2021 19:43:46 +0000 (14:43 -0500)]
dump_common_audit_data(): fix racy accesses to ->d_name

We are not guaranteed the locking environment that would prevent
dentry getting renamed right under us.  And it's possible for
old long name to be freed after rename, leading to UAF here.

Cc: stable@kernel.org # v2.6.2+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
3 years agoMerge tag 'block-5.11-2021-01-16' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 16 Jan 2021 19:39:58 +0000 (11:39 -0800)]
Merge tag 'block-5.11-2021-01-16' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "Just an nvme pull request via Christoph:

   - don't initialize hwmon for discover controllers (Sagi Grimberg)

   - fix iov_iter handling in nvme-tcp (Sagi Grimberg)

   - fix a preempt warning in nvme-tcp (Sagi Grimberg)

   - fix a possible NULL pointer dereference in nvme (Israel Rukshin)"

* tag 'block-5.11-2021-01-16' of git://git.kernel.dk/linux-block:
  nvme: don't intialize hwmon for discovery controllers
  nvme-tcp: fix possible data corruption with bio merges
  nvme-tcp: Fix warning with CONFIG_DEBUG_PREEMPT
  nvmet-rdma: Fix NULL deref when setting pi_enable and traddr INADDR_ANY

3 years agoMerge tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 16 Jan 2021 19:12:02 +0000 (11:12 -0800)]
Merge tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block

Pull io_uring fixes from Jens Axboe:
 "We still have a pending fix for a cancelation issue, but it's still
  being investigated. In the meantime:

   - Dead mm handling fix (Pavel)

   - SQPOLL setup error handling (Pavel)

   - Flush timeout sequence fix (Marcelo)

   - Missing finish_wait() for one exit case"

* tag 'io_uring-5.11-2021-01-16' of git://git.kernel.dk/linux-block:
  io_uring: ensure finish_wait() is always called in __io_uring_task_cancel()
  io_uring: flush timeouts that should already have expired
  io_uring: do sqo disable on install_fd error
  io_uring: fix null-deref in io_disable_sqo_submit
  io_uring: don't take files/mm for a dead task
  io_uring: drop mm and files after task_work_run

3 years agoMerge tag 'riscv-for-linus-5.11-rc4' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 16 Jan 2021 19:00:08 +0000 (11:00 -0800)]
Merge tag 'riscv-for-linus-5.11-rc4' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Palmer Dabbelt:
 "There are a few more fixes than a normal rc4, largely due to the
  bubble introduced by the holiday break:

   - return -ENOSYS for syscall number -1, which previously returned an
     uninitialized value.

   - ensure of_clk_init() has been called in time_init(), without which
     clock drivers may not be initialized.

   - fix sifive,uart0 driver to properly display the baud rate. A fix to
     initialize MPIE that allows interrupts to be processed during
     system calls.

   - avoid erronously begin tracing IRQs when interrupts are disabled,
     which at least triggers suprious lockdep failures.

   - workaround for a warning related to calling smp_processor_id()
     while preemptible. The warning itself is suprious on currently
     availiable systems.

   - properly include the generic time VDSO calls. A fix to our kasan
     address mapping. A fix to the HiFive Unleashed device tree, which
     allows the Ethernet PHY to be properly initialized by Linux (as
     opposed to relying on the bootloader).

   - defconfig update to include SiFive's GPIO driver, which is present
     on the HiFive Unleashed and necessary to initialize the PHY.

   - avoid allocating memory while initializing reserved memory.

   - avoid allocating the last 4K of memory, as pointers there alias
     with syscall errors.

  There are also two cleanups that should have no functional effect but
  do fix build warnings:

   - drop a duplicated definition of PAGE_KERNEL_EXEC.

   - properly declare the asm register SP shim.

   - cleanup the rv32 memory size Kconfig entry, to reflect the actual
     size of memory availiable"

* tag 'riscv-for-linus-5.11-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  RISC-V: Fix maximum allowed phsyical memory for RV32
  RISC-V: Set current memblock limit
  RISC-V: Do not allocate memblock while iterating reserved memblocks
  riscv: stacktrace: Move register keyword to beginning of declaration
  riscv: defconfig: enable gpio support for HiFive Unleashed
  dts: phy: add GPIO number and active state used for phy reset
  dts: phy: fix missing mdio device and probe failure of vsc8541-01 device
  riscv: Fix KASAN memory mapping.
  riscv: Fixup CONFIG_GENERIC_TIME_VSYSCALL
  riscv: cacheinfo: Fix using smp_processor_id() in preemptible
  riscv: Trace irq on only interrupt is enabled
  riscv: Drop a duplicated PAGE_KERNEL_EXEC
  riscv: Enable interrupts during syscalls with M-Mode
  riscv: Fix sifive serial driver
  riscv: Fix kernel time_init()
  riscv: return -ENOSYS for syscall -1

3 years agomm: don't play games with pinned pages in clear_page_refs
Linus Torvalds [Sun, 10 Jan 2021 01:09:10 +0000 (17:09 -0800)]
mm: don't play games with pinned pages in clear_page_refs

Turning a pinned page read-only breaks the pinning after COW.  Don't do it.

