muen/linux.git
2 years agoxfs: clean up xfs_dir2_leaf_addname
Darrick J. Wong [Sun, 10 Mar 2019 18:46:28 +0000 (11:46 -0700)]
xfs: clean up xfs_dir2_leaf_addname

Remove typedefs and consolidate local variable initialization.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
2 years agoxfs: zero initialize highstale and lowstale in xfs_dir2_leaf_addname
Darrick J. Wong [Sun, 10 Mar 2019 18:41:31 +0000 (11:41 -0700)]
xfs: zero initialize highstale and lowstale in xfs_dir2_leaf_addname

Smatch complains about the following:

fs/xfs/libxfs/xfs_dir2_leaf.c:848 xfs_dir2_leaf_addname() error:
uninitialized symbol 'lowstale'.

fs/xfs/libxfs/xfs_dir2_leaf.c:849 xfs_dir2_leaf_addname() error:
uninitialized symbol 'highstale'.

I don't think there's any incorrect behavior associated with the
uninitialized variable, but as the author of the previous zero-init
patch points out, it's best not to be passing around pointers to
uninitialized stack areas.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Bill O'Donnell <billodo@redhat.com>
2 years agoxfs: clean up xfs_dir2_leafn_add
Darrick J. Wong [Fri, 8 Mar 2019 00:50:11 +0000 (16:50 -0800)]
xfs: clean up xfs_dir2_leafn_add

Remove typedefs and consolidate local variable initialization.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
2 years agoxfs: Zero initialize highstale and lowstale in xfs_dir2_leafn_add
Nathan Chancellor [Fri, 8 Mar 2019 00:48:11 +0000 (16:48 -0800)]
xfs: Zero initialize highstale and lowstale in xfs_dir2_leafn_add

When building with -Wsometimes-uninitialized, Clang warns:

fs/xfs/libxfs/xfs_dir2_node.c:481:6: warning: variable 'lowstale' is
used uninitialized whenever 'if' condition is false
[-Wsometimes-uninitialized]
fs/xfs/libxfs/xfs_dir2_node.c:481:6: warning: variable 'highstale' is
used uninitialized whenever 'if' condition is false
[-Wsometimes-uninitialized]

While it isn't technically wrong, it isn't a problem in practice because
highstale and lowstale are only initialized in xfs_dir2_leafn_add when
compact is not zero then they are passed to xfs_dir3_leaf_find_entry,
where they are initialized before use when compact is zero. Regardless,
it's better not to be passing around uninitialized stack memory so zero
initialize these variables, which silences this warning.

Link: https://github.com/ClangBuiltLinux/linux/issues/393
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: fix reporting supported extra file attributes for statx()
Luis R. Rodriguez [Fri, 1 Mar 2019 16:14:57 +0000 (08:14 -0800)]
xfs: fix reporting supported extra file attributes for statx()

statx(2) notes that any attribute that is not indicated as supported by
stx_attributes_mask has no usable value. Commit 5f955f26f3d42d ("xfs: report
crtime and attribute flags to statx") added support for informing userspace
of extra file attributes but forgot to list these flags as supported
making reporting them rather useless for the pedantic userspace author.

$ git describe --contains 5f955f26f3d42d04aba65590a32eb70eedb7f37d
v4.11-rc6~5^2^2~2

Fixes: 5f955f26f3d42d ("xfs: report crtime and attribute flags to statx")
Signed-off-by: Luis R. Rodriguez <mcgrof@kernel.org>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
[darrick: add a comment reminding people to keep attributes_mask up to date]
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: fix backwards endian conversion in scrub
Darrick J. Wong [Mon, 25 Feb 2019 17:36:36 +0000 (09:36 -0800)]
xfs: fix backwards endian conversion in scrub

Fix a backwards endian conversion of a constant.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
2 years agoxfs: fix uninitialized error variables
Darrick J. Wong [Mon, 25 Feb 2019 17:35:34 +0000 (09:35 -0800)]
xfs: fix uninitialized error variables

smatch complained about some uninitialized error returns, so fix those.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
2 years agoxfs: rework breaking of shared extents in xfs_file_iomap_begin
Darrick J. Wong [Fri, 22 Feb 2019 00:36:17 +0000 (16:36 -0800)]
xfs: rework breaking of shared extents in xfs_file_iomap_begin

Rework the data flow in xfs_file_iomap_begin where we decide if we have
to break shared extents.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: don't pass iomap flags to xfs_reflink_allocate_cow
Darrick J. Wong [Fri, 22 Feb 2019 00:26:35 +0000 (16:26 -0800)]
xfs: don't pass iomap flags to xfs_reflink_allocate_cow

Don't pass raw iomap flags to xfs_reflink_allocate_cow; signal our
intention with a boolean argument.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: fix uninitialized error variable
Colin Ian King [Thu, 21 Feb 2019 15:53:20 +0000 (07:53 -0800)]
xfs: fix uninitialized error variable

A previous commit removed the initialization of variable 'error' to zero,
and can cause a bogus error return.  This occurs when error contains a
non-zero garbage value and the call to xchk_should_terminate detects a
pending fatal signal and checks for a zero error before setting it
to -EAGAIN. Fix the issue by initializing error to zero.

Fixes: b9454fe056bd ("xfs: clean up the inode cluster checking in the inobt scrub")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: introduce an always_cow mode
Christoph Hellwig [Mon, 18 Feb 2019 17:38:49 +0000 (09:38 -0800)]
xfs: introduce an always_cow mode

Add a mode where XFS never overwrites existing blocks in place.  This
is to aid debugging our COW code, and also put infatructure in place
for things like possible future support for zoned block devices, which
can't support overwrites.

This mode is enabled globally by doing a:

    echo 1 > /sys/fs/xfs/debug/always_cow

Note that the parameter is global to allow running all tests in xfstests
easily in this mode, which would not easily be possible with a per-fs
sysfs file.

In always_cow mode persistent preallocations are disabled, and fallocate
will fail when called with a 0 mode (with our without
FALLOC_FL_KEEP_SIZE), and not create unwritten extent for zeroed space
when called with FALLOC_FL_ZERO_RANGE or FALLOC_FL_UNSHARE_RANGE.

There are a few interesting xfstests failures when run in always_cow
mode:

 - generic/392 fails because the bytes used in the file used to test
   hole punch recovery are less after the log replay.  This is
   because the blocks written and then punched out are only freed
   with a delay due to the logging mechanism.
 - xfs/170 will fail as the already fragile file streams mechanism
   doesn't seem to interact well with the COW allocator
 - xfs/180 xfs/182 xfs/192 xfs/198 xfs/204 and xfs/208 will claim
   the file system is badly fragmented, but there is not much we
   can do to avoid that when always writing out of place
 - xfs/205 fails because overwriting a file in always_cow mode
   will require new space allocation and the assumption in the
   test thus don't work anymore.
 - xfs/326 fails to modify the file at all in always_cow mode after
   injecting the refcount error, leading to an unexpected md5sum
   after the remount, but that again is expected

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: report IOMAP_F_SHARED from xfs_file_iomap_begin_delay
Christoph Hellwig [Mon, 18 Feb 2019 17:38:48 +0000 (09:38 -0800)]
xfs: report IOMAP_F_SHARED from xfs_file_iomap_begin_delay

No user of it in the iomap code at the moment, but we should not
actively report wrong information if we can trivially get it right.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: make COW fork unwritten extent conversions more robust
Christoph Hellwig [Mon, 18 Feb 2019 17:38:48 +0000 (09:38 -0800)]
xfs: make COW fork unwritten extent conversions more robust

If we have racing buffered and direct I/O COW fork extents under
writeback can have been moved to the data fork by the time we call
xfs_reflink_convert_cow from xfs_submit_ioend.  This would be mostly
harmless as the block numbers don't change by this move, except for
the fact that xfs_bmapi_write will crash or trigger asserts when
not finding existing extents, even despite trying to paper over this
with the XFS_BMAPI_CONVERT_ONLY flag.

