platform/kernel/linux-rpi.git
3 years agobtrfs: use cached value of fs_info::csum_size everywhere
David Sterba [Tue, 30 Jun 2020 00:01:31 +0000 (02:01 +0200)]
btrfs: use cached value of fs_info::csum_size everywhere

btrfs_get_16 shows up in the system performance profiles (helper to read
16bit values from on-disk structures). This is partially because of the
checksum size that's frequently read along with data reads/writes, other
u16 uses are from item size or directory entries.

Replace all calls to btrfs_super_csum_size by the cached value from
fs_info.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: precalculate checksums per leaf once
David Sterba [Thu, 2 Jul 2020 08:54:11 +0000 (10:54 +0200)]
btrfs: precalculate checksums per leaf once

btrfs_csum_bytes_to_leaves shows up in system profiles, which makes it a
candidate for optimizations. After the 64bit division has been replaced
by shift, there's still a calculation done each time the function is
called: checksums per leaf.

As this is a constant value for the entire filesystem lifetime, we
can calculate it once at mount time and reuse. This also allows to
reduce the division to 64bit/32bit as we know the constant will always
fit the 32bit type.

Replace the open-coded rounding up with a macro that internally handles
the 64bit division and as it's now a short function, make it static
inline (slight code increase, slight stack usage reduction).

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: store precalculated csum_size in fs_info
David Sterba [Thu, 2 Jul 2020 09:10:18 +0000 (11:10 +0200)]
btrfs: store precalculated csum_size in fs_info

In many places we need the checksum size and it is inefficient to read
it from the raw superblock. Store the value into fs_info, actual use
will be in followup patches.  The size is u32 as it allows to generate
better assembly than with u16.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: replace s_blocksize_bits with fs_info::sectorsize_bits
David Sterba [Wed, 1 Jul 2020 19:19:09 +0000 (21:19 +0200)]
btrfs: replace s_blocksize_bits with fs_info::sectorsize_bits

The value of super_block::s_blocksize_bits is the same as
fs_info::sectorsize_bits, but we don't need to do the extra dereferences
in many functions and storing the bits as u32 (in fs_info) generates
shorter assembly.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: replace div_u64 by shift in free_space_bitmap_size
David Sterba [Wed, 1 Jul 2020 19:07:40 +0000 (21:07 +0200)]
btrfs: replace div_u64 by shift in free_space_bitmap_size

Change free_space_bitmap_size to take btrfs_fs_info so we can get the
sectorsize_bits to do calculations.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use precalculated sectorsize_bits from fs_info
David Sterba [Wed, 1 Jul 2020 18:45:04 +0000 (20:45 +0200)]
btrfs: use precalculated sectorsize_bits from fs_info

We do a lot of calculations where we divide or multiply by sectorsize.
We also know and make sure that sectorsize is a power of two, so this
means all divisions can be turned to shifts and avoid eg. expensive
u64/u32 divisions.

The type is u32 as it's more register friendly on x86_64 compared to u8
and the resulting assembly is smaller (movzbl vs movl).

There's also superblock s_blocksize_bits but it's usually one more
pointer dereference farther than fs_info.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: rename page_size to io_size in submit_extent_page
Qu Wenruo [Wed, 21 Oct 2020 06:25:01 +0000 (14:25 +0800)]
btrfs: rename page_size to io_size in submit_extent_page

The variable @page_size in submit_extent_page() is not related to page
size.

It can already be smaller than PAGE_SIZE, so rename it to io_size to
reduce confusion, this is especially important for later subpage
support.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: only require sector size alignment for page read
Qu Wenruo [Wed, 21 Oct 2020 06:24:58 +0000 (14:24 +0800)]
btrfs: only require sector size alignment for page read

If we're reading partial page, btrfs will warn about this as read/write
is always done in sector size, which now equals page size.

But for the upcoming subpage read-only support, our data read is only
aligned to sectorsize, which can be smaller than page size.

Thus here we change the warning condition to check it against
sectorsize, the behavior is not changed for regular sectorsize ==
PAGE_SIZE case, and won't report error for subpage read.

Also, pass the proper start/end with bv_offset for check_data_csum() to
handle.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: rename pages_locked in process_pages_contig()
Qu Wenruo [Wed, 21 Oct 2020 06:24:57 +0000 (14:24 +0800)]
btrfs: rename pages_locked in process_pages_contig()

Function process_pages_contig() does not only handle page locking but
also other operations.  Rename the local variable pages_locked to
pages_processed to reduce confusion.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: sink parameter start and len to check_data_csum
Qu Wenruo [Wed, 21 Oct 2020 06:24:54 +0000 (14:24 +0800)]
btrfs: sink parameter start and len to check_data_csum

For check_data_csum(), the page we're using is directly from the inode
mapping, thus it has valid page_offset().

We can use (page_offset() + pg_off) to replace @start parameter
completely, while the @len should always be sectorsize.

Since we're here, also add some comment, as there are quite some
confusion in words like start/offset, without explaining whether it's
file_offset or logical bytenr.

This should not affect the existing behavior, as for current sectorsize
== PAGE_SIZE case, @pgoff should always be 0, and len is always
PAGE_SIZE (or sectorsize from the dio read path).

Reviewed-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: replace fs_info and private_data with inode in btrfs_wq_submit_bio
Qu Wenruo [Wed, 21 Oct 2020 06:24:53 +0000 (14:24 +0800)]
btrfs: replace fs_info and private_data with inode in btrfs_wq_submit_bio

All callers of btrfs_wq_submit_bio() pass struct inode as @private_data,
so there is no need for it to be (void *), replace it with "struct inode
*inode".

While we can extract fs_info from struct inode, also remove the @fs_info
parameter.

Since we're here, also replace all the (void *private_data) into (struct
inode *inode).

Reviewed-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: sink the failed_start parameter to set_extent_bit
Qu Wenruo [Wed, 21 Oct 2020 06:24:51 +0000 (14:24 +0800)]
btrfs: sink the failed_start parameter to set_extent_bit

The @failed_start parameter is only paired with @exclusive_bits, and
those parameters are only used for EXTENT_LOCKED bit, which have their
own wrappers lock_extent_bits().

Thus for regular set_extent_bit() calls, the failed_start makes no
sense, just sink the parameter.

Also, since @failed_start and @exclusive_bits are used in pairs, add
an assert to make it obvious.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: update the comment for find_first_extent_bit
Qu Wenruo [Wed, 21 Oct 2020 06:24:50 +0000 (14:24 +0800)]
btrfs: update the comment for find_first_extent_bit

The pitfall here is, if the parameter @bits has multiple bits set, we
will return the first range which just has one of the specified bits
set.

This is a little tricky if we want an exact match.  Anyway, update the
comment to make that clear.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: fix the comment on lock_extent_buffer_for_io
Qu Wenruo [Wed, 21 Oct 2020 06:24:49 +0000 (14:24 +0800)]
btrfs: fix the comment on lock_extent_buffer_for_io

The return value of that function is completely wrong.