The whole "track page soft dirty" state doesn't work with pinned pages
anyway, since the page might be dirtied by the pinning entity without
ever being noticed in the page tables.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm: fix clear_refs_write locking
Linus Torvalds [Fri, 8 Jan 2021 21:13:41 +0000 (13:13 -0800)]
mm: fix clear_refs_write locking

Turning page table entries read-only requires the mmap_sem held for
writing.

So stop doing the odd games with turning things from read locks to write
locks and back.  Just get the write lock.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agoRISC-V: Fix maximum allowed phsyical memory for RV32
Atish Patra [Mon, 11 Jan 2021 23:45:04 +0000 (15:45 -0800)]
RISC-V: Fix maximum allowed phsyical memory for RV32

Linux kernel can only map 1GB of address space for RV32 as the page offset
is set to 0xC0000000. The current description in the Kconfig is confusing
as it indicates that RV32 can support 2GB of physical memory. That is
simply not true for current kernel. In future, a 2GB split support can be
added to allow 2GB physical address space.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: Set current memblock limit
Atish Patra [Mon, 11 Jan 2021 23:45:02 +0000 (15:45 -0800)]
RISC-V: Set current memblock limit

Currently, linux kernel can not use last 4k bytes of addressable space
because IS_ERR_VALUE macro treats those as an error. This will be an issue
for RV32 as any memblock allocator potentially allocate chunk of memory
from the end of DRAM (2GB) leading bad address error even though the
address was technically valid.

Fix this issue by limiting the memblock if available memory spans the
entire address space.

Reviewed-by: Anup Patel <anup@brainfault.org>
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoRISC-V: Do not allocate memblock while iterating reserved memblocks
Atish Patra [Mon, 11 Jan 2021 23:45:01 +0000 (15:45 -0800)]
RISC-V: Do not allocate memblock while iterating reserved memblocks

Currently, resource tree allocates memory blocks while iterating on the
list. It leads to following kernel warning because memblock allocation
also invokes memory block reservation API.

[    0.000000] ------------[ cut here ]------------
[    0.000000] WARNING: CPU: 0 PID: 0 at kernel/resource.c:795
__insert_resource+0x8e/0xd0
[    0.000000] Modules linked in:
[    0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted
5.10.0-00022-ge20097fb37e2-dirty #549
[    0.000000] epc: c00125c2 ra : c001262c sp : c1c01f50
[    0.000000]  gp : c1d456e0 tp : c1c0a980 t0 : ffffcf20
[    0.000000]  t1 : 00000000 t2 : 00000000 s0 : c1c01f60
[    0.000000]  s1 : ffffcf00 a0 : ffffff00 a1 : c1c0c0c4
[    0.000000]  a2 : 80c12b15 a3 : 80402000 a4 : 80402000
[    0.000000]  a5 : c1c0c0c4 a6 : 80c12b15 a7 : f5faf600
[    0.000000]  s2 : c1c0c0c4 s3 : c1c0e000 s4 : c1009a80
[    0.000000]  s5 : c1c0c000 s6 : c1d48000 s7 : c1613b4c
[    0.000000]  s8 : 00000fff s9 : 80000200 s10: c1613b40
[    0.000000]  s11: 00000000 t3 : c1d4a000 t4 : ffffffff

This is also unnecessary as we can pre-compute the total memblocks required
for each memory region and allocate it before the loop. It save precious
boot time not going through memblock allocation code every time.

Fixes: 00ab027a3b82 ("RISC-V: Add kernel image sections to the resource tree")

Reviewed-by: Anup Patel <anup@brainfault.org>
Tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Atish Patra <atish.patra@wdc.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
3 years agoiov_iter: fix the uaccess area in copy_compat_iovec_from_user
Christoph Hellwig [Mon, 11 Jan 2021 17:19:26 +0000 (18:19 +0100)]
iov_iter: fix the uaccess area in copy_compat_iovec_from_user

sizeof needs to be called on the compat pointer, not the native one.

Fixes: 89cd35c58bc2 ("iov_iter: transparently handle compat iovecs in import_iovec")
Reported-by: David Laight <David.Laight@ACULAB.COM>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
3 years agoMerge tag 'for-5.11/dm-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 16 Jan 2021 02:01:17 +0000 (18:01 -0800)]
Merge tag 'for-5.11/dm-fixes-1' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - Fix DM-raid's raid1 discard limits so discards work.

 - Select missing Kconfig dependencies for DM integrity and zoned
   targets.

 - Four fixes for DM crypt target's support to optionally bypass kcryptd
   workqueues.

 - Fix DM snapshot merge supports missing data flushes before committing
   metadata.

 - Fix DM integrity data device flushing when external metadata is used.

 - Fix DM integrity's maximum number of supported constructor arguments
   that user can request when creating an integrity device.

 - Eliminate DM core ioctl logging noise when an ioctl is issued without
   required CAP_SYS_RAWIO permission.