Instead of special casing non-transaction conversions in the already
way too complicated xfs_bmapi_write just add a new helper for the much
simpler non-transactional COW fork case, which simplify ignores not
found extents.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: merge COW handling into xfs_file_iomap_begin_delay
Christoph Hellwig [Mon, 18 Feb 2019 17:38:47 +0000 (09:38 -0800)]
xfs: merge COW handling into xfs_file_iomap_begin_delay

Besides simplifying the code a bit this allows to actually implement
the behavior of using COW preallocation for non-COW data mentioned
in the current comments.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: also truncate holes covered by COW blocks
Christoph Hellwig [Mon, 18 Feb 2019 17:38:47 +0000 (09:38 -0800)]
xfs: also truncate holes covered by COW blocks

This only matters if we want to write data through the COW fork that is
not actually an overwrite of existing data.  Reasons for that are
speculative COW fork allocations using the cowextsize, or a mode where
we always write through the COW fork.  Currently both can't actually
happen, but I plan to enable them.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: don't use delalloc extents for COW on files with extsize hints
Christoph Hellwig [Mon, 18 Feb 2019 17:38:46 +0000 (09:38 -0800)]
xfs: don't use delalloc extents for COW on files with extsize hints

While using delalloc for extsize hints is generally a good idea, the
current code that does so only for COW doesn't help us much and creates
a lot of special cases.  Switch it to use real allocations like we
do for direct I/O.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: fix SEEK_DATA for speculative COW fork preallocation
Christoph Hellwig [Mon, 18 Feb 2019 17:38:46 +0000 (09:38 -0800)]
xfs: fix SEEK_DATA for speculative COW fork preallocation

We speculatively allocate extents in the COW fork to reduce
fragmentation.  But when we write data into such COW fork blocks that
do now shadow an allocation in the data fork SEEK_DATA will not
correctly report it, as it only looks at the data fork extents.
The only reason why that hasn't been an issue so far is because
we even use these speculative COW fork preallocations over holes in
the data fork at all for buffered writes, and blocks in the COW
fork that are written by direct writes are moved into the data
fork immediately at I/O completion time.

Add a new set of iomap_ops for SEEK_HOLE/SEEK_DATA which looks into
both the COW and data fork, and reports all COW extents as unwritten
to the iomap layer.  While this isn't strictly true for COW fork
extents that were already converted to real extents, the practical
semantics that you can't read data from them until they are moved
into the data fork are very similar, and this will force the iomap
layer into probing the extents for actually present data.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: make xfs_bmbt_to_iomap more useful
Christoph Hellwig [Mon, 18 Feb 2019 17:38:46 +0000 (09:38 -0800)]
xfs: make xfs_bmbt_to_iomap more useful

Move checking for invalid zero blocks and setting of various iomap flags
into this helper.  Also make it deal with "raw" delalloc extents to
avoid clutter in the callers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: fix xfs_buf magic number endian checks
Darrick J. Wong [Sat, 16 Feb 2019 19:47:28 +0000 (11:47 -0800)]
xfs: fix xfs_buf magic number endian checks

Create a separate magic16 check function so that we don't run afoul of
static checkers.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: retry COW fork delalloc conversion when no extent was found
Christoph Hellwig [Fri, 15 Feb 2019 16:02:50 +0000 (08:02 -0800)]
xfs: retry COW fork delalloc conversion when no extent was found

While we can only truncate a block under the page lock for the current
page, there is no high-level synchronization for moving extents from the
COW to the data fork.  This means that for example we can have another
thread doing a direct I/O completion that moves extents from the COW to
the data fork race with writeback.  While this race is very hard to hit
the always_cow seems to reproduce it reasonably well, and it also exists
without that.  Because of that there is a chance that a delalloc
conversion for the COW fork might not find any extents to convert.  In
that case we should retry the whole block lookup and now find the blocks
in the data fork.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: remove the truncate short cut in xfs_map_blocks
Christoph Hellwig [Fri, 15 Feb 2019 16:02:50 +0000 (08:02 -0800)]
xfs: remove the truncate short cut in xfs_map_blocks

Now that we properly handle the race with truncate in the delalloc
allocator there is no need to short cut this exceptional case earlier
on.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: move xfs_iomap_write_allocate to xfs_aops.c
Christoph Hellwig [Fri, 15 Feb 2019 16:02:49 +0000 (08:02 -0800)]
xfs: move xfs_iomap_write_allocate to xfs_aops.c

This function is a small wrapper only used by the writeback code, so
move it together with the writeback code and simplify it down to the
glorified do { } while loop that is now is.

A few bits intentionally got lost here: no need to call xfs_qm_dqattach
because quotas are always attached when we create the delalloc
reservation, and no need for the imap->br_startblock == 0 check given
that xfs_bmapi_convert_delalloc already has a WARN_ON_ONCE for exactly
that condition.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: move stat accounting to xfs_bmapi_convert_delalloc
Christoph Hellwig [Fri, 15 Feb 2019 16:02:49 +0000 (08:02 -0800)]
xfs: move stat accounting to xfs_bmapi_convert_delalloc

This way we can actually count how many bytes got converted and how many
calls we need, unlike in the caller which doesn't have the detailed
view.

Note that this includes a slight change in behavior as the
xs_xstrat_quick is now bumped for every allocation instead of just the
one covering the requested writeback offset, which makes a lot more
sense.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: move transaction handling to xfs_bmapi_convert_delalloc
Christoph Hellwig [Fri, 15 Feb 2019 16:02:49 +0000 (08:02 -0800)]
xfs: move transaction handling to xfs_bmapi_convert_delalloc

No need to deal with the transaction and the inode locking in the
caller. Note that we also switch to passing whichfork as the second
paramter, matching what most related functions do.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: split XFS_BMAPI_DELALLOC handling from xfs_bmapi_write
Christoph Hellwig [Fri, 15 Feb 2019 16:02:48 +0000 (08:02 -0800)]
xfs: split XFS_BMAPI_DELALLOC handling from xfs_bmapi_write

Delalloc conversion has traditionally been part of our function to
allocate blocks on disk (first xfs_bmapi, then xfs_bmapi_write), but
delalloc conversion is a little special as we really do not want
to allocate blocks over holes, for which we don't have reservations.

Split the delalloc conversions into a separate helper to keep the
code simple and structured.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: factor out two helpers from xfs_bmapi_write
Christoph Hellwig [Fri, 15 Feb 2019 16:02:48 +0000 (08:02 -0800)]
xfs: factor out two helpers from xfs_bmapi_write

We want to be able to reuse them for the upcoming dedidcated delalloc
convert routine.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: simplify the xfs_bmap_btree_to_extents calling conventions
Christoph Hellwig [Fri, 15 Feb 2019 16:02:47 +0000 (08:02 -0800)]
xfs: simplify the xfs_bmap_btree_to_extents calling conventions

Move boilerplate code from the callers into xfs_bmap_btree_to_extents:

 - exit early without failure if we don't need to convert to the
   extent format
 - assert that we have a btree cursor
 - don't reinitialize the passed in logflags argument

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: remove the s_maxbytes checks in xfs_map_blocks
Christoph Hellwig [Fri, 15 Feb 2019 16:02:47 +0000 (08:02 -0800)]
xfs: remove the s_maxbytes checks in xfs_map_blocks

We already ensure all data fits into s_maxbytes in the write / fault
path.  The only reason we have them here is that they were copy and
pasted from xfs_bmapi_read when we stopped using that function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: remove the io_type field from the writeback context and ioend
Christoph Hellwig [Fri, 15 Feb 2019 16:02:46 +0000 (08:02 -0800)]
xfs: remove the io_type field from the writeback context and ioend

The io_type field contains what is basically a summary of information
from the inode fork and the imap.  But we can just as easily use that
information directly, simplifying a few bits here and there and
improving the trace points.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: don't ever put nlink > 0 inodes on the unlinked list
Darrick J. Wong [Wed, 13 Feb 2019 19:15:17 +0000 (11:15 -0800)]
xfs: don't ever put nlink > 0 inodes on the unlinked list

When XFS creates an O_TMPFILE file, the inode is created with nlink = 1,
put on the unlinked list, and then the VFS sets nlink = 0 in d_tmpfile.
If we crash before anything logs the inode (it's dirty incore but the
vfs doesn't tell us it's dirty so we never log that change), the iunlink
processing part of recovery will then explode with a pile of:

XFS: Assertion failed: VFS_I(ip)->i_nlink == 0, file:
fs/xfs/xfs_log_recover.c, line: 5072

Worse yet, since nlink is nonzero, the inodes also don't get cleaned up
and they just leak until the next xfs_repair run.