That function only returns 0 if the extent buffer doesn't need to be
submitted.  The "ret = 1" and "ret = 0" are determined by the return
value of "test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)".

And if we get ret == 1, it's because the extent buffer is dirty, and we
set its status to EXTENT_BUFFER_WRITE_BACK, and continue to page
locking.

While if we get ret == 0, it means the extent is not dirty from the
beginning, so we don't need to write it back.

The caller also follows this, in btree_write_cache_pages(), if
lock_extent_buffer_for_io() returns 0, we just skip the extent buffer
completely.

So the comment is completely wrong.

Since we're here, also change the description a little.  The write bio
flushing won't be visible to the caller, thus it's not an major feature.
In the main description, only describe the locking part to make the
point more clear.

For reference, added in commit 2e3c25136adf ("btrfs: extent_io: add
proper error handling to lock_extent_buffer_for_io()")

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: remove unnecessary casts in printk
David Sterba [Tue, 15 Sep 2020 12:18:23 +0000 (14:18 +0200)]
btrfs: remove unnecessary casts in printk

Long time ago the explicit casts were necessary for u64 but we don't
need it.  Remove casts where the type matches, leaving only cases that
cast sector_t or loff_t.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: add set/get accessors for root_item::drop_level
David Sterba [Tue, 15 Sep 2020 19:44:52 +0000 (21:44 +0200)]
btrfs: add set/get accessors for root_item::drop_level

The drop_level member is used directly unlike all the other int types in
root_item. Add the definition and use it everywhere. The type is u8 so
there's no conversion necessary and the helpers are properly inlined,
this is for consistency.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use root_item helpers for limit and flags in btrfs_create_tree
David Sterba [Tue, 15 Sep 2020 19:00:04 +0000 (21:00 +0200)]
btrfs: use root_item helpers for limit and flags in btrfs_create_tree

For consistency use the available helpers to set flags and limit.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: check-integrity: use proper helper to access btrfs_header
David Sterba [Tue, 15 Sep 2020 12:26:27 +0000 (14:26 +0200)]
btrfs: check-integrity: use proper helper to access btrfs_header

There's one raw use of le->cpu conversion but we have a helper to do
that for us, so use it.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: send: use helpers to access root_item::ctransid
David Sterba [Tue, 15 Sep 2020 12:30:15 +0000 (14:30 +0200)]
btrfs: send: use helpers to access root_item::ctransid

We have helpers to access the on-disk item members, use that for
root_item::ctransid instead of raw le64_to_cpu.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: generate lockdep keyset names at compile time
David Sterba [Tue, 29 Sep 2020 12:56:39 +0000 (14:56 +0200)]
btrfs: generate lockdep keyset names at compile time

The names in btrfs_lockdep_keysets are generated from a simple pattern
using snprintf but we can generate them directly with some macro magic
and remove the helpers.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use the right number of levels for lockdep keysets
David Sterba [Tue, 27 Oct 2020 14:54:08 +0000 (15:54 +0100)]
btrfs: use the right number of levels for lockdep keysets

BTRFS_MAX_LEVEL is 8 and the keyset table is supposed to have a key for
each level, but we'll never have more than 8 levels.  The values passed
to btrfs_set_buffer_lockdep_class are always derived from a valid extent
buffer.  Set the array sizes to the right value.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: remove dio iomap DSYNC workaround
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:21 +0000 (11:39 -0500)]
btrfs: remove dio iomap DSYNC workaround

This effectively reverts 09745ff88d93 ("btrfs: dio iomap DSYNC
workaround") now that the iomap API has been updated to allow
iomap_dio_complete() not to be called under i_rwsem anymore.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: call iomap_dio_complete() without inode_lock
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:20 +0000 (11:39 -0500)]
btrfs: call iomap_dio_complete() without inode_lock

If direct writes are called with O_DIRECT | O_DSYNC, it will result in a
deadlock because iomap_dio_rw() is called under i_rwsem which calls:

  iomap_dio_complete()
    generic_write_sync()
      btrfs_sync_file()

btrfs_sync_file() requires i_rwsem, so call __iomap_dio_rw() with the
i_rwsem locked, and call iomap_dio_complete() after unlocking i_rwsem.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: remove btrfs_inode::dio_sem
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:19 +0000 (11:39 -0500)]
btrfs: remove btrfs_inode::dio_sem

The inode dio_sem can be eliminated because all DIO synchronization is
now performed through inode->i_rwsem that provides the same guarantees.

This reduces btrfs_inode size by 40 bytes.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use shared lock for direct writes within EOF
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:18 +0000 (11:39 -0500)]
btrfs: use shared lock for direct writes within EOF

Direct writes within EOF are safe to be performed with inode shared lock
to improve parallelization with other direct writes or reads because EOF
is not changed and there is no race with truncate().

Direct reads are already performed under shared inode lock.

This patch is precursor to removing btrfs_inode->dio_sem.

Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: push inode locking and unlocking into buffered/direct write
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:17 +0000 (11:39 -0500)]
btrfs: push inode locking and unlocking into buffered/direct write

Push inode locking and unlocking closer to where we perform the I/O. For
this we need to move the write checks inside the respective functions as
well.

pos is evaluated after generic_write_checks because O_APPEND can change
iocb->ki_pos.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: introduce btrfs_inode_lock()/unlock()
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:16 +0000 (11:39 -0500)]
btrfs: introduce btrfs_inode_lock()/unlock()

btrfs_inode_lock/unlock() are wrappers around inode locks, separating
the type of lock and actual locking.

- 0 - default, exclusive lock
- BTRFS_ILOCK_SHARED - for shared locks, for possible parallel DIO
- BTRFS_ILOCK_TRY - for the RWF_NOWAIT sequence

The bits SHARED and TRY can be combined together.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: introduce btrfs_write_check()
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:15 +0000 (11:39 -0500)]
btrfs: introduce btrfs_write_check()

btrfs_write_check() checks write parameters in one place before
beginning a write. This does away with inode_unlock() after every check.
In the later patches, it will help push inode_lock/unlock() in buffered
and direct write functions respectively.

generic_write_checks needs to be called before as it could truncate
iov_iter and its return used as count.

Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: check FS error state bit early during write
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:14 +0000 (11:39 -0500)]
btrfs: check FS error state bit early during write

fs_info::fs_state is a filesystem bit check as opposed to inode and can
be performed before we begin with write checks. This eliminates inode
lock/unlock in case the error bit is set.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: move pos increment and pagecache extension to btrfs_buffered_write
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:13 +0000 (11:39 -0500)]
btrfs: move pos increment and pagecache extension to btrfs_buffered_write

While we do this, correct the call to pagecache_isize_extended:

 - pagecache_isize_extended needs to be called to the start of the write
   as opposed to i_size

 - we don't need to check range before the call, this is done in the
   function

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: split btrfs_direct_IO to read and write
Goldwyn Rodrigues [Thu, 24 Sep 2020 16:39:12 +0000 (11:39 -0500)]
btrfs: split btrfs_direct_IO to read and write

The read and write DIO don't have anything in common except for the
call to iomap_dio_rw. Extract the write call into a new function to get
rid of conditional statements for direct write.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: sysfs: add per-fs attribute for read policy
Anand Jain [Wed, 28 Oct 2020 13:14:47 +0000 (21:14 +0800)]
btrfs: sysfs: add per-fs attribute for read policy

Add

 /sys/fs/btrfs/UUID/read_policy

attribute so that the read policy for the raid1, raid1c34 and raid10 can
be tuned.

When this attribute is read, it will show all available policies, with
active policy in [ ]. The read_policy attribute can be written using one
of the items listed in there.

For example:
  $ cat /sys/fs/btrfs/UUID/read_policy
  [pid]
  $ echo pid > /sys/fs/btrfs/UUID/read_policy

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: create read policy framework
Anand Jain [Wed, 28 Oct 2020 13:14:46 +0000 (21:14 +0800)]
btrfs: create read policy framework

As of now, we use the pid method to read striped mirrored data, which
means process id determines the stripe id to read. This type of routing
typically helps in a system with many small independent processes tying
to read random data. On the other hand, the pid based read IO policy is
inefficient because if there is a single process trying to read a large
file, the overall disk bandwidth remains underutilized.

So this patch introduces a read policy framework so that we could add
more read policies, such as IO routing based on the device's wait-queue
or manual when we have a read-preferred device or a policy based on the
target storage caching.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: add helper for string match ignoring leading/trailing whitespace
Anand Jain [Wed, 28 Oct 2020 13:14:45 +0000 (21:14 +0800)]
btrfs: add helper for string match ignoring leading/trailing whitespace

Add a generic helper to match the string in a given buffer, and ignore
the leading and trailing whitespace.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ rename variables, add comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: do not start and wait for delalloc on snapshot roots on transaction commit
Filipe Manana [Tue, 27 Oct 2020 12:40:06 +0000 (12:40 +0000)]
btrfs: do not start and wait for delalloc on snapshot roots on transaction commit

We do not need anymore to start writeback for delalloc of roots that are
being snapshotted and wait for it to complete. This was done in commit
609e804d771f59 ("Btrfs: fix file corruption after snapshotting due to mix
of buffered/DIO writes") to fix a type of file corruption where files in a
snapshot end up having their i_size updated in a non-ordered way, leaving
implicit file holes, when buffered IO writes that increase a file's size
are followed by direct IO writes that also increase the file's size.

This is not needed anymore because we now have a more generic mechanism
to prevent a non-ordered i_size update since commit 9ddc959e802bf7
("btrfs: use the file extent tree infrastructure"), which addresses this
scenario involving snapshots as well.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: switch extent buffer tree lock to rw_semaphore
Josef Bacik [Thu, 20 Aug 2020 15:46:09 +0000 (11:46 -0400)]
btrfs: switch extent buffer tree lock to rw_semaphore

Historically we've implemented our own locking because we wanted to be
able to selectively spin or sleep based on what we were doing in the
tree.  For instance, if all of our nodes were in cache then there's
rarely a reason to need to sleep waiting for node locks, as they'll
likely become available soon.  At the time this code was written the
rw_semaphore didn't do adaptive spinning, and thus was orders of
magnitude slower than our home grown locking.

However now the opposite is the case.  There are a few problems with how
we implement blocking locks, namely that we use a normal waitqueue and
simply wake everybody up in reverse sleep order.  This leads to some
suboptimal performance behavior, and a lot of context switches in highly
contended cases.  The rw_semaphores actually do this properly, and also
have adaptive spinning that works relatively well.

The locking code is also a bit of a bear to understand, and we lose the
benefit of lockdep for the most part because the blocking states of the
lock are simply ad-hoc and not mapped into lockdep.

So rework the locking code to drop all of this custom locking stuff, and
simply use a rw_semaphore for everything.  This makes the locking much
simpler for everything, as we can now drop a lot of cruft and blocking
transitions.  The performance numbers vary depending on the workload,
because generally speaking there doesn't tend to be a lot of contention
on the btree.  However, on my test system which is an 80 core single
socket system with 256GiB of RAM and a 2TiB NVMe drive I get the
following results (with all debug options off):

  dbench 200 baseline
  Throughput 216.056 MB/sec  200 clients  200 procs  max_latency=1471.197 ms

  dbench 200 with patch
  Throughput 737.188 MB/sec  200 clients  200 procs  max_latency=714.346 ms

Previously we also used fs_mark to test this sort of contention, and
those results are far less impressive, mostly because there's not enough
tasks to really stress the locking

  fs_mark -d /d[0-15] -S 0 -L 20 -n 100000 -s 0 -t 16

  baseline
    Average Files/sec:     160166.7
    p50 Files/sec:         165832
    p90 Files/sec:         123886
    p99 Files/sec:         123495

    real    3m26.527s
    user    2m19.223s
    sys     48m21.856s

  patched
    Average Files/sec:     164135.7
    p50 Files/sec:         171095
    p90 Files/sec:         122889
    p99 Files/sec:         113819

    real    3m29.660s
    user    2m19.990s
    sys     44m12.259s

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: open code insert_orphan_item
Nikolay Borisov [Thu, 22 Oct 2020 15:40:46 +0000 (18:40 +0300)]
btrfs: open code insert_orphan_item

Just open code it in its sole caller and remove a level of indirection.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: introduce mount option rescue=all
Josef Bacik [Fri, 16 Oct 2020 15:29:20 +0000 (11:29 -0400)]
btrfs: introduce mount option rescue=all

Now that we have the building blocks for some better recovery options
with corrupted file systems, add a rescue=all option to enable all of
the relevant rescue options.  This will allow distros to simply default
to rescue=all for the "oh dear lord the world's on fire" recovery
without needing to know all the different options that we have and may
add in the future.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: introduce mount option rescue=ignoredatacsums
Josef Bacik [Fri, 16 Oct 2020 15:29:19 +0000 (11:29 -0400)]
btrfs: introduce mount option rescue=ignoredatacsums

There are cases where you can end up with bad data csums because of
misbehaving applications.  This happens when an application modifies a
buffer in-flight when doing an O_DIRECT write.  In order to recover the
file we need a way to turn off data checksums so you can copy the file
off, and then you can delete the file and restore it properly later.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: introduce mount option rescue=ignorebadroots
Josef Bacik [Fri, 16 Oct 2020 15:29:18 +0000 (11:29 -0400)]
btrfs: introduce mount option rescue=ignorebadroots