* tag 'for-5.11/dm-fixes-1' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm crypt: defer decryption to a tasklet if interrupts disabled
  dm integrity: fix the maximum number of arguments
  dm crypt: do not call bio_endio() from the dm-crypt tasklet
  dm integrity: fix flush with external metadata device
  dm: eliminate potential source of excessive kernel log noise
  dm snapshot: flush merged data before committing metadata
  dm crypt: use GFP_ATOMIC when allocating crypto requests from softirq
  dm crypt: do not wait for backlogged crypto request completion in softirq
  dm zoned: select CONFIG_CRC32
  dm integrity: select CRYPTO_SKCIPHER
  dm raid: fix discard limits for raid1

3 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Fri, 15 Jan 2021 23:25:45 +0000 (15:25 -0800)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "10 patches.

  Subsystems affected by this patch series: MAINTAINERS and mm (slub,
  pagealloc, memcg, kasan, vmalloc, migration, hugetlb, memory-failure,
  and process_vm_access)"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  mm/process_vm_access.c: include compat.h
  mm,hwpoison: fix printing of page flags
  MAINTAINERS: add Vlastimil as slab allocators maintainer
  mm/hugetlb: fix potential missing huge page size info
  mm: migrate: initialize err in do_migrate_pages
  mm/vmalloc.c: fix potential memory leak
  arm/kasan: fix the array size of kasan_early_shadow_pte[]
  mm/memcontrol: fix warning in mem_cgroup_page_lruvec()
  mm/page_alloc: add a missing mm_page_alloc_zone_locked() tracepoint
  mm, slub: consider rest of partial list if acquire_slab() fails

3 years agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma
Linus Torvalds [Fri, 15 Jan 2021 23:14:40 +0000 (15:14 -0800)]
Merge tag 'for-linus' of git://git./linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "A fairly modest set of bug fixes, nothing abnormal from the merge
  window

  The ucma patch is a bit on the larger side, but given the regression
  was recently added I've opted to forward it to the rc stream.

   - Fix a ucma memory leak introduced in v5.9 while fixing the
     Syzkaller bugs

   - Don't fail when the xarray wraps for user verbs objects

   - User triggerable oops regression from the umem page size rework

   - Error unwind bugs in usnic, ocrdma, mlx5 and cma"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/cma: Fix error flow in default_roce_mode_store
  RDMA/mlx5: Fix wrong free of blue flame register on error
  IB/mlx5: Fix error unwinding when set_has_smi_cap fails
  RDMA/umem: Avoid undefined behavior of rounddown_pow_of_two()
  RDMA/ocrdma: Fix use after free in ocrdma_dealloc_ucontext_pd()
  RDMA/usnic: Fix memleak in find_free_vf_and_create_qp_grp
  RDMA/restrack: Don't treat as an error allocation ID wrapping
  RDMA/ucma: Do not miss ctx destruction steps in some cases

3 years agoio_uring: ensure finish_wait() is always called in __io_uring_task_cancel()
Jens Axboe [Fri, 15 Jan 2021 23:04:23 +0000 (16:04 -0700)]
io_uring: ensure finish_wait() is always called in __io_uring_task_cancel()

If we enter with requests pending and performm cancelations, we'll have
a different inflight count before and after calling prepare_to_wait().
This causes the loop to restart. If we actually ended up canceling
everything, or everything completed in-between, then we'll break out
of the loop without calling finish_wait() on the waitqueue. This can
trigger a warning on exit_signals(), as we leave the task state in
TASK_UNINTERRUPTIBLE.

Put a finish_wait() after the loop to catch that case.

Cc: stable@vger.kernel.org # 5.9+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 15 Jan 2021 22:54:24 +0000 (14:54 -0800)]
Merge tag 'ext4_for_linus_stable' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 fixes from Ted Ts'o:
 "A number of bug fixes for ext4:

   - Fix for the new fast_commit feature

   - Fix some error handling codepaths in whiteout handling and
     mountpoint sampling

   - Fix how we write ext4_error information so it goes through the
     journal when journalling is active, to avoid races that can lead to
     lost error information, superblock checksum failures, or DIF/DIX
     features"

* tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: remove expensive flush on fast commit
  ext4: fix bug for rename with RENAME_WHITEOUT
  ext4: fix wrong list_splice in ext4_fc_cleanup
  ext4: use IS_ERR instead of IS_ERR_OR_NULL and set inode null when IS_ERR
  ext4: don't leak old mountpoint samples
  ext4: drop ext4_handle_dirty_super()
  ext4: fix superblock checksum failure when setting password salt
  ext4: use sbi instead of EXT4_SB(sb) in ext4_update_super()
  ext4: save error info to sb through journal if available
  ext4: protect superblock modifications with a buffer lock
  ext4: drop sync argument of ext4_commit_super()
  ext4: combine ext4_handle_error() and save_error_info()

3 years agoMerge tag '5.11-rc3-smb3' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Fri, 15 Jan 2021 22:39:21 +0000 (14:39 -0800)]
Merge tag '5.11-rc3-smb3' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Two small cifs fixes for stable (including an important handle leak
  fix) and three small cleanup patches"

* tag '5.11-rc3-smb3' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: style: replace one-element array with flexible-array
  cifs: connect: style: Simplify bool comparison
  fs: cifs: remove unneeded variable in smb3_fs_context_dup
  cifs: fix interrupted close commands
  cifs: check pointer before freeing

3 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 15 Jan 2021 21:11:51 +0000 (13:11 -0800)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Catalin Marinas:

 - Set the minimum GCC version to 5.1 for arm64 due to earlier compiler
   bugs.