Therefore, change xfs_iunlink to require that inodes being put on the
unlinked list have nlink == 0, change the tmpfile callers to instantiate
nodes that way, and set the nlink to 1 just prior to calling d_tmpfile.
Fix the comment for xfs_iunlink while we're at it.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2 years agoxfs: reserve blocks for ifree transaction during log recovery
Darrick J. Wong [Wed, 13 Feb 2019 19:46:16 +0000 (11:46 -0800)]
xfs: reserve blocks for ifree transaction during log recovery

Log recovery frees all the inodes stored in the unlinked list, which can
cause expansion of the free inode btree.  The ifree code skips block
reservations if it thinks there's a per-AG space reservation, but we
don't set up the reservation until after log recovery, which means that
a finobt expansion blows up in xfs_trans_mod_sb when we exceed the
transaction's block reservation.

To fix this, we set the "no finobt reservation" flag to true when we
create the xfs_mount and only set it to false if we confirm that every
AG had enough free space to put aside for the finobt.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
2 years agoxfs: rename m_inotbt_nores to m_finobt_nores
Darrick J. Wong [Thu, 14 Feb 2019 17:33:15 +0000 (09:33 -0800)]
xfs: rename m_inotbt_nores to m_finobt_nores

Rename this flag variable to imply more strongly that it's related to
the free inode btree (finobt) operation.  No functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
2 years agoxfs: don't overflow xattr listent buffer
Darrick J. Wong [Wed, 13 Feb 2019 19:15:17 +0000 (11:15 -0800)]
xfs: don't overflow xattr listent buffer

For VFS listxattr calls, xfs_xattr_put_listent calls
__xfs_xattr_put_listent twice if it sees an attribute
"trusted.SGI_ACL_FILE": once for that name, and again for
"system.posix_acl_access".  Unfortunately, if we happen to run out of
buffer space while emitting the first name, we set count to -1 (so that
we can feed ERANGE to the caller).  The second invocation doesn't check that
the context parameters make sense and overwrites the byte before the
buffer, triggering a KASAN report:

==================================================================
BUG: KASAN: slab-out-of-bounds in strncpy+0xb3/0xd0
Write of size 1 at addr ffff88807fbd317f by task syz/1113

CPU: 3 PID: 1113 Comm: syz Not tainted 5.0.0-rc6-xfsx #rc6
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-1ubuntu1 04/01/2014
Call Trace:
 dump_stack+0xcc/0x180
 print_address_description+0x6c/0x23c
 kasan_report.cold.3+0x1c/0x35
 strncpy+0xb3/0xd0
 __xfs_xattr_put_listent+0x1a9/0x2c0 [xfs]
 xfs_attr_list_int_ilocked+0x11af/0x1800 [xfs]
 xfs_attr_list_int+0x20c/0x2e0 [xfs]
 xfs_vn_listxattr+0x225/0x320 [xfs]
 listxattr+0x11f/0x1b0
 path_listxattr+0xbd/0x130
 do_syscall_64+0x139/0x560

While we're at it we add an assert to the other put_listent to avoid
this sort of thing ever happening to the attrlist_by_handle code.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2 years agoxfs: compile time offset checks for common v4/v5 metadata
Brian Foster [Fri, 8 Feb 2019 17:02:29 +0000 (09:02 -0800)]
xfs: compile time offset checks for common v4/v5 metadata

The v5 superblock format added various metadata fields (such as crc,
metadata lsn, owner uuid, etc.) to v4 metadata headers or created
new v5 headers for blocks where no such headers existed on v4. Where
v4 headers did exist, the v5 structures are careful to place v4
metadata at the original location. For example, the magic value is
expected to be at the same location in certain blocks to facilitate
version detection.

While failure of this invariant is likely to cause severe and
obvious problems at runtime, we can detect this condition at compile
time via the more recently added on-disk format check
infrastructure. Since there is no runtime cost, add some offset
checks that start with v5 structure definitions, traverse down to
the first bit of common metadata with v4 and ensure that common
metadata is at the expected offset. Note that we don't care about
blocks which had no v4 header because there is no common metadata in
those cases. No functional changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: use buf ops magic to detect btree block type
Darrick J. Wong [Wed, 6 Feb 2019 18:20:54 +0000 (10:20 -0800)]
xfs: use buf ops magic to detect btree block type

Now that we encode block magic numbers in all the buffer ops, use that
for block type detection in the ag header repair code instead of
encoding magics directly in the repair code.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: add magic numbers to dquot buffer ops
Darrick J. Wong [Wed, 6 Feb 2019 18:05:04 +0000 (10:05 -0800)]
xfs: add magic numbers to dquot buffer ops

Add dquot magic numbers to the buffer ops type, in case we ever want to
use them.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: add inode magic to inode verifier
Darrick J. Wong [Wed, 6 Feb 2019 18:01:50 +0000 (10:01 -0800)]
xfs: add inode magic to inode verifier

Use xfs_verify_magic to check the magic numbers of inodes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: factor xfs_da3_blkinfo verification into common helper
Brian Foster [Thu, 7 Feb 2019 18:45:48 +0000 (10:45 -0800)]
xfs: factor xfs_da3_blkinfo verification into common helper

With the verifier magic value helper in place, we've left a bit more
duplicate code across the verifiers that involve struct
xfs_da3_blkinfo. This includes the da node, xattr leaf and dir leaf
verifiers, all of which perform similar checks for v4 and v5
filesystems.

Create a common helper to verify an xfs_da3_blkinfo structure,
taking care to only access v5 fields where appropriate, and refactor
the aforementioned verifiers to use the helper. No functional
changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: miscellaneous verifier magic value fixups
Brian Foster [Thu, 7 Feb 2019 18:45:48 +0000 (10:45 -0800)]
xfs: miscellaneous verifier magic value fixups

Most buffer verifiers have hardcoded magic value checks
conditionalized on the version of the filesystem. The magic value
field of the verifier structure facilitates abstraction of some of
this code. Populate the ->magic field of various verifiers to take
advantage of this abstraction. No functional changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: use verifier magic field in dir2 leaf verifiers
Brian Foster [Thu, 7 Feb 2019 18:45:47 +0000 (10:45 -0800)]
xfs: use verifier magic field in dir2 leaf verifiers

The dir2 leaf verifiers share the same underlying structure
verification code, but implement six accessor functions to multiplex
the code across the two verifiers. Further, the magic value isn't
sufficiently abstracted such that the common helper has to manually
fix up the magic from the caller on v5 filesystems.

Use the magic field in the verifier structure to eliminate the
duplicate code and clean this all up. No functional change.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: distinguish between bnobt and cntbt magic values
Brian Foster [Thu, 7 Feb 2019 18:45:47 +0000 (10:45 -0800)]
xfs: distinguish between bnobt and cntbt magic values

The allocation btree verifiers share code that is unable to detect
cross-tree magic value corruptions such as a bnobt block with a
cntbt magic value. Populate the b_ops->magic field of the associated
verifier structures such that the structure verifier can check the
magic value against the expected value based on tree type.

The btree level check requires knowledge of the tree type to
determine the appropriate maximum value. This was previously part of
the hardcoded magic value checks. With that code removed, peek at
the first magic value in the verifier to determine the expected tree
type of the current block.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: split up allocation btree verifier
Brian Foster [Thu, 7 Feb 2019 18:45:47 +0000 (10:45 -0800)]
xfs: split up allocation btree verifier

Similar to the inode btree verifier, the same allocation btree
verifier structure is shared between the by-bno (bnobt) and by-size
(cntbt) btrees. This prevents the ability to distinguish magic
values between them. Separate the verifier into two, one for each
tree, and assign them appropriately. No functional changes.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: distinguish between inobt and finobt magic values
Brian Foster [Thu, 7 Feb 2019 18:45:46 +0000 (10:45 -0800)]
xfs: distinguish between inobt and finobt magic values

The inode btree verifier code is shared between the inode btree and
free inode btree because the underlying metadata formats are
essentially equivalent. A side effect of this is that the verifier
cannot determine whether a particular btree block should have an
inobt or finobt magic value.

This logic allows an unfortunate xfs_repair bug to escape detection
where certain level > 0 nodes of the finobt are stamped with inobt
magic by xfs_repair finobt reconstruction. This is fortunately not a
severe problem since the inode btree magic values do not contribute
to any changes in kernel behavior, but we do need a means to detect
and prevent this problem in the future.