In the face of extent root corruption, or any other core fs wide root
corruption we will fail to mount the file system.  This makes recovery
kind of a pain, because you need to fall back to userspace tools to
scrape off data.  Instead provide a mechanism to gracefully handle bad
roots, so we can at least mount read-only and possibly recover data from
the file system.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: show rescue=usebackuproot in /proc/mounts
Josef Bacik [Fri, 16 Oct 2020 15:29:17 +0000 (11:29 -0400)]
btrfs: show rescue=usebackuproot in /proc/mounts

The standalone option usebackuproot was intended as one-time use and it
was not necessary to keep it in the option list. Now that we're going to
have more rescue options, it's desirable to keep them intact as it could
be confusing why the option disappears.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ remove the btrfs_clear_opt part from open_ctree ]
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: add a helper to print out rescue= options
Josef Bacik [Fri, 16 Oct 2020 15:29:16 +0000 (11:29 -0400)]
btrfs: add a helper to print out rescue= options

We're going to have a lot of rescue options, add a helper to collapse
the /proc/mounts output to rescue=option1:option2:option3 format.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: sysfs: export supported rescue= mount options
Josef Bacik [Fri, 16 Oct 2020 15:29:15 +0000 (11:29 -0400)]
btrfs: sysfs: export supported rescue= mount options

We're going to be adding a variety of different rescue options, we
should advertise which ones we support to make user spaces life easier
in the future.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: push the NODATASUM check into btrfs_lookup_bio_sums
Josef Bacik [Fri, 16 Oct 2020 15:29:14 +0000 (11:29 -0400)]
btrfs: push the NODATASUM check into btrfs_lookup_bio_sums

When we move to being able to handle NULL csum_roots it'll be cleaner to
just check in btrfs_lookup_bio_sums instead of at all of the caller
locations, so push the NODATASUM check into it as well so it's unified.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: unify the ro checking for mount options
Josef Bacik [Fri, 16 Oct 2020 15:29:13 +0000 (11:29 -0400)]
btrfs: unify the ro checking for mount options

We're going to be adding more options that require RDONLY, so add a
helper to do the check and error out if we don't have RDONLY set.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: do not start readahead for csum tree when scrubbing non-data block groups
Filipe Manana [Mon, 12 Oct 2020 10:55:26 +0000 (11:55 +0100)]
btrfs: do not start readahead for csum tree when scrubbing non-data block groups

When scrubbing a stripe of a block group we always start readahead for the
checksums btree and wait for it to complete, however when the blockgroup is
not a data block group (or a mixed block group) it is a waste of time to do
it, since there are no checksums for metadata extents in that btree.

So skip that when the block group does not have the data flag set, saving
some time doing memory allocations, queueing a job in the readahead work
queue, waiting for it to complete and potentially avoiding some IO as well
(when csum tree extents are not in memory already).

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: assert we are holding the reada_lock when releasing a readahead zone
Filipe Manana [Mon, 12 Oct 2020 10:55:25 +0000 (11:55 +0100)]
btrfs: assert we are holding the reada_lock when releasing a readahead zone

When we drop the last reference of a zone, we end up releasing it through
the callback reada_zone_release(), which deletes the zone from a device's
reada_zones radix tree. This tree is protected by the global readahead
lock at fs_info->reada_lock. Currently all places that are sure that they
are dropping the last reference on a zone, are calling kref_put() in a
critical section delimited by this lock, while all other places that are
sure they are not dropping the last reference, do not bother calling
kref_put() while holding that lock.

When working on the previous fix for hangs and use-after-frees in the
readahead code, my initial attempts were different and I actually ended
up having reada_zone_release() called when not holding the lock, which
resulted in weird and unexpected problems. So just add an assertion
there to detect such problem more quickly and make the dependency more
obvious.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: set EXTENT_NORESERVE bits side btrfs_dirty_pages()
Goldwyn Rodrigues [Wed, 14 Oct 2020 14:55:45 +0000 (09:55 -0500)]
btrfs: set EXTENT_NORESERVE bits side btrfs_dirty_pages()

Set the extent bits EXTENT_NORESERVE inside btrfs_dirty_pages() as
opposed to calling set_extent_bits again later.

Fold check for written length within the function.

Note: EXTENT_NORESERVE is set before unlocking extents.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use round_down while calculating start position in btrfs_dirty_pages()
Goldwyn Rodrigues [Wed, 14 Oct 2020 14:55:44 +0000 (09:55 -0500)]
btrfs: use round_down while calculating start position in btrfs_dirty_pages()

round_down looks prettier than the bit mask operations.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use iosize while reading compressed pages
Goldwyn Rodrigues [Tue, 15 Sep 2020 15:41:40 +0000 (10:41 -0500)]
btrfs: use iosize while reading compressed pages

While using compression, a submitted bio is mapped with a compressed bio
which performs the read from disk, decompresses and returns uncompressed
data to original bio. The original bio must reflect the uncompressed
size (iosize) of the I/O to be performed, or else the page just gets the
decompressed I/O length of data (disk_io_size). The compressed bio
checks the extent map and gets the correct length while performing the
I/O from disk.

This came up in subpage work when only compressed length of the original
bio was filled in the page. This worked correctly for pagesize ==
sectorsize because both compressed and uncompressed data are at pagesize
boundaries, and would end up filling the requested page.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: calculate num_pages, reserve_bytes once in btrfs_buffered_write
Goldwyn Rodrigues [Fri, 25 Sep 2020 20:36:38 +0000 (15:36 -0500)]
btrfs: calculate num_pages, reserve_bytes once in btrfs_buffered_write

write_bytes can change in btrfs_check_nocow_lock(). Calculate variables
such as num_pages and reserve_bytes once we are sure of the value of
write_bytes so there is no need to re-calculate.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: calculate more accurate remaining time to sleep in transaction_kthread
Nikolay Borisov [Tue, 20 Oct 2020 09:44:17 +0000 (12:44 +0300)]
btrfs: calculate more accurate remaining time to sleep in transaction_kthread

If transaction_kthread is woken up before btrfs_fs_info::commit_interval
seconds have elapsed it will sleep for a fixed period of 5 seconds. This
is not a problem per-se but is not accurate. Instead the code should
sleep for an interval which guarantees on next wakeup commit_interval
would have passed. Since time tracking is not precise subtract 1 second
from delta to ensure the delay we end up waiting will be longer than
than the wake up period.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: record delta directly in transaction_kthread
Nikolay Borisov [Thu, 8 Oct 2020 12:24:29 +0000 (15:24 +0300)]
btrfs: record delta directly in transaction_kthread

Rename 'now' to 'delta' and store there the delta between transaction
start time and current time. This is in preparation for optimising the
sleep logic in the next patch. No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: remove redundant time check in transaction kthread loop
Nikolay Borisov [Thu, 8 Oct 2020 12:24:28 +0000 (15:24 +0300)]
btrfs: remove redundant time check in transaction kthread loop

The value obtained from ktime_get_seconds() is guaranteed to be
monotonically increasing since it's taken from CLOCK_MONOTONIC. As
transaction_kthread obtains a reference to the currently running
transaction under holding btrfs_fs_info::trans_lock it's guaranteed to:

a) see an initialized 'cur', whose start_time is guaranteed to be smaller
   than 'now'

or

b) not obtain a 'cur' and simply go to sleep.