 - Make atomic helpers __always_inline to avoid a section mismatch when
   compiling with clang.

 - Fix the CMA and crashkernel reservations to use ZONE_DMA (remove the
   arm64_dma32_phys_limit variable, no longer needed with a dynamic
   ZONE_DMA sizing in 5.11).

 - Remove redundant IRQ flag tracing that was leaving lockdep
   inconsistent with the hardware state.

 - Revert perf events based hard lockup detector that was causing
   smp_processor_id() to be called in preemptible context.

 - Some trivial cleanups - spelling fix, renaming S_FRAME_SIZE to
   PT_REGS_SIZE, function prototypes added.

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: selftests: Fix spelling of 'Mismatch'
  arm64: syscall: include prototype for EL0 SVC functions
  compiler.h: Raise minimum version of GCC to 5.1 for arm64
  arm64: make atomic helpers __always_inline
  arm64: rename S_FRAME_SIZE to PT_REGS_SIZE
  Revert "arm64: Enable perf events based hard lockup detector"
  arm64: entry: remove redundant IRQ flag tracing
  arm64: Remove arm64_dma32_phys_limit and its uses

3 years agoMerge tag 'mips_fixes_5.11.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mips...
Linus Torvalds [Fri, 15 Jan 2021 21:07:33 +0000 (13:07 -0800)]
Merge tag 'mips_fixes_5.11.1' of git://git./linux/kernel/git/mips/linux

Pull MIPS fixes from Thomas Bogendoerfer:

 - fix coredumps on 64bit kernels

 - fix for alignment bugs preventing booting

 - fix checking for failed irq_alloc_desc calls

* tag 'mips_fixes_5.11.1' of git://git.kernel.org/pub/scm/linux/kernel/git/mips/linux:
  MIPS: OCTEON: fix unreachable code in octeon_irq_init_ciu
  MIPS: relocatable: fix possible boot hangup with KASLR enabled
  MIPS: Fix malformed NT_FILE and NT_SIGINFO in 32bit coredumps
  MIPS: boot: Fix unaligned access with CONFIG_MIPS_RAW_APPENDED_DTB

3 years agoperf inject: Correct event attribute sizes
Al Grant [Tue, 24 Nov 2020 19:58:17 +0000 (19:58 +0000)]
perf inject: Correct event attribute sizes

When 'perf inject' reads a perf.data file from an older version of perf,
it writes event attributes into the output with the original size field,
but lays them out as if they had the size currently used. Readers see a
corrupt file. Update the size field to match the layout.

Signed-off-by: Al Grant <al.grant@foss.arm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20201124195818.30603-1-al.grant@arm.com
Signed-off-by: Denis Nikitin <denik@chromium.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf intel-pt: Fix 'CPU too large' error
Adrian Hunter [Thu, 7 Jan 2021 17:41:59 +0000 (19:41 +0200)]
perf intel-pt: Fix 'CPU too large' error

In some cases, the number of cpus (nr_cpus_online) is confused with the
maximum cpu number (nr_cpus_avail), which results in the error in the
example below:

Example on system with 8 cpus:

 Before:
   # echo 0 > /sys/devices/system/cpu/cpu2/online
   # ./perf record --kcore -e intel_pt// taskset --cpu-list 7 uname
   Linux
   [ perf record: Woken up 1 times to write data ]
   [ perf record: Captured and wrote 0.147 MB perf.data ]
   # ./perf script --itrace=e
   Requested CPU 7 too large. Consider raising MAX_NR_CPUS
   0x25908 [0x8]: failed to process type: 68 [Invalid argument]

 After:
   # ./perf script --itrace=e
   #

Fixes: 8c7274691f0d ("perf machine: Replace MAX_NR_CPUS with perf_env::nr_cpus_online")
Fixes: 7df4e36a4785 ("perf session: Replace MAX_NR_CPUS with perf_env::nr_cpus_online")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: stable@vger.kernel.org
Link: http://lore.kernel.org/lkml/20210107174159.24897-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf stat: Take cgroups into account for shadow stats
Namhyung Kim [Fri, 15 Jan 2021 07:11:39 +0000 (16:11 +0900)]
perf stat: Take cgroups into account for shadow stats

As of now it doesn't consider cgroups when collecting shadow stats and
metrics so counter values from different cgroups will be saved in a same
slot.  This resulted in incorrect numbers when those cgroups have
different workloads.

For example, let's look at the scenario below: cgroups A and C runs same
workload which burns a cpu while cgroup B runs a light workload.