Add a field to xfs_buf_ops to store the v4 and v5 superblock magic
values expected by a particular verifier. Add a helper to check an
on-disk magic value against the value expected by the verifier. Call
the helper from the shared [f]inobt verifier code for magic value
verification. This ensures that the inode btree blocks each have the
appropriate magic value based on specific tree type and superblock
version.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: create a separate finobt verifier
Brian Foster [Thu, 7 Feb 2019 18:45:46 +0000 (10:45 -0800)]
xfs: create a separate finobt verifier

The inobt verifier is reused for the inobt and finobt, which
prevents the ability to distinguish between magic values on a
per-tree basis. Create a separate finobt structure in preparation
for changes to enforce the appropriate magic value for the
associated tree. This patch has no functional change.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: always check magic values in on-disk byte order
Brian Foster [Thu, 7 Feb 2019 18:45:45 +0000 (10:45 -0800)]
xfs: always check magic values in on-disk byte order

Most verifiers that check on-disk magic values convert the CPU
endian magic value constant to disk endian to facilitate compile
time optimization of the byte swap and reduce the need for runtime
byte swaps in buffer verifiers. Several buffer verifiers do not
follow this pattern. Update those verifiers for consistency.

Also fix up a random typo in the inode readahead verifier name.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: clarify documentation for the function to reverify buffers
Brian Foster [Wed, 6 Feb 2019 17:25:29 +0000 (09:25 -0800)]
xfs: clarify documentation for the function to reverify buffers

Improve the documentation around xfs_buf_ensure_ops, which is the
function that is responsible for cleaning up the b_ops state of buffers
that go through xrep_findroot_block but don't match anything.  Rename
the function to xfs_buf_reverify.

[darrick: this started off as bfoster mods of a previous patch of mine,
but the renaming part is now this separate patch.]

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: cache unlinked pointers in an rhashtable
Darrick J. Wong [Thu, 7 Feb 2019 18:37:16 +0000 (10:37 -0800)]
xfs: cache unlinked pointers in an rhashtable

Use a rhashtable to cache the unlinked list incore.  This should speed
up unlinked processing considerably when there are a lot of inodes on
the unlinked list because iunlink_remove no longer has to traverse an
entire bucket list to find which inode points to the one being removed.

The incore list structure records "X.next_unlinked = Y" relations, with
the rhashtable using Y to index the records.  This makes finding the
inode X that points to a inode Y very quick.  If our cache fails to find
anything we can always fall back on the old method.

FWIW this drastically reduces the amount of time it takes to remove
inodes from the unlinked list.  I wrote a program to open a lot of
O_TMPFILE files and then close them in the same order, which takes
a very long time if we have to traverse the unlinked lists.  With the
ptach, I see:

+ /d/t/tmpfile/tmpfile
Opened 193531 files in 6.33s.
Closed 193531 files in 5.86s

real    0m12.192s
user    0m0.064s
sys     0m11.619s
+ cd /
+ umount /mnt

real    0m0.050s
user    0m0.004s
sys     0m0.030s

And without the patch:

+ /d/t/tmpfile/tmpfile
Opened 193588 files in 6.35s.
Closed 193588 files in 751.61s

real    12m38.853s
user    0m0.084s
sys     12m34.470s
+ cd /
+ umount /mnt

real    0m0.086s
user    0m0.000s
sys     0m0.060s

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: add tracepoints for high level iunlink operations
Darrick J. Wong [Thu, 7 Feb 2019 18:37:16 +0000 (10:37 -0800)]
xfs: add tracepoints for high level iunlink operations

Add tracepoints so we can associate high level operations with low level
updates.  No functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: refactor inode update in iunlink_remove
Darrick J. Wong [Thu, 7 Feb 2019 18:37:15 +0000 (10:37 -0800)]
xfs: refactor inode update in iunlink_remove

In xfs_iunlink_remove we have two identical calls to
xfs_iunlink_update_inode, so move it out of the if statement to simplify
the code some more.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: refactor unlinked list search and mapping to a separate function
Darrick J. Wong [Thu, 7 Feb 2019 18:37:15 +0000 (10:37 -0800)]
xfs: refactor unlinked list search and mapping to a separate function

There's a loop that searches an unlinked bucket list to find the inode
that points to a given inode.  Hoist this into a separate function;
later we'll use our iunlink backref cache to bypass the slow list
operation.  No functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: refactor inode unlinked pointer update functions
Darrick J. Wong [Thu, 7 Feb 2019 18:37:15 +0000 (10:37 -0800)]
xfs: refactor inode unlinked pointer update functions

Hoist the functions that update an inode's unlinked pointer updates into
a helper.  No functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: strengthen AGI unlinked inode bucket pointer checks
Darrick J. Wong [Thu, 7 Feb 2019 18:37:14 +0000 (10:37 -0800)]
xfs: strengthen AGI unlinked inode bucket pointer checks

Strengthen our checking of the AGI unlinked pointers when we start to
use them for updating the metadata.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: refactor AGI unlinked bucket updates
Darrick J. Wong [Thu, 7 Feb 2019 18:37:14 +0000 (10:37 -0800)]
xfs: refactor AGI unlinked bucket updates

Split the AGI unlinked bucket updates into a separate function.  No
functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2 years agoxfs: add xfs_verify_agino_or_null helper
Darrick J. Wong [Thu, 7 Feb 2019 18:37:13 +0000 (10:37 -0800)]
xfs: add xfs_verify_agino_or_null helper

Add a new helper to check that a per-AG inode pointer is either null or
points somewhere valid within that AG.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2 years agoxfs: clean up iunlink functions
Darrick J. Wong [Thu, 7 Feb 2019 18:37:13 +0000 (10:37 -0800)]
xfs: clean up iunlink functions

Fix some indentation issues with the iunlink functions and reorganize
the tops of the functions to be identical.  No functional changes.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
2 years agoxfs: use the latest extent at writeback delalloc conversion time
Brian Foster [Fri, 1 Feb 2019 17:14:24 +0000 (09:14 -0800)]
xfs: use the latest extent at writeback delalloc conversion time

The writeback delalloc conversion code is racy with respect to
changes in the currently cached file mapping outside of the current
page. This is because the ilock is cycled between the time the
caller originally looked up the mapping and across each real
allocation of the provided file range. This code has collected
various hacks over the years to help combat the symptoms of these
races (i.e., truncate race detection, allocation into hole
detection, etc.), but none address the fundamental problem that the
imap may not be valid at allocation time.

Rather than continue to use race detection hacks, update writeback
delalloc conversion to a model that explicitly converts the delalloc
extent backing the current file offset being processed. The current
file offset is the only block we can trust to remain once the ilock
is dropped because any operation that can remove the block
(truncate, hole punch, etc.) must flush and discard pagecache pages
first.

Modify xfs_iomap_write_allocate() to use the xfs_bmapi_delalloc()
mechanism to request allocation of the entire delalloc extent
backing the current offset instead of assuming the extent passed by
the caller is unchanged. Record the range specified by the caller
and apply it to the resulting allocated extent so previous checks by
the caller for COW fork overlap are not lost. Finally, overload the
bmapi delalloc flag with the range reval flag behavior since this is
the only use case for both.

This ensures that writeback always picks up the correct
and current extent associated with the page, regardless of races
with other extent modifying operations. If operating on a data fork
and the COW overlap state has changed since the ilock was cycled,
the caller revalidates against the COW fork sequence number before
using the imap for the next block.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: create delalloc bmapi wrapper for full extent allocation
Brian Foster [Fri, 1 Feb 2019 17:14:23 +0000 (09:14 -0800)]
xfs: create delalloc bmapi wrapper for full extent allocation

The writeback delalloc conversion code is racy with respect to
changes in the currently cached file mapping. This stems from the
fact that the bmapi allocation code requires a file range to
allocate and the writeback conversion code assumes the range of the
currently cached mapping is still valid with respect to the fork. It
may not be valid, however, because the ilock is cycled (potentially
multiple times) between the time the cached mapping was populated
and the delalloc conversion occurs.

To facilitate a solution to this problem, create a new
xfs_bmapi_delalloc() wrapper to xfs_bmapi_write() that takes a file
(FSB) offset and attempts to allocate whatever delalloc extent backs
the offset. Use a new bmapi flag to cause xfs_bmapi_write() to set
the range based on the extent backing the bno parameter unless bno
lands in a hole. If bno does land in a hole, fall back to the
current behavior (which may result in an error or quietly skipping
holes in the specified range depending on other parameters). This
patch does not change behavior.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: remove superfluous writeback mapping eof trimming
Brian Foster [Fri, 1 Feb 2019 17:14:23 +0000 (09:14 -0800)]
xfs: remove superfluous writeback mapping eof trimming

Now that the cached writeback mapping is explicitly invalidated on
data fork changes, the EOF trimming band-aid is no longer necessary.
Remove xfs_trim_extent_eof() as well since it has no other users.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: validate writeback mapping using data fork seq counter
Brian Foster [Fri, 1 Feb 2019 17:14:23 +0000 (09:14 -0800)]
xfs: validate writeback mapping using data fork seq counter

The writeback code caches the current extent mapping across multiple
xfs_do_writepage() calls to avoid repeated lookups for sequential
pages backed by the same extent. This is known to be slightly racy
with extent fork changes in certain difficult to reproduce
scenarios. The cached extent is trimmed to within EOF to help avoid
the most common vector for this problem via speculative
preallocation management, but this is a band-aid that does not
address the fundamental problem.