Given this remove the unnecessary check, if it sees
now < cur->start_time this would imply there are far greater problems on
the machine.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: use helpers to convert from seconds to jiffies in transaction_kthread
Nikolay Borisov [Thu, 8 Oct 2020 12:24:27 +0000 (15:24 +0300)]
btrfs: use helpers to convert from seconds to jiffies in transaction_kthread

The kernel provides easy to understand helpers to convert from human
understandable units to the kernel-friendly 'jiffies'. So let's use
those to make the code easier to understand. No functional changes.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agobtrfs: sysfs: export filesystem generation
Anand Jain [Wed, 7 Oct 2020 07:20:03 +0000 (15:20 +0800)]
btrfs: sysfs: export filesystem generation

Matching with the information that's available from the ioctl
FS_INFO, add generation to the per-filesystem directory
/sys/fs/btrfs/UUID/generation, which could be used by scripts.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
3 years agoLinux 5.10-rc7
Linus Torvalds [Sun, 6 Dec 2020 22:25:12 +0000 (14:25 -0800)]
Linux 5.10-rc7

3 years agoMerge tag 'char-misc-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregk...
Linus Torvalds [Sun, 6 Dec 2020 19:48:17 +0000 (11:48 -0800)]
Merge tag 'char-misc-5.10-rc7' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver fixes from Greg KH:
 "Here are some small driver fixes, and one "large" revert, for
  5.10-rc7.

  They include:

   - revert mei patch from 5.10-rc1 that was using a reserved userspace
     value. It will be resubmitted once the proper id has been assigned
     by the virtio people.

   - habanalabs fixes found by the fall-through audit from Gustavo

   - speakup driver fixes for reported issues

   - fpga config build fix for reported issue.

  All of these except the revert have been in linux-next with no
  reported issues. The revert is "clean" and just removes a
  previously-added driver, so no real issue there"

* tag 'char-misc-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  Revert "mei: virtio: virtualization frontend driver"
  fpga: Specify HAS_IOMEM dependency for FPGA_DFL
  habanalabs: put devices before driver removal
  habanalabs: free host huge va_range if not used
  speakup: Reject setting the speakup line discipline outside of speakup

3 years agoMerge tag 'tty-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Sun, 6 Dec 2020 19:43:50 +0000 (11:43 -0800)]
Merge tag 'tty-5.10-rc7' of git://git./linux/kernel/git/gregkh/tty

Pull tty fixes from Greg KH:
 "Here are two tty core fixes for 5.10-rc7.

  They resolve some reported locking issues in the tty core. While they
  have not been in a released linux-next yet, they have passed all of
  the 0-day bot testing as well as the submitter's testing"

* tag 'tty-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  tty: Fix ->session locking
  tty: Fix ->pgrp locking in tiocspgrp()

3 years agoMerge tag 'usb-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 6 Dec 2020 19:38:36 +0000 (11:38 -0800)]
Merge tag 'usb-5.10-rc7' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are some small USB fixes for 5.10-rc7 that resolve a number of
  reported issues, and add some new device ids.

  Nothing major here, but these solve some problems that people were
  having with the 5.10-rc tree:

   - reverts for USB storage dma settings that broke working devices

   - thunderbolt use-after-free fix

   - cdns3 driver fixes

   - gadget driver userspace copy fix

   - new device ids

  All of these except for the reverts have been in linux-next with no
  reported issues. The reverts are "clean" and were tested by Hans, as
  well as passing the 0-day tests"

* tag 'usb-5.10-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  usb: gadget: f_fs: Use local copy of descriptors for userspace copy
  usb: ohci-omap: Fix descriptor conversion
  Revert "usb-storage: fix sdev->host->dma_dev"
  Revert "uas: fix sdev->host->dma_dev"
  Revert "uas: bump hw_max_sectors to 2048 blocks for SS or faster drives"
  USB: serial: kl5kusb105: fix memleak on open
  USB: serial: ch341: sort device-id entries
  USB: serial: ch341: add new Product ID for CH341A
  USB: serial: option: fix Quectel BG96 matching
  usb: cdns3: core: fix goto label for error path
  usb: cdns3: gadget: clear trb->length as zero after preparing every trb
  usb: cdns3: Fix hardware based role switch
  USB: serial: option: add support for Thales Cinterion EXS82
  USB: serial: option: add Fibocom NL668 variants
  thunderbolt: Fix use-after-free in remove_unplugged_switch()

3 years agoMerge tag 'x86-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 6 Dec 2020 19:22:39 +0000 (11:22 -0800)]
Merge tag 'x86-urgent-2020-12-06' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "A set of fixes for x86:

   - Make the AMD L3 QoS code and data priorization enable/disable
     mechanism work correctly.

     The control bit was only set/cleared on one of the CPUs in a L3
     domain, but it has to be modified on all CPUs in the domain. The
     initial documentation was not clear about this, but the updated one
     from Oct 2020 spells it out.

   - Fix an off by one in the UV platform detection code which causes
     the UV hubs to be identified wrongly.

     The chip revisions start at 1 not at 0.

   - Fix a long standing bug in the evaluation of prefixes in the
     uprobes code which fails to handle repeated prefixes properly.

     The aggregate size of the prefixes can be larger than the bytes
     array but the code blindly iterated over the aggregate size beyond
     the array boundary. Add a macro to handle this case properly and
     use it at the affected places"

* tag 'x86-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/sev-es: Use new for_each_insn_prefix() macro to loop over prefixes bytes
  x86/insn-eval: Use new for_each_insn_prefix() macro to loop over prefixes bytes
  x86/uprobes: Do not use prefixes.nbytes when looping over prefixes.bytes
  x86/platform/uv: Fix UV4 hub revision adjustment
  x86/resctrl: Fix AMD L3 QOS CDP enable/disable

3 years agoMerge tag 'perf-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 6 Dec 2020 19:20:18 +0000 (11:20 -0800)]
Merge tag 'perf-urgent-2020-12-06' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Thomas Gleixner:
 "Two fixes for performance monitoring on X86:

   - Add recursion protection to another callchain invoked from
     x86_pmu_stop() which can recurse back into x86_pmu_stop(). The
     first attempt to fix this missed this extra code path.