  $ perf stat -a -e cycles,instructions --for-each-cgroup A,B,C  sleep 1

   Performance counter stats for 'system wide':

     3,958,116,522      cycles                A
     6,722,650,929      instructions          A #    2.53  insn per cycle
         1,132,741      cycles                B
           571,743      instructions          B #    0.00  insn per cycle
     4,007,799,935      cycles                C
     6,793,181,523      instructions          C #    2.56  insn per cycle

       1.001050869 seconds time elapsed

When I run 'perf stat' with single workload, it usually shows IPC around
1.7.  We can verify it (6,722,650,929.0 / 3,958,116,522 = 1.698) for cgroup A.

But in this case, since cgroups are ignored, cycles are averaged so it
used the lower value for IPC calculation and resulted in around 2.5.

  avg cycle: (3958116522 + 1132741 + 4007799935) / 3 = 2655683066
  IPC (A)  :  6722650929 / 2655683066 = 2.531
  IPC (B)  :      571743 / 2655683066 = 0.0002
  IPC (C)  :  6793181523 / 2655683066 = 2.557

We can simply compare cgroup pointers in the evsel and it'll be NULL
when cgroups are not specified.  With this patch, I can see correct
numbers like below:

  $ perf stat -a -e cycles,instructions --for-each-cgroup A,B,C  sleep 1

  Performance counter stats for 'system wide':

     4,171,051,687      cycles                A
     7,219,793,922      instructions          A #    1.73  insn per cycle
         1,051,189      cycles                B
           583,102      instructions          B #    0.55  insn per cycle
     4,171,124,710      cycles                C
     7,192,944,580      instructions          C #    1.72  insn per cycle

       1.007909814 seconds time elapsed

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210115071139.257042-2-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf stat: Introduce struct runtime_stat_data
Namhyung Kim [Fri, 15 Jan 2021 07:11:38 +0000 (16:11 +0900)]
perf stat: Introduce struct runtime_stat_data

To pass more info to the saved_value in the runtime_stat, add a new
struct runtime_stat_data.  Currently it only has 'ctx' field but later
patch will add more.

Note that we intentionally pass 0 as ctx to clock-related events for
compatibility.  It was already there in a few places.  So move the code
into the saved_value_lookup() explicitly and add a comment.

Suggested-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210115071139.257042-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agolibperf tests: Fail when failing to get a tracepoint id
Ian Rogers [Thu, 14 Jan 2021 18:02:50 +0000 (10:02 -0800)]
libperf tests: Fail when failing to get a tracepoint id

Permissions are necessary to get a tracepoint id. Fail the test when the
read fails.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210114180250.3853825-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agolibperf tests: If a test fails return non-zero
Ian Rogers [Thu, 14 Jan 2021 18:02:49 +0000 (10:02 -0800)]
libperf tests: If a test fails return non-zero

If a test fails return -1 rather than 0. This is consistent with the
return value in test-cpumap.c

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210114180250.3853825-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agolibperf tests: Avoid uninitialized variable warning
Ian Rogers [Thu, 14 Jan 2021 21:23:04 +0000 (13:23 -0800)]
libperf tests: Avoid uninitialized variable warning

The variable 'bf' is read (for a write call) without being initialized
triggering a memory sanitizer warning. Use 'bf' in the read and switch
the write to reading from a string.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210114212304.4018119-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoext4: remove expensive flush on fast commit
Daejun Park [Wed, 6 Jan 2021 01:32:42 +0000 (10:32 +0900)]
ext4: remove expensive flush on fast commit

In the fast commit, it adds REQ_FUA and REQ_PREFLUSH on each fast
commit block when barrier is enabled.  However, in recovery phase,
ext4 compares CRC value in the tail.  So it is sufficient to add
REQ_FUA and REQ_PREFLUSH on the block that has tail.

Signed-off-by: Daejun Park <daejun7.park@samsung.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20210106013242epcms2p5b6b4ed8ca86f29456fdf56aa580e74b4@epcms2p5
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
3 years agoext4: fix bug for rename with RENAME_WHITEOUT
yangerkun [Tue, 5 Jan 2021 06:28:57 +0000 (14:28 +0800)]
ext4: fix bug for rename with RENAME_WHITEOUT

We got a "deleted inode referenced" warning cross our fsstress test. The
bug can be reproduced easily with following steps:

  cd /dev/shm
  mkdir test/
  fallocate -l 128M img
  mkfs.ext4 -b 1024 img
  mount img test/
  dd if=/dev/zero of=test/foo bs=1M count=128
  mkdir test/dir/ && cd test/dir/
  for ((i=0;i<1000;i++)); do touch file$i; done # consume all block
  cd ~ && renameat2(AT_FDCWD, /dev/shm/test/dir/file1, AT_FDCWD,
    /dev/shm/test/dir/dst_file, RENAME_WHITEOUT) # ext4_add_entry in
    ext4_rename will return ENOSPC!!
  cd /dev/shm/ && umount test/ && mount img test/ && ls -li test/dir/file1
  We will get the output:
  "ls: cannot access 'test/dir/file1': Structure needs cleaning"
  and the dmesg show:
  "EXT4-fs error (device loop0): ext4_lookup:1626: inode #2049: comm ls:
  deleted inode referenced: 139"

ext4_rename will create a special inode for whiteout and use this 'ino'
to replace the source file's dir entry 'ino'. Once error happens
latter(the error above was the ENOSPC return from ext4_add_entry in
ext4_rename since all space has been consumed), the cleanup do drop the
nlink for whiteout, but forget to restore 'ino' with source file. This
will trigger the bug describle as above.