Now that we have an xfs_ifork sequence counter mechanism used to
facilitate COW writeback, we can use the same mechanism to validate
consistency between the data fork and cached writeback mappings. On
its face, this is somewhat of a big hammer approach because any
change to the data fork invalidates any mapping currently cached by
a writeback in progress regardless of whether the data fork change
overlaps with the range under writeback. In practice, however, the
impact of this approach is minimal in most cases.

First, data fork changes (delayed allocations) caused by sustained
sequential buffered writes are amortized across speculative
preallocations. This means that a cached mapping won't be
invalidated by each buffered write of a common file copy workload,
but rather only on less frequent allocation events. Second, the
extent tree is always entirely in-core so an additional lookup of a
usable extent mostly costs a shared ilock cycle and in-memory tree
lookup. This means that a cached mapping reval is relatively cheap
compared to the I/O itself. Third, spurious invalidations don't
impact ioend construction. This means that even if the same extent
is revalidated multiple times across multiple writepage instances,
we still construct and submit the same size ioend (and bio) if the
blocks are physically contiguous.

Update struct xfs_writepage_ctx with a new field to hold the
sequence number of the data fork associated with the currently
cached mapping. Check the wpc seqno against the data fork when the
mapping is validated and reestablish the mapping whenever the fork
has changed since the mapping was cached. This ensures that
writeback always uses a valid extent mapping and thus prevents lost
writebacks and stale delalloc block problems.

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 <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: update fork seq counter on data fork changes
Brian Foster [Fri, 1 Feb 2019 17:14:22 +0000 (09:14 -0800)]
xfs: update fork seq counter on data fork changes

The sequence counter in the xfs_ifork structure is only updated on
COW forks. This is because the counter is currently only used to
optimize out repetitive COW fork checks at writeback time.

Tweak the extent code to update the seq counter regardless of the
fork type in preparation for using this counter on data forks as
well.

Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: Introduce XFS_PTAG_VERIFIER_ERROR panic mask
Marco Benatto [Fri, 1 Feb 2019 17:12:20 +0000 (09:12 -0800)]
xfs: Introduce XFS_PTAG_VERIFIER_ERROR panic mask

Currently we have a few PTAGs in place allowing us to transform a filesystem
error in a BUG() call.  However, we don't have a panic tag for corrupt
metadata, so introduce XFS_PTAG_VERIFIER_ERROR so that the administrator can
use the fs.xfs.panic_mask sysctl knob to convert any error detected by buffer
verifiers into a kernel panic.

Signed-off-by: Marco Benatto <mbenatto@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
[darrick: light editing of commit message]
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: remove duplicated xfs_defer.h
YueHaibing [Fri, 1 Feb 2019 17:09:50 +0000 (09:09 -0800)]
xfs: remove duplicated xfs_defer.h

Remove duplicated include.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2 years agoxfs: check attribute name validity
Darrick J. Wong [Fri, 1 Feb 2019 17:08:54 +0000 (09:08 -0800)]
xfs: check attribute name validity

Check extended attribute entry names for invalid characters.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: check directory name validity
Darrick J. Wong [Fri, 1 Feb 2019 17:08:54 +0000 (09:08 -0800)]
xfs: check directory name validity

Check directory entry names for invalid characters.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: fix off-by-one error in rtbitmap cross-reference
Darrick J. Wong [Fri, 1 Feb 2019 17:08:53 +0000 (09:08 -0800)]
xfs: fix off-by-one error in rtbitmap cross-reference

Fix an off-by-one error in the realtime bitmap "is used" cross-reference
helper function if the realtime extent size is a single block.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: scrub should flag dir/attr offsets that aren't mappable with xfs_dablk_t
Darrick J. Wong [Fri, 1 Feb 2019 17:08:53 +0000 (09:08 -0800)]
xfs: scrub should flag dir/attr offsets that aren't mappable with xfs_dablk_t

Teach scrub to flag extent maps that exceed the range that can be mapped
with a xfs_dablk_t.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: abort xattr scrub if fatal signals are pending
Darrick J. Wong [Fri, 1 Feb 2019 17:08:52 +0000 (09:08 -0800)]
xfs: abort xattr scrub if fatal signals are pending

The extended attribute scrubber should abort the "read all attrs" loop
if there's a fatal signal pending on the process.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: consolidate scrub dinode mapping code into a single function
Darrick J. Wong [Fri, 1 Feb 2019 17:08:52 +0000 (09:08 -0800)]
xfs: consolidate scrub dinode mapping code into a single function

Move all the confusing dinode mapping code that's split between
xchk_iallocbt_check_cluster and xchk_iallocbt_check_cluster_ifree into
the first function so that it's clearer how we find the dinode for a
given inode.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: scrub big block inode btrees correctly
Darrick J. Wong [Fri, 1 Feb 2019 17:08:52 +0000 (09:08 -0800)]
xfs: scrub big block inode btrees correctly

Teach scrub how to handle the case that there are one or more inobt
records covering a given inode cluster.  This fixes the operation on big
block filesystems (e.g. 64k blocks, 512 byte inodes).

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: clean up the inode cluster checking in the inobt scrub
Darrick J. Wong [Fri, 1 Feb 2019 17:08:51 +0000 (09:08 -0800)]
xfs: clean up the inode cluster checking in the inobt scrub

The code to check inobt records against inode clusters is a mess of
poorly named variables and unnecessary parameters.  Clean the
unnecessary inode number parameters out of _check_cluster_freemask in
favor of computing them inside the function instead of making the caller
do it.  In xchk_iallocbt_check_cluster, rename the variables to make it
more obvious just what chunk_ino and cluster_ino represent.

Add a tracepoint to make it easier to track each inode cluster as we
scrub it.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: hoist inode cluster checks out of loop
Darrick J. Wong [Fri, 1 Feb 2019 17:08:51 +0000 (09:08 -0800)]
xfs: hoist inode cluster checks out of loop

Hoist the inode cluster checks out of the inobt record check loop into
a separate function in preparation for refactoring of that loop.  No
functional changes here; that's in the next patch.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: check inobt record alignment on big block filesystems
Darrick J. Wong [Fri, 1 Feb 2019 17:08:50 +0000 (09:08 -0800)]
xfs: check inobt record alignment on big block filesystems

On a big block filesystem, there may be multiple inobt records covering
a single inode cluster.  These records obviously won't be aligned to
cluster alignment rules, and they must cover the entire cluster.  Teach
scrub to check for these things.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: check the ir_startino alignment directly
Darrick J. Wong [Fri, 1 Feb 2019 17:08:50 +0000 (09:08 -0800)]
xfs: check the ir_startino alignment directly

In xchk_iallocbt_rec, check the alignment of ir_startino by converting
the inode cluster block alignment into units of inodes instead of the
other way around (converting ir_startino to blocks).  This prevents us
from tripping over off-by-one errors in ir_startino which are obscured
by the inode -> block conversion.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoxfs: never try to scrub more than 64 inodes per inobt record
Darrick J. Wong [Fri, 1 Feb 2019 17:08:49 +0000 (09:08 -0800)]
xfs: never try to scrub more than 64 inodes per inobt record

Make sure we never check more than XFS_INODES_PER_CHUNK inodes for any
given inobt record since there can be more than one inobt record mapped
to an inode cluster.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
2 years agoLinux 5.0-rc6 v5.0-rc6
Linus Torvalds [Sun, 10 Feb 2019 22:42:20 +0000 (14:42 -0800)]
Linux 5.0-rc6

2 years agoMerge tag 'dmaengine-fix-5.0-rc6' of git://git.infradead.org/users/vkoul/slave-dma
Linus Torvalds [Sun, 10 Feb 2019 18:39:37 +0000 (10:39 -0800)]
Merge tag 'dmaengine-fix-5.0-rc6' of git://git.infradead.org/users/vkoul/slave-dma