   - Use the already filtered status variable to check for PEBS counter
     overflow bits and not the unfiltered full status read from
     IA32_PERF_GLOBAL_STATUS which can have unrelated bits check which
     would be evaluated incorrectly"

* tag 'perf-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/intel: Check PEBS status correctly
  perf/x86/intel: Fix a warning on x86_pmu_stop() with large PEBS

3 years agoMerge tag 'irq-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 6 Dec 2020 19:15:55 +0000 (11:15 -0800)]
Merge tag 'irq-urgent-2020-12-06' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Thomas Gleixner:
 "A set of updates for the interrupt subsystem:

   - Make multiqueue devices which use the managed interrupt affinity
     infrastructure work on PowerPC/Pseries. PowerPC does not use the
     generic infrastructure for setting up PCI/MSI interrupts and the
     multiqueue changes failed to update the legacy PCI/MSI
     infrastructure. Make this work by passing the affinity setup
     information down to the mapping and allocation functions.

   - Move Jason Cooper from MAINTAINERS to CREDITS as his mail is
     bouncing and he's not reachable. We hope all is well with him and
     say thanks for his work over the years"

* tag 'irq-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  powerpc/pseries: Pass MSI affinity to irq_create_mapping()
  genirq/irqdomain: Add an irq_create_mapping_affinity() function
  MAINTAINERS: Move Jason Cooper to CREDITS

3 years agoMerge tag 'locking-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 6 Dec 2020 19:11:32 +0000 (11:11 -0800)]
Merge tag 'locking-urgent-2020-12-06' of git://git./linux/kernel/git/tip/tip

Pull intel_idle build fix from Thomas Gleixner:
 "A tiny build fix for a recent change in the intel_idle driver which
  missed a CONFIG dependency and broke the build for certain
  configurations"

* tag 'locking-urgent-2020-12-06' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  intel_idle: Build fix

3 years agoMerge tag 'kbuild-fixes-v5.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 6 Dec 2020 18:31:39 +0000 (10:31 -0800)]
Merge tag 'kbuild-fixes-v5.10-2' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - Move -Wcast-align to W=3, which tends to be false-positive and there
   is no tree-wide solution.

 - Pass -fmacro-prefix-map to KBUILD_CPPFLAGS because it is a
   preprocessor option and makes sense for .S files as well.

 - Disable -gdwarf-2 for Clang's integrated assembler to avoid warnings.

 - Disable --orphan-handling=warn for LLD 10.0.1 to avoid warnings.

 - Fix undesirable line breaks in *.mod files.

* tag 'kbuild-fixes-v5.10-2' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  kbuild: avoid split lines in .mod files
  kbuild: Disable CONFIG_LD_ORPHAN_WARN for ld.lld 10.0.1
  kbuild: Hoist '--orphan-handling' into Kconfig
  Kbuild: do not emit debug info for assembly with LLVM_IAS=1
  kbuild: use -fmacro-prefix-map for .S sources
  Makefile.extrawarn: move -Wcast-align to W=3

3 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sun, 6 Dec 2020 18:20:59 +0000 (10:20 -0800)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "12 patches.

  Subsystems affected by this patch series: mm (memcg, zsmalloc, swap,
  mailmap, selftests, pagecache, hugetlb, pagemap), lib, and coredump"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  mm/mmap.c: fix mmap return value when vma is merged after call_mmap()
  hugetlb_cgroup: fix offline of hugetlb cgroup with reservations
  mm/filemap: add static for function __add_to_page_cache_locked
  userfaultfd: selftests: fix SIGSEGV if huge mmap fails
  tools/testing/selftests/vm: fix build error
  mailmap: add two more addresses of Uwe Kleine-König
  mm/swapfile: do not sleep with a spin lock held
  mm/zsmalloc.c: drop ZSMALLOC_PGTABLE_MAPPING
  mm: list_lru: set shrinker map bit when child nr_items is not zero
  mm: memcg/slab: fix obj_cgroup_charge() return value handling
  coredump: fix core_pattern parse error
  zlib: export S390 symbols for zlib modules

3 years agomm/mmap.c: fix mmap return value when vma is merged after call_mmap()
Liu Zixian [Sun, 6 Dec 2020 06:15:15 +0000 (22:15 -0800)]
mm/mmap.c: fix mmap return value when vma is merged after call_mmap()

On success, mmap should return the begin address of newly mapped area,
but patch "mm: mmap: merge vma after call_mmap() if possible" set
vm_start of newly merged vma to return value addr.  Users of mmap will
get wrong address if vma is merged after call_mmap().  We fix this by
moving the assignment to addr before merging vma.

We have a driver which changes vm_flags, and this bug is found by our
testcases.

Fixes: d70cec898324 ("mm: mmap: merge vma after call_mmap() if possible")
Signed-off-by: Liu Zixian <liuzixian4@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Hongxiang Lou <louhongxiang@huawei.com>
Cc: Hu Shiyuan <hushiyuan@huawei.com>
Cc: Matthew Wilcox <willy@infradead.org>
Link: https://lkml.kernel.org/r/20201203085350.22624-1-liuzixian4@huawei.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agohugetlb_cgroup: fix offline of hugetlb cgroup with reservations
Mike Kravetz [Sun, 6 Dec 2020 06:15:12 +0000 (22:15 -0800)]
hugetlb_cgroup: fix offline of hugetlb cgroup with reservations

Adrian Moreno was ruuning a kubernetes 1.19 + containerd/docker workload
using hugetlbfs.  In this environment the issue is reproduced by:

 - Start a simple pod that uses the recently added HugePages medium
   feature (pod yaml attached)

 - Start a DPDK app. It doesn't need to run successfully (as in transfer
   packets) nor interact with real hardware. It seems just initializing
   the EAL layer (which handles hugepage reservation and locking) is
   enough to trigger the issue

 - Delete the Pod (or let it "Complete").

This would result in a kworker thread going into a tight loop (top output):

   1425 root      20   0       0      0      0 R  99.7   0.0   5:22.45 kworker/28:7+cgroup_destroy

'perf top -g' reports:

  -   63.28%     0.01%  [kernel]                    [k] worker_thread
     - 49.97% worker_thread
        - 52.64% process_one_work
           - 62.08% css_killed_work_fn
              - hugetlb_cgroup_css_offline
                   41.52% _raw_spin_lock
                 - 2.82% _cond_resched
                      rcu_all_qs
                   2.66% PageHuge
        - 0.57% schedule
           - 0.57% __schedule

We are spinning in the do-while loop in hugetlb_cgroup_css_offline.
Worse yet, we are holding the master cgroup lock (cgroup_mutex) while
infinitely spinning.  Little else can be done on the system as the
cgroup_mutex can not be acquired.

Do note that the issue can be reproduced by simply offlining a hugetlb
cgroup containing pages with reservation counts.

The loop in hugetlb_cgroup_css_offline is moving page counts from the
cgroup being offlined to the parent cgroup.  This is done for each
hstate, and is repeated until hugetlb_cgroup_have_usage returns false.
The routine moving counts (hugetlb_cgroup_move_parent) is only moving
'usage' counts.  The routine hugetlb_cgroup_have_usage is checking for
both 'usage' and 'reservation' counts.  Discussion about what to do with
reservation counts when reparenting was discussed here:

https://lore.kernel.org/linux-kselftest/CAHS8izMFAYTgxym-Hzb_JmkTK1N_S9tGN71uS6MFV+R7swYu5A@mail.gmail.com/

The decision was made to leave a zombie cgroup for with reservation
counts.  Unfortunately, the code checking reservation counts was
incorrectly added to hugetlb_cgroup_have_usage.