Signed-off-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: stable@vger.kernel.org
Fixes: cd808deced43 ("ext4: support RENAME_WHITEOUT")
Link: https://lore.kernel.org/r/20210105062857.3566-1-yangerkun@huawei.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
3 years agoext4: fix wrong list_splice in ext4_fc_cleanup
Daejun Park [Wed, 30 Dec 2020 09:48:51 +0000 (18:48 +0900)]
ext4: fix wrong list_splice in ext4_fc_cleanup

After full/fast commit, entries in staging queue are promoted to main
queue. In ext4_fs_cleanup function, it splice to staging queue to
staging queue.

Fixes: aa75f4d3daaeb ("ext4: main fast-commit commit path")
Signed-off-by: Daejun Park <daejun7.park@samsung.com>
Reviewed-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://lore.kernel.org/r/20201230094851epcms2p6eeead8cc984379b37b2efd21af90fd1a@epcms2p6
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
3 years agoext4: use IS_ERR instead of IS_ERR_OR_NULL and set inode null when IS_ERR
Yi Li [Wed, 30 Dec 2020 03:38:27 +0000 (11:38 +0800)]
ext4: use IS_ERR instead of IS_ERR_OR_NULL and set inode null when IS_ERR

1: ext4_iget/ext4_find_extent never returns NULL, use IS_ERR
instead of IS_ERR_OR_NULL to fix this.

2: ext4_fc_replay_inode should set the inode to NULL when IS_ERR.
and go to call iput properly.

Fixes: 8016e29f4362 ("ext4: fast commit recovery path")
Signed-off-by: Yi Li <yili@winhong.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20201230033827.3996064-1-yili@winhong.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
3 years agoperf test: Fix shadow stat test for non-bash shells
Namhyung Kim [Thu, 14 Jan 2021 05:06:09 +0000 (14:06 +0900)]
perf test: Fix shadow stat test for non-bash shells

It was using some bash-specific features and failed to parse when
running with a different shell like below:

  root@kbl-ppc:~/kbl-ws/perf-dev/lck-9077/acme.tmp/tools/perf# ./perf test 83 -vv
  83: perf stat metrics (shadow stat) test                            :
  --- start ---
  test child forked, pid 3922
  ./tests/shell/stat+shadow_stat.sh: 19: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 24: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 30: ./tests/shell/stat+shadow_stat.sh: [[: not found
  (standard_in) 2: syntax error
  ./tests/shell/stat+shadow_stat.sh: 36: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 19: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 24: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 30: ./tests/shell/stat+shadow_stat.sh: [[: not found
  (standard_in) 2: syntax error
  ./tests/shell/stat+shadow_stat.sh: 36: ./tests/shell/stat+shadow_stat.sh: [[: not found
  ./tests/shell/stat+shadow_stat.sh: 45: ./tests/shell/stat+shadow_stat.sh: declare: not found
  test child finished with -1
  ---- end ----
  perf stat metrics (shadow stat) test: FAILED!

Reported-by: Jin Yao <yao.jin@linux.intel.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Laight <david.laight@aculab.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lore.kernel.org/lkml/20210114050609.1258820-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agotools headers: Syncronize linux/build_bug.h with the kernel sources
Arnaldo Carvalho de Melo [Thu, 14 Jan 2021 15:35:27 +0000 (12:35 -0300)]
tools headers: Syncronize linux/build_bug.h with the kernel sources

To pick up the changes in:

  3a176b94609a18f5 ("Revert "kbuild: avoid static_assert for genksyms"")

And silence this perf build warning:

  Warning: Kernel ABI header at 'tools/include/linux/build_bug.h' differs from latest version at 'include/linux/build_bug.h'
  diff -u tools/include/linux/build_bug.h include/linux/build_bug.h

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masahiro Yamada <masahiroy@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agotools headers UAPI: Sync kvm.h headers with the kernel sources
Arnaldo Carvalho de Melo [Thu, 14 Jan 2021 15:26:08 +0000 (12:26 -0300)]
tools headers UAPI: Sync kvm.h headers with the kernel sources

To pick the changes in:

  647daca25d24fb6e ("KVM: SVM: Add support for booting APs in an SEV-ES guest")

That don't cause any tooling change, just silences this perf build
warning:

  Warning: Kernel ABI header at 'tools/include/uapi/linux/kvm.h' differs from latest version at 'include/uapi/linux/kvm.h'
  diff -u tools/include/uapi/linux/kvm.h include/uapi/linux/kvm.h