Pull dmaengine fixes from Vinod Koul:
 - Fix in at_xdmac fr wrongful channel state
 - Fix for imx driver for wrong callback invocation
 - Fix to bcm driver for interrupt race & transaction abort.
 - Fix in dmatest to abort in mapping error

* tag 'dmaengine-fix-5.0-rc6' of git://git.infradead.org/users/vkoul/slave-dma:
  dmaengine: dmatest: Abort test in case of mapping error
  dmaengine: bcm2835: Fix abort of transactions
  dmaengine: bcm2835: Fix interrupt race on RT
  dmaengine: imx-dma: fix wrong callback invoke
  dmaengine: at_xdmac: Fix wrongfull report of a channel as in use

2 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 10 Feb 2019 17:57:42 +0000 (09:57 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "A handful of fixes:

   - Fix an MCE corner case bug/crash found via MCE injection testing

   - Fix 5-level paging boot crash

   - Fix MCE recovery cache invalidation bug

   - Fix regression on Xen guests caused by a recent PMD level mremap
     speedup optimization"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mm: Make set_pmd_at() paravirt aware
  x86/mm/cpa: Fix set_mce_nospec()
  x86/boot/compressed/64: Do not corrupt EDX on EFER.LME=1 setting
  x86/MCE: Initialize mce.bank in the case of a fatal error in mce_no_way_out()

2 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 10 Feb 2019 17:54:19 +0000 (09:54 -0800)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Ingo Molnar:
 "irqchip driver fixes: most of them are race fixes for ARM GIC (General
  Interrupt Controller) variants, but also a fix for the ARM MMP
  (Marvell PXA168 et al) irqchip affecting OLPC keyboards"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3-its: Fix ITT_entry_size accessor
  irqchip/mmp: Only touch the PJ4 IRQ & FIQ bits on enable/disable
  irqchip/gic-v3-its: Gracefully fail on LPI exhaustion
  irqchip/gic-v3-its: Plug allocation race for devices sharing a DevID
  irqchip/gic-v4: Fix occasional VLPI drop

2 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 10 Feb 2019 17:48:18 +0000 (09:48 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "A couple of kernel side fixes:

   - Fix the Intel uncore driver on certain hardware configurations

   - Fix a CPU hotplug related memory allocation bug

   - Remove a spurious WARN()

  ... plus also a handful of perf tooling fixes"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf script python: Add Python3 support to tests/attr.py
  perf trace: Support multiple "vfs_getname" probes
  perf symbols: Filter out hidden symbols from labels
  perf symbols: Add fallback definitions for GELF_ST_VISIBILITY()
  tools headers uapi: Sync linux/in.h copy from the kernel sources
  perf clang: Do not use 'return std::move(something)'
  perf mem/c2c: Fix perf_mem_events to support powerpc
  perf tests evsel-tp-sched: Fix bitwise operator
  perf/core: Don't WARN() for impossible ring-buffer sizes
  perf/x86/intel: Delay memory deallocation until x86_pmu_dead_cpu()
  perf/x86/intel/uncore: Add Node ID mask

2 years agoMerge branch 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 10 Feb 2019 17:44:52 +0000 (09:44 -0800)]
Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fixes from Ingo Molnar:
 "An rtmutex (PI-futex) deadlock scenario fix, plus a locking
  documentation fix"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  futex: Handle early deadlock return correctly
  futex: Fix barrier comment

2 years agox86/mm: Make set_pmd_at() paravirt aware
Juergen Gross [Sun, 10 Feb 2019 07:40:56 +0000 (08:40 +0100)]
x86/mm: Make set_pmd_at() paravirt aware

set_pmd_at() calls native_set_pmd() unconditionally on x86. This was
fine as long as only huge page entries were written via set_pmd_at(),
as Xen pv guests don't support those.

Commit 2c91bd4a4e2e53 ("mm: speed up mremap by 20x on large regions")
introduced a usage of set_pmd_at() possible on pv guests, leading to
failures like:

BUG: unable to handle kernel paging request at ffff888023e26778
#PF error: [PROT] [WRITE]
RIP: e030:move_page_tables+0x7c1/0xae0
move_vma.isra.3+0xd1/0x2d0
__se_sys_mremap+0x3c6/0x5b0
 do_syscall_64+0x49/0x100
entry_SYSCALL_64_after_hwframe+0x44/0xa9

Make set_pmd_at() paravirt aware by just letting it use set_pmd().

Fixes: 2c91bd4a4e2e53 ("mm: speed up mremap by 20x on large regions")
Reported-by: Sander Eikelenboom <linux@eikelenboom.it>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: xen-devel@lists.xenproject.org
Cc: boris.ostrovsky@oracle.com
Cc: sstabellini@kernel.org
Cc: hpa@zytor.com
Cc: bp@alien8.de
Cc: torvalds@linux-foundation.org
Link: https://lkml.kernel.org/r/20190210074056.11842-1-jgross@suse.com
2 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sat, 9 Feb 2019 21:43:12 +0000 (13:43 -0800)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "One PM related driver bugfix and a MAINTAINERS update"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  MAINTAINERS: Update the ocores i2c bus driver maintainer, etc
  i2c: omap: Use noirq system sleep pm ops to idle device for suspend

2 years agoMerge tag 'mips_fixes_5.0_3' of git://git.kernel.org/pub/scm/linux/kernel/git/mips...
Linus Torvalds [Sat, 9 Feb 2019 20:41:14 +0000 (12:41 -0800)]
Merge tag 'mips_fixes_5.0_3' of git://git./linux/kernel/git/mips/linux

Pull MIPS fixes from Paul Burton:
 "A batch of MIPS fixes for 5.0, nothing too scary.

   - A workaround for a Loongson 3 CPU bug is the biggest change, but
     still fairly straightforward. It adds extra memory barriers (sync
     instructions) around atomics to avoid a CPU bug that can break
     atomicity.

   - Loongson64 also sees a fix for powering off some systems which
     would incorrectly reboot rather than waiting for the power down
     sequence to complete.

   - We have DT fixes for the Ingenic JZ4740 SoC & the JZ4780-based Ci20
     board, and a DT warning fix for the Nexsys4/MIPSfpga board.

   - The Cavium Octeon platform sees a further fix to the behaviour of
     the pcie_disable command line argument that was introduced in v3.3.

   - The VDSO, introduced in v4.4, sees build fixes for configurations
     of GCC that were built using the --with-fp-32= flag to specify a
     default 32-bit floating point ABI.

   - get_frame_info() sees a fix for configurations with
     CONFIG_KALLSYMS=n, for which it previously always returned an
     error.

   - If the MIPS Coherence Manager (CM) reports an error then we'll now
     clear that error correctly so that the GCR_ERROR_CAUSE register
     will be updated with information about any future errors"

* tag 'mips_fixes_5.0_3' of git://git.kernel.org/pub/scm/linux/kernel/git/mips/linux:
  mips: cm: reprime error cause
  mips: loongson64: remove unreachable(), fix loongson_poweroff().
  MIPS: Remove function size check in get_frame_info()
  MIPS: Use lower case for addresses in nexys4ddr.dts
  MIPS: Loongson: Introduce and use loongson_llsc_mb()
  MIPS: VDSO: Include $(ccflags-vdso) in o32,n32 .lds builds
  MIPS: VDSO: Use same -m%-float cflag as the kernel proper
  MIPS: OCTEON: don't set octeon_dma_bar_type if PCI is disabled
  DTS: CI20: Fix bugs in ci20's device tree.
  MIPS: DTS: jz4740: Correct interrupt number of DMA core

2 years agoMerge tag 'for-linus-20190209' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 9 Feb 2019 18:26:09 +0000 (10:26 -0800)]
Merge tag 'for-linus-20190209' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:

 - NVMe pull request from Christoph, fixing namespace locking when
   dealing with the effects log, and a rapid add/remove issue (Keith)

 - blktrace tweak, ensuring requests with -1 sectors are shown (Jan)

 - link power management quirk for a Smasung SSD (Hans)

 - m68k nfblock dynamic major number fix (Chengguang)

 - series fixing blk-iolatency inflight counter issue (Liu)

 - ensure that we clear ->private when setting up the aio kiocb (Mike)

 - __find_get_block_slow() rate limit print (Tetsuo)

* tag 'for-linus-20190209' of git://git.kernel.dk/linux-block:
  blk-mq: remove duplicated definition of blk_mq_freeze_queue
  Blk-iolatency: warn on negative inflight IO counter
  blk-iolatency: fix IO hang due to negative inflight counter
  blktrace: Show requests without sector
  fs: ratelimit __find_get_block_slow() failure message.
  m68k: set proper major_num when specifying module param major_num
  libata: Add NOLPM quirk for SAMSUNG MZ7TE512HMHP-000L1 SSD
  nvme-pci: fix rapid add remove sequence
  nvme: lock NS list changes while handling command effects
  aio: initialize kiocb private in case any filesystems expect it.