To fix the issue, simply remove the check for reservation counts.  While
fixing this issue, a related bug in hugetlb_cgroup_css_offline was
noticed.  The hstate index is not reinitialized each time through the
do-while loop.  Fix this as well.

Fixes: 1adc4d419aa2 ("hugetlb_cgroup: add interface for charge/uncharge hugetlb reservations")
Reported-by: Adrian Moreno <amorenoz@redhat.com>
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Adrian Moreno <amorenoz@redhat.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20201203220242.158165-1-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm/filemap: add static for function __add_to_page_cache_locked
Alex Shi [Sun, 6 Dec 2020 06:15:09 +0000 (22:15 -0800)]
mm/filemap: add static for function __add_to_page_cache_locked

  mm/filemap.c:830:14: warning: no previous prototype for `__add_to_page_cache_locked' [-Wmissing-prototypes]

Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Link: https://lkml.kernel.org/r/1604661895-5495-1-git-send-email-alex.shi@linux.alibaba.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agouserfaultfd: selftests: fix SIGSEGV if huge mmap fails
Axel Rasmussen [Sun, 6 Dec 2020 06:15:05 +0000 (22:15 -0800)]
userfaultfd: selftests: fix SIGSEGV if huge mmap fails

The error handling in hugetlb_allocate_area() was incorrect for the
hugetlb_shared test case.

Previously the behavior was:

- mmap a hugetlb area
  - If this fails, set the pointer to NULL, and carry on
- mmap an alias of the same hugetlb fd
  - If this fails, munmap the original area

If the original mmap failed, it's likely the second one did too.  If
both failed, we'd blindly try to munmap a NULL pointer, causing a
SIGSEGV.  Instead, "goto fail" so we return before trying to mmap the
alias.

This issue can be hit "in real life" by forgetting to set
/proc/sys/vm/nr_hugepages (leaving it at 0), and then trying to run the
hugetlb_shared test.

Another small improvement is, when the original mmap fails, don't just
print "it failed": perror(), so we can see *why*.  :)

Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Joe Perches <joe@perches.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: David Alan Gilbert <dgilbert@redhat.com>
Link: https://lkml.kernel.org/r/20201204203443.2714693-1-axelrasmussen@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agotools/testing/selftests/vm: fix build error
Xingxing Su [Sun, 6 Dec 2020 06:15:02 +0000 (22:15 -0800)]
tools/testing/selftests/vm: fix build error

Only x86 and PowerPC implement the pkey-xxx.h, and an error was reported
when compiling protection_keys.c.

Add a Arch judgment to compile "protection_keys" in the Makefile.

If other arch implement this, add the arch name to the Makefile.
eg:
    ifneq (,$(findstring $(ARCH),powerpc mips ... ))

Following build errors:

    pkey-helpers.h:93:2: error: #error Architecture not supported
     #error Architecture not supported
    pkey-helpers.h:96:20: error: `PKEY_DISABLE_ACCESS' undeclared
     #define PKEY_MASK (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE)
                        ^
    protection_keys.c:218:45: error: `PKEY_DISABLE_WRITE' undeclared
     pkey_assert(flags & (PKEY_DISABLE_ACCESS | PKEY_DISABLE_WRITE));
                                                ^

Signed-off-by: Xingxing Su <suxingxing@loongson.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Shuah Khan <shuah@kernel.org>
Cc: Sandipan Das <sandipan@linux.ibm.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Brian Geffon <bgeffon@google.com>
Cc: Mina Almasry <almasrymina@google.com>
Link: https://lkml.kernel.org/r/1606826876-30656-1-git-send-email-suxingxing@loongson.cn
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomailmap: add two more addresses of Uwe Kleine-König
Uwe Kleine-König [Sun, 6 Dec 2020 06:14:58 +0000 (22:14 -0800)]
mailmap: add two more addresses of Uwe Kleine-König

This fixes attribution for the commits (among others)

 - d4097456cd1d ("video/framebuffer: move the probe func into
   .devinit.text in Blackfin LCD driver")

 - 0312e024d6cd ("mfd: mc13xxx: Add support for mc34708")

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: https://lkml.kernel.org/r/20201127213358.3440830-1-u.kleine-koenig@pengutronix.de
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm/swapfile: do not sleep with a spin lock held
Qian Cai [Sun, 6 Dec 2020 06:14:55 +0000 (22:14 -0800)]
mm/swapfile: do not sleep with a spin lock held

We can't call kvfree() with a spin lock held, so defer it.  Fixes a
might_sleep() runtime warning.

Fixes: 873d7bcfd066 ("mm/swapfile.c: use kvzalloc for swap_info_struct allocation")
Signed-off-by: Qian Cai <qcai@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20201202151549.10350-1-qcai@redhat.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm/zsmalloc.c: drop ZSMALLOC_PGTABLE_MAPPING
Minchan Kim [Sun, 6 Dec 2020 06:14:51 +0000 (22:14 -0800)]
mm/zsmalloc.c: drop ZSMALLOC_PGTABLE_MAPPING

While I was doing zram testing, I found sometimes decompression failed
since the compression buffer was corrupted.  With investigation, I found
below commit calls cond_resched unconditionally so it could make a
problem in atomic context if the task is reschedule.

  BUG: sleeping function called from invalid context at mm/vmalloc.c:108
  in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 946, name: memhog
  3 locks held by memhog/946:
   #0: ffff9d01d4b193e8 (&mm->mmap_lock#2){++++}-{4:4}, at: __mm_populate+0x103/0x160
   #1: ffffffffa3d53de0 (fs_reclaim){+.+.}-{0:0}, at: __alloc_pages_slowpath.constprop.0+0xa98/0x1160
   #2: ffff9d01d56b8110 (&zspage->lock){.+.+}-{3:3}, at: zs_map_object+0x8e/0x1f0
  CPU: 0 PID: 946 Comm: memhog Not tainted 5.9.3-00011-gc5bfc0287345-dirty #316
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1 04/01/2014
  Call Trace:
    unmap_kernel_range_noflush+0x2eb/0x350
    unmap_kernel_range+0x14/0x30
    zs_unmap_object+0xd5/0xe0
    zram_bvec_rw.isra.0+0x38c/0x8e0
    zram_rw_page+0x90/0x101
    bdev_write_page+0x92/0xe0
    __swap_writepage+0x94/0x4a0
    pageout+0xe3/0x3a0
    shrink_page_list+0xb94/0xd60
    shrink_inactive_list+0x158/0x460

We can fix this by removing the ZSMALLOC_PGTABLE_MAPPING feature (which
contains the offending calling code) from zsmalloc.