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoperf bpf examples: Fix bpf.h header include directive in 5sec.c example
Arnaldo Carvalho de Melo [Tue, 29 Dec 2020 18:41:19 +0000 (15:41 -0300)]
perf bpf examples: Fix bpf.h header include directive in 5sec.c example

It was looking at bpf/bpf.h, which caused this problem:

  # perf trace -e tools/perf/examples/bpf/5sec.c
  /home/acme/git/perf/tools/perf/examples/bpf/5sec.c:42:10: fatal error: 'bpf/bpf.h' file not found
  #include <bpf/bpf.h>
           ^~~~~~~~~~~
  1 error generated.
  ERROR: unable to compile tools/perf/examples/bpf/5sec.c
  Hint: Check error message shown above.
  Hint: You can also pre-compile it into .o using:
        clang -target bpf -O2 -c tools/perf/examples/bpf/5sec.c
        with proper -I and -D options.
  event syntax error: 'tools/perf/examples/bpf/5sec.c'
                       \___ Failed to load tools/perf/examples/bpf/5sec.c from source: Error when compiling BPF scriptlet
  #

Change that to plain bpf.h, to make it work again:

  # perf trace -e tools/perf/examples/bpf/5sec.c sleep 5s
       0.000 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1776891872, rqtp: 5000000000)
  # perf trace -e tools/perf/examples/bpf/5sec.c/max-stack=16/ sleep 5s
       0.000 perf_bpf_probe:hrtimer_nanosleep(__probe_ip: -1776891872, rqtp: 5000000000)
                                         hrtimer_nanosleep ([kernel.kallsyms])
                                         common_nsleep ([kernel.kallsyms])
                                         __x64_sys_clock_nanosleep ([kernel.kallsyms])
                                         do_syscall_64 ([kernel.kallsyms])
                                         entry_SYSCALL_64_after_hwframe ([kernel.kallsyms])
                                         __clock_nanosleep_2 (/usr/lib64/libc-2.32.so)
  # perf trace -e tools/perf/examples/bpf/5sec.c sleep 4s
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
3 years agoMerge tag 'acpi-5.11-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 15 Jan 2021 18:55:33 +0000 (10:55 -0800)]
Merge tag 'acpi-5.11-rc4' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fixes from Rafael Wysocki:
 "These address a device ID bounds check error in the device enumeration
  code and fix a mistake in the documentation.

  Specifics:

   - Harden the ACPI device enumeration code against device ID length
     overflows to address a Linux VM cash on Hyper-V (Dexuan Cui).

   - Fix a mistake in the documentation of error type values for PCIe
     errors (Qiuxu Zhuo)"

* tag 'acpi-5.11-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  Documentation: ACPI: EINJ: Fix error type values for PCIe errors
  ACPI: scan: Harden acpi_device_add() against device ID overflows

3 years agoMerge tag 'for-linus-5.11-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 15 Jan 2021 18:52:00 +0000 (10:52 -0800)]
Merge tag 'for-linus-5.11-rc4-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:

 - A series to fix a regression when running as a fully virtualized
   guest on an old Xen hypervisor not supporting PV interrupt callbacks
   for HVM guests.

 - A patch to add support to query Xen resource sizes (setting was
   possible already) from user mode.

* tag 'for-linus-5.11-rc4-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  x86/xen: Fix xen_hvm_smp_init() when vector callback not available
  x86/xen: Don't register Xen IPIs when they aren't going to be used
  x86/xen: Add xen_no_vector_callback option to test PCI INTX delivery
  xen: Set platform PCI device INTX affinity to CPU0
  xen: Fix event channel callback via INTX/GSI
  xen/privcmd: allow fetching resource sizes

3 years agoMerge tag 'iommu-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Fri, 15 Jan 2021 18:48:58 +0000 (10:48 -0800)]
Merge tag 'iommu-fixes' of git://git./linux/kernel/git/arm64/linux

Pull iommu fixes from Will Deacon:
 "Three IOMMU fixes for -rc4.

  The main one is a change to the Intel IOMMU driver to fix the handling
  of unaligned addresses when invalidating the TLB.

  The fix itself is a bit ugly (the caller does a bunch of shifting
  which is then effectively undone later in the callchain), but Lu has
  patches to clean all of this up in 5.12.

  Summary:

   - Fix address alignment handling for VT-D TLB invalidation

   - Enable workarounds for buggy Qualcomm firmware on two more SoCs

   - Drop duplicate #include"

* tag 'iommu-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  iommu/vt-d: Fix duplicate included linux/dma-map-ops.h
  iommu: arm-smmu-qcom: Add sdm630/msm8998 compatibles for qcom quirks
  iommu/vt-d: Fix unaligned addresses for intel_flush_svm_range_dev()

3 years agoMerge tag 'topic/nouveau-ampere-modeset-2021-01-15' of git://anongit.freedesktop...
Linus Torvalds [Fri, 15 Jan 2021 18:45:54 +0000 (10:45 -0800)]
Merge tag 'topic/nouveau-ampere-modeset-2021-01-15' of git://anongit.freedesktop.org/drm/drm

Pull drm nouveau ampere display support from Dave Airlie:
 "Ben has requested if we can include Ampere modesetting support under
  fixes, it's for new GPUs and shouldn't affect existing hardware.