2 years agoMerge tag 'mtd/fixes-for-5.0-rc6' of git://git.infradead.org/linux-mtd
Linus Torvalds [Sat, 9 Feb 2019 18:17:01 +0000 (10:17 -0800)]
Merge tag 'mtd/fixes-for-5.0-rc6' of git://git.infradead.org/linux-mtd

Pull mtd fixes from Boris Brezillon:

 - Fix a problem with the imx28 ECC engine

 - Remove a debug trace introduced in 2b6f0090a333 ("mtd: Check
   add_mtd_device() ret code")

 - Make sure partitions of size 0 can be registered

 - Fix kernel-doc warning in the rawnand core

 - Fix the error path of spinand_init() (missing manufacturer cleanup in
   a few places)

 - Address a problem with the SPI NAND PROGRAM LOAD operation which does
   not work as expected on some parts.

* tag 'mtd/fixes-for-5.0-rc6' of git://git.infradead.org/linux-mtd:
  mtd: rawnand: gpmi: fix MX28 bus master lockup problem
  mtd: Make sure mtd->erasesize is valid even if the partition is of size 0
  mtd: Remove a debug trace in mtdpart.c
  mtd: rawnand: fix kernel-doc warnings
  mtd: spinand: Fix the error/cleanup path in spinand_init()
  mtd: spinand: Handle the case where PROGRAM LOAD does not reset the cache

2 years agoMerge tag 'for-linus-5.0-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 9 Feb 2019 17:44:08 +0000 (09:44 -0800)]
Merge tag 'for-linus-5.0-rc6-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:
 "Two very minor fixes: one remove of a #include for an unused header
  and a fix of the xen ML address in MAINTAINERS"

* tag 'for-linus-5.0-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  MAINTAINERS: unify reference to xen-devel list
  arch/arm/xen: Remove duplicate header

2 years agoMerge tag 'perf-urgent-for-mingo-5.0-20190205' of git://git.kernel.org/pub/scm/linux...
Ingo Molnar [Sat, 9 Feb 2019 12:13:45 +0000 (13:13 +0100)]
Merge tag 'perf-urgent-for-mingo-5.0-20190205' of git://git./linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

perf trace:

  Arnaldo Carvalho de Melo:

    Fix handling of probe:vfs_getname when the probed routine is
    inlined in multiple places, fixing the collection of the 'filename'
    parameter in open syscalls.

perf test:

  Gustavo A. R. Silva:

    Fix bitwise operator usage in evsel-tp-sched test, which made tat
    test always detect fields as signed.

  Jiri Olsa:

    Filter out hidden symbols from labels, added in systems where the
    annobin plugin is used, such as RHEL8, which, if left in place make
    the DWARF unwind 'perf test' to fail on PPC.

  Tony Jones:

    Fix 'perf_event_attr' tests when building with python3.

perf mem/c2c:

  Ravi Bangoria:

    Fix perf_mem_events on PowerPC.

tools headers UAPI:

  Arnaldo Carvalho de Melo:

    Sync linux/in.h copy from the kernel sources, silencing a perf build warning.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
2 years agoMerge tag 'armsoc-fixes-5.0' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Sat, 9 Feb 2019 00:23:41 +0000 (16:23 -0800)]
Merge tag 'armsoc-fixes-5.0' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Arnd Bergmann:
 "This is a bit larger than normal, as we had not managed to send out a
  pull request before traveling for a week without my signing key.

  There are multiple code fixes for older bugs, all of which should get
  backported into stable kernels:

   - tango: one fix for multiplatform configurations broken on other
     platforms when tango is enabled

   - arm_scmi: device unregistration fix

   - iop32x: fix kernel oops from extraneous __init annotation

   - pxa: remove a double kfree

   - fsl qbman: close an interrupt clearing race

  The rest is the usual collection of smaller fixes for device tree
  files, on the renesas, allwinner, meson, omap, davinci, qualcomm and
  imx platforms.

  Some of these are for compile-time warnings, most are for board
  specific functionality that fails to work because of incorrect
  settings"

* tag 'armsoc-fixes-5.0' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc: (30 commits)
  ARM: tango: Improve ARCH_MULTIPLATFORM compatibility
  firmware: arm_scmi: provide the mandatory device release callback
  ARM: iop32x/n2100: fix PCI IRQ mapping
  arm64: dts: add msm8996 compatible to gicv3
  ARM: dts: am335x-shc.dts: fix wrong cd pin level
  ARM: dts: n900: fix mmc1 card detect gpio polarity
  ARM: dts: omap3-gta04: Fix graph_port warning
  ARM: pxa: ssp: unneeded to free devm_ allocated data
  ARM: dts: r8a7743: Convert to new LVDS DT bindings
  soc: fsl: qbman: avoid race in clearing QMan interrupt
  arm64: dts: renesas: r8a77965: Enable DMA for SCIF2
  arm64: dts: renesas: r8a7796: Enable DMA for SCIF2
  arm64: dts: renesas: r8a774a1: Enable DMA for SCIF2
  ARM: dts: da850: fix interrupt numbers for clocksource
  dt-bindings: imx8mq: Number clocks consecutively
  arm64: dts: meson: Fix mmc cd-gpios polarity
  ARM: dts: imx6sx: correct backward compatible of gpt
  ARM: dts: imx: replace gpio-key,wakeup with wakeup-source property
  ARM: dts: vf610-bk4: fix incorrect #address-cells for dspi3
  ARM: dts: meson8m2: mxiii-plus: mark the SD card detection GPIO active-low
  ...

2 years agoMerge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
Linus Torvalds [Sat, 9 Feb 2019 00:21:33 +0000 (16:21 -0800)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Will Deacon:
 "Two arm64 fixes for -rc6. They resolve a kernel NULL dereference in
  kexec and bogus kernel page table dumping when userspace is configured
  for 52-bit virtual addressing.

  Summary:

   - Fix kernel oops when attemping kexec_file() with a NULL cmdline

   - Fix page table output in debugfs when ARM64_USER_VA_BITS_52=y"

* tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux:
  arm64: kexec_file: handle empty command-line
  arm64: ptdump: Don't iterate kernel page tables using PTRS_PER_PXX

2 years agoMerge tag 'powerpc-5.0-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sat, 9 Feb 2019 00:04:12 +0000 (16:04 -0800)]
Merge tag 'powerpc-5.0-4' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "Just two fixes, both going to stable.

   - Our support for split pmd page table lock had a bug which could
     lead to a crash on mremap() when using the Radix MMU (Power9 only).

   - A fix for the PAPR SCM driver (nvdimm) we added last release, which
     had a bug where we might mis-handle a hypervisor response leading
     to us failing to attach the memory region.

  Thanks to: Aneesh Kumar K.V, Oliver O'Halloran"

* tag 'powerpc-5.0-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/papr_scm: Use the correct bind address
  powerpc/radix: Fix kernel crash with mremap()

2 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Fri, 8 Feb 2019 23:39:28 +0000 (15:39 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull signal fixes from Eric Biederman:
 "This contains four small fixes for signal handling. A missing range
  check, a regression fix, prioritizing signals we have already started
  a signal group exit for, and better detection of synchronous signals.

  The confused decision of which signals to handle failed spectacularly
  when a timer was pointed at SIGBUS and the stack overflowed. Resulting
  in an unkillable process in an infinite loop instead of a SIGSEGV and
  core dump"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  signal: Better detection of synchronous signals
  signal: Always notice exiting tasks
  signal: Always attempt to allocate siginfo for SIGSTOP
  signal: Make siginmask safe when passed a signal of 0

2 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Fri, 8 Feb 2019 23:37:17 +0000 (15:37 -0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "This is a set of five minor fixes (although, tecnhincally, the aicxxx
  fix is for a major problem in that the driver won't load without it,
  but I think the fact it's taken us since 4.10 to discover this
  indicates that the user base for these things has declined)"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: cxlflash: Prevent deadlock when adapter probe fails
  Revert "scsi: libfc: Add WARN_ON() when deleting rports"
  scsi: sd_zbc: Fix zone information messages
  scsi: target: make the pi_prot_format ConfigFS path readable
  scsi: aic94xx: fix module loading

2 years agoMerge tag 'iommu-fixes-v5.0-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 8 Feb 2019 23:34:10 +0000 (15:34 -0800)]
Merge tag 'iommu-fixes-v5.0-rc5' of git://git./linux/kernel/git/joro/iommu

Pull IOMMU fix from Joerg Roedel:
 "Intel decided to leave the newly added Scalable Mode Feature
  default-disabled for now. The patch here accomplishes that"

* tag 'iommu-fixes-v5.0-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
  iommu/vt-d: Leave scalable mode default off

2 years agoMerge tag 'pci-v5.0-fixes-4' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
Linus Torvalds [Fri, 8 Feb 2019 23:32:10 +0000 (15:32 -0800)]
Merge tag 'pci-v5.0-fixes-4' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fix from Bjorn Helgaas:
 "Work around Synopsys duplicate Device ID (HAPS USB3, NXP i.MX) that
  breaks PCIe on I.MX SoCs (Thinh Nguyen)"

* tag 'pci-v5.0-fixes-4' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: Work around Synopsys duplicate Device ID (HAPS USB3, NXP i.MX)

2 years agoMerge tag 'acpi-5.0-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Fri, 8 Feb 2019 23:30:02 +0000 (15:30 -0800)]
Merge tag 'acpi-5.0-rc6' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "This prevents excessive ACPI debug messages from being printed to the
  kernel log, which has started to happen after one of the recent ACPICA
  commits (Erik Schmauss)"

* tag 'acpi-5.0-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: Set debug output flags independent of ACPICA

2 years agoMAINTAINERS: Update the ocores i2c bus driver maintainer, etc
Andrew Lunn [Fri, 8 Feb 2019 16:56:31 +0000 (17:56 +0100)]
MAINTAINERS: Update the ocores i2c bus driver maintainer, etc

The listed maintainer has not been responding to emails for a while.
Add myself as a second maintainer.

Add the platform data include file, which was not listed.

Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Wolfram Sang <wsa@the-dreams.de>
2 years agoblk-mq: remove duplicated definition of blk_mq_freeze_queue
Liu Bo [Fri, 25 Jan 2019 00:12:49 +0000 (08:12 +0800)]
blk-mq: remove duplicated definition of blk_mq_freeze_queue

As the prototype has been defined in "include/linux/blk-mq.h", the one
in "block/blk-mq.h" can be removed then.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoBlk-iolatency: warn on negative inflight IO counter
Liu Bo [Fri, 25 Jan 2019 00:12:48 +0000 (08:12 +0800)]
Blk-iolatency: warn on negative inflight IO counter

This is to catch any unexpected negative value of inflight IO counter.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoblk-iolatency: fix IO hang due to negative inflight counter
Liu Bo [Fri, 25 Jan 2019 00:12:47 +0000 (08:12 +0800)]
blk-iolatency: fix IO hang due to negative inflight counter

Our test reported the following stack, and vmcore showed that
->inflight counter is -1.

[ffffc9003fcc38d0] __schedule at ffffffff8173d95d
[ffffc9003fcc3958] schedule at ffffffff8173de26
[ffffc9003fcc3970] io_schedule at ffffffff810bb6b6
[ffffc9003fcc3988] blkcg_iolatency_throttle at ffffffff813911cb
[ffffc9003fcc3a20] rq_qos_throttle at ffffffff813847f3
[ffffc9003fcc3a48] blk_mq_make_request at ffffffff8137468a
[ffffc9003fcc3b08] generic_make_request at ffffffff81368b49
[ffffc9003fcc3b68] submit_bio at ffffffff81368d7d
[ffffc9003fcc3bb8] ext4_io_submit at ffffffffa031be00 [ext4]
[ffffc9003fcc3c00] ext4_writepages at ffffffffa03163de [ext4]
[ffffc9003fcc3d68] do_writepages at ffffffff811c49ae
[ffffc9003fcc3d78] __filemap_fdatawrite_range at ffffffff811b6188
[ffffc9003fcc3e30] filemap_write_and_wait_range at ffffffff811b6301
[ffffc9003fcc3e60] ext4_sync_file at ffffffffa030cee8 [ext4]
[ffffc9003fcc3ea8] vfs_fsync_range at ffffffff8128594b
[ffffc9003fcc3ee8] do_fsync at ffffffff81285abd
[ffffc9003fcc3f18] sys_fsync at ffffffff81285d50
[ffffc9003fcc3f28] do_syscall_64 at ffffffff81003c04
[ffffc9003fcc3f50] entry_SYSCALL_64_after_swapgs at ffffffff81742b8e

The ->inflight counter may be negative (-1) if

1) blk-iolatency was disabled when the IO was issued,

2) blk-iolatency was enabled before this IO reached its endio,

3) the ->inflight counter is decreased from 0 to -1 in endio()

In fact the hang can be easily reproduced by the below script,

H=/sys/fs/cgroup/unified/
P=/sys/fs/cgroup/unified/test

echo "+io" > $H/cgroup.subtree_control
mkdir -p $P

echo $$ > $P/cgroup.procs

xfs_io -f -d -c "pwrite 0 4k" /dev/sdg

echo "`cat /sys/block/sdg/dev` target=1000000" > $P/io.latency

xfs_io -f -d -c "pwrite 0 4k" /dev/sdg

This fixes the problem by freezing the queue so that while
enabling/disabling iolatency, there is no inflight rq running.

Note that quiesce_queue is not needed as this only updating iolatency
configuration about which dispatching request_queue doesn't care.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Fri, 8 Feb 2019 19:21:54 +0000 (11:21 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:
 "This pull request is dedicated to the upcoming snowpocalypse parts 2
  and 3 in the Pacific Northwest:

   1) Drop profiles are broken because some drivers use dev_kfree_skb*
      instead of dev_consume_skb*, from Yang Wei.

   2) Fix IWLWIFI kconfig deps, from Luca Coelho.

   3) Fix percpu maps updating in bpftool, from Paolo Abeni.

   4) Missing station release in batman-adv, from Felix Fietkau.

   5) Fix some networking compat ioctl bugs, from Johannes Berg.

   6) ucc_geth must reset the BQL queue state when stopping the device,
      from Mathias Thore.

   7) Several XDP bug fixes in virtio_net from Toshiaki Makita.

   8) TSO packets must be sent always on queue 0 in stmmac, from Jose
      Abreu.

   9) Fix socket refcounting bug in RDS, from Eric Dumazet.

  10) Handle sparse cpu allocations in bpf selftests, from Martynas
      Pumputis.

  11) Make sure mgmt frames have enough tailroom in mac80211, from Felix
      Feitkau.

  12) Use safe list walking in sctp_sendmsg() asoc list traversal, from
      Greg Kroah-Hartman.

  13) Make DCCP's ccid_hc_[rt]x_parse_options always check for NULL
      ccid, from Eric Dumazet.

  14) Need to reload WoL password into bcmsysport device after deep
      sleeps, from Florian Fainelli.

  15) Remove filter from mask before freeing in cls_flower, from Petr
      Machata.

  16) Missing release and use after free in error paths of s390 qeth
      code, from Julian Wiedmann.

  17) Fix lockdep false positive in dsa code, from Marc Zyngier.

  18) Fix counting of ATU violations in mv88e6xxx, from Andrew Lunn.

  19) Fix EQ firmware assert in qed driver, from Manish Chopra.

  20) Don't default Caivum PTP to Y in kconfig, from Bjorn Helgaas"

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (116 commits)
  net: dsa: b53: Fix for failure when irq is not defined in dt
  sit: check if IPv6 enabled before calling ip6_err_gen_icmpv6_unreach()
  geneve: should not call rt6_lookup() when ipv6 was disabled
  net: Don't default Cavium PTP driver to 'y'
  net: broadcom: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: via-velocity: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: tehuti: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: sun: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: fsl_ucc_hdlc: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: fec_mpc52xx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: smsc: epic100: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: dscc4: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: tulip: de2104x: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net: defxx: replace dev_kfree_skb_irq by dev_consume_skb_irq for drop profiles
  net/mlx5e: Don't overwrite pedit action when multiple pedit used
  net/mlx5e: Update hw flows when encap source mac changed
  qed*: Advance drivers version to 8.37.0.20
  qed: Change verbosity for coalescing message.
  qede: Fix system crash on configuring channels.
  qed: Consider TX tcs while deriving the max num_queues for PF.
  ...