  It's a bit bigger than just adding a PCI ID, but It has no effect on
  older GPUs"

* tag 'topic/nouveau-ampere-modeset-2021-01-15' of git://anongit.freedesktop.org/drm/drm:
  drm/nouveau/disp/ga10[24]: initial support
  drm/nouveau/dmaobj/ga10[24]: initial support
  drm/nouveau/i2c/ga10[024]: initial support
  drm/nouveau/gpio/ga10[024]: initial support
  drm/nouveau/bar/ga10[024]: initial support
  drm/nouveau/mmu/ga10[024]: initial support
  drm/nouveau/timer/ga10[024]: initial support
  drm/nouveau/fb/ga10[024]: initial support
  drm/nouveau/imem/ga10[024]: initial support
  drm/nouveau/privring/ga10[024]: initial support
  drm/nouveau/mc/ga10[024]: initial support
  drm/nouveau/devinit/ga10[024]: initial support
  drm/nouveau/bios/ga10[024]: initial support
  drm/nouveau/pci/ga10[024]: initial support
  drm/nouveau/core: recognise GA10[024]

3 years agoMerge branch 'acpi-docs'
Rafael J. Wysocki [Fri, 15 Jan 2021 18:15:49 +0000 (19:15 +0100)]
Merge branch 'acpi-docs'

* acpi-docs:
  Documentation: ACPI: EINJ: Fix error type values for PCIe errors

3 years agoio_uring: flush timeouts that should already have expired
Marcelo Diop-Gonzalez [Fri, 15 Jan 2021 16:54:40 +0000 (11:54 -0500)]
io_uring: flush timeouts that should already have expired

Right now io_flush_timeouts() checks if the current number of events
is equal to ->timeout.target_seq, but this will miss some timeouts if
there have been more than 1 event added since the last time they were
flushed (possible in io_submit_flush_completions(), for example). Fix
it by recording the last sequence at which timeouts were flushed so
that the number of events seen can be compared to the number of events
needed without overflow.

Signed-off-by: Marcelo Diop-Gonzalez <marcelo827@gmail.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoarm64: selftests: Fix spelling of 'Mismatch'
Mark Brown [Fri, 8 Jan 2021 18:31:44 +0000 (18:31 +0000)]
arm64: selftests: Fix spelling of 'Mismatch'

The SVE and FPSIMD stress tests have a spelling mistake in the output, fix
it.

Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20210108183144.673-1-broonie@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
3 years agoarm64: syscall: include prototype for EL0 SVC functions
Mark Rutland [Thu, 14 Jan 2021 12:48:12 +0000 (12:48 +0000)]
arm64: syscall: include prototype for EL0 SVC functions

The kbuild test robot reports that when building with W=1, GCC will warn
for a couple of missing prototypes in syscall.c:

|  arch/arm64/kernel/syscall.c:157:6: warning: no previous prototype for 'do_el0_svc' [-Wmissing-prototypes]
|    157 | void do_el0_svc(struct pt_regs *regs)
|        |      ^~~~~~~~~~
|  arch/arm64/kernel/syscall.c:164:6: warning: no previous prototype for 'do_el0_svc_compat' [-Wmissing-prototypes]
|    164 | void do_el0_svc_compat(struct pt_regs *regs)
|        |      ^~~~~~~~~~~~~~~~~

While this isn't a functional problem, as a general policy we should
include the prototype for functions wherever possible to catch any
accidental divergence between the prototype and implementation. Here we
can easily include <asm/exception.h>, so let's do so.

While there are a number of warnings elsewhere and some warnings enabled
under W=1 are of questionable benefit, this change helps to make the
code more robust as it evolved and reduces the noise somewhat, so it
seems worthwhile.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/202101141046.n8iPO3mw-lkp@intel.com
Link: https://lore.kernel.org/r/20210114124812.17754-1-mark.rutland@arm.com
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
3 years agocompiler.h: Raise minimum version of GCC to 5.1 for arm64
Will Deacon [Tue, 12 Jan 2021 22:48:32 +0000 (22:48 +0000)]
compiler.h: Raise minimum version of GCC to 5.1 for arm64

GCC versions >= 4.9 and < 5.1 have been shown to emit memory references
beyond the stack pointer, resulting in memory corruption if an interrupt
is taken after the stack pointer has been adjusted but before the
reference has been executed. This leads to subtle, infrequent data
corruption such as the EXT4 problems reported by Russell King at the
link below.

Life is too short for buggy compilers, so raise the minimum GCC version
required by arm64 to 5.1.

Reported-by: Russell King <linux@armlinux.org.uk>
Suggested-by: Arnd Bergmann <arnd@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
Tested-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: <stable@vger.kernel.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Florian Weimer <fweimer@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/r/20210105154726.GD1551@shell.armlinux.org.uk
Link: https://lore.kernel.org/r/20210112224832.10980-1-will@kernel.org
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>