Even though this option showed some amount improvement(e.g., 30%) in
some arm32 platforms, it has been headache to maintain since it have
abused APIs[1](e.g., unmap_kernel_range in atomic context).

Since we are approaching to deprecate 32bit machines and already made
the config option available for only builtin build since v5.8, lastly it
has been not default option in zsmalloc, it's time to drop the option
for better maintenance.

[1] http://lore.kernel.org/linux-mm/20201105170249.387069-1-minchan@kernel.org

Fixes: e47110e90584 ("mm/vunmap: add cond_resched() in vunmap_pmd_range")
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Tony Lindgren <tony@atomide.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Harish Sriram <harish@linux.ibm.com>
Cc: Uladzislau Rezki <urezki@gmail.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20201117202916.GA3856507@google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm: list_lru: set shrinker map bit when child nr_items is not zero
Yang Shi [Sun, 6 Dec 2020 06:14:48 +0000 (22:14 -0800)]
mm: list_lru: set shrinker map bit when child nr_items is not zero

When investigating a slab cache bloat problem, significant amount of
negative dentry cache was seen, but confusingly they neither got shrunk
by reclaimer (the host has very tight memory) nor be shrunk by dropping
cache.  The vmcore shows there are over 14M negative dentry objects on
lru, but tracing result shows they were even not scanned at all.

Further investigation shows the memcg's vfs shrinker_map bit is not set.
So the reclaimer or dropping cache just skip calling vfs shrinker.  So
we have to reboot the hosts to get the memory back.

I didn't manage to come up with a reproducer in test environment, and
the problem can't be reproduced after rebooting.  But it seems there is
race between shrinker map bit clear and reparenting by code inspection.
The hypothesis is elaborated as below.

The memcg hierarchy on our production environment looks like:

                root
               /    \
          system   user

The main workloads are running under user slice's children, and it
creates and removes memcg frequently.  So reparenting happens very often
under user slice, but no task is under user slice directly.

So with the frequent reparenting and tight memory pressure, the below
hypothetical race condition may happen:

       CPU A                            CPU B
reparent
    dst->nr_items == 0
                                 shrinker:
                                     total_objects == 0
    add src->nr_items to dst
    set_bit
                                     return SHRINK_EMPTY
                                     clear_bit
child memcg offline
    replace child's kmemcg_id with
    parent's (in memcg_offline_kmem())
                                  list_lru_del() between shrinker runs
                                     see parent's kmemcg_id
                                     dec dst->nr_items
reparent again
    dst->nr_items may go negative
    due to concurrent list_lru_del()

                                 The second run of shrinker:
                                     read nr_items without any
                                     synchronization, so it may
                                     see intermediate negative
                                     nr_items then total_objects
                                     may return 0 coincidently

                                     keep the bit cleared
    dst->nr_items != 0
    skip set_bit
    add scr->nr_item to dst

After this point dst->nr_item may never go zero, so reparenting will not
set shrinker_map bit anymore.  And since there is no task under user
slice directly, so no new object will be added to its lru to set the
shrinker map bit either.  That bit is kept cleared forever.

How does list_lru_del() race with reparenting? It is because reparenting
replaces children's kmemcg_id to parent's without protecting from
nlru->lock, so list_lru_del() may see parent's kmemcg_id but actually
deleting items from child's lru, but dec'ing parent's nr_items, so the
parent's nr_items may go negative as commit 2788cf0c401c ("memcg:
reparent list_lrus and free kmemcg_id on css offline") says.

Since it is impossible that dst->nr_items goes negative and
src->nr_items goes zero at the same time, so it seems we could set the
shrinker map bit iff src->nr_items != 0.  We could synchronize
list_lru_count_one() and reparenting with nlru->lock, but it seems
checking src->nr_items in reparenting is the simplest and avoids lock
contention.

Fixes: fae91d6d8be5 ("mm/list_lru.c: set bit in memcg shrinker bitmap on first list_lru item appearance")
Suggested-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: <stable@vger.kernel.org> [4.19]
Link: https://lkml.kernel.org/r/20201202171749.264354-1-shy828301@gmail.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agomm: memcg/slab: fix obj_cgroup_charge() return value handling
Roman Gushchin [Sun, 6 Dec 2020 06:14:45 +0000 (22:14 -0800)]
mm: memcg/slab: fix obj_cgroup_charge() return value handling

Commit 10befea91b61 ("mm: memcg/slab: use a single set of kmem_caches
for all allocations") introduced a regression into the handling of the
obj_cgroup_charge() return value.  If a non-zero value is returned
(indicating of exceeding one of memory.max limits), the allocation
should fail, instead of falling back to non-accounted mode.

To make the code more readable, move memcg_slab_pre_alloc_hook() and
memcg_slab_post_alloc_hook() calling conditions into bodies of these
hooks.

Fixes: 10befea91b61 ("mm: memcg/slab: use a single set of kmem_caches for all allocations")
Signed-off-by: Roman Gushchin <guro@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20201127161828.GD840171@carbon.dhcp.thefacebook.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agocoredump: fix core_pattern parse error
Menglong Dong [Sun, 6 Dec 2020 06:14:42 +0000 (22:14 -0800)]
coredump: fix core_pattern parse error

'format_corename()' will splite 'core_pattern' on spaces when it is in
pipe mode, and take helper_argv[0] as the path to usermode executable.
It works fine in most cases.

However, if there is a space between '|' and '/file/path', such as
'| /usr/lib/systemd/systemd-coredump %P %u %g', then helper_argv[0] will
be parsed as '', and users will get a 'Core dump to | disabled'.

It is not friendly to users, as the pattern above was valid previously.
Fix this by ignoring the spaces between '|' and '/file/path'.

Fixes: 315c69261dd3 ("coredump: split pipe command whitespace before expanding template")
Signed-off-by: Menglong Dong <dong.menglong@zte.com.cn>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Paul Wise <pabs3@bonedaddy.net>
Cc: Jakub Wilk <jwilk@jwilk.net> [https://bugs.debian.org/924398]
Cc: Neil Horman <nhorman@tuxdriver.com>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/5fb62870.1c69fb81.8ef5d.af76@mx.google.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agozlib: export S390 symbols for zlib modules
Randy Dunlap [Sun, 6 Dec 2020 06:14:38 +0000 (22:14 -0800)]
zlib: export S390 symbols for zlib modules

Fix build errors when ZLIB_INFLATE=m and ZLIB_DEFLATE=m and ZLIB_DFLTCC=y
by exporting the 2 needed symbols in dfltcc_inflate.c.

Fixes these build errors:

  ERROR: modpost: "dfltcc_inflate" [lib/zlib_inflate/zlib_inflate.ko] undefined!
  ERROR: modpost: "dfltcc_can_inflate" [lib/zlib_inflate/zlib_inflate.ko] undefined!

Fixes: 126196100063 ("lib/zlib: add s390 hardware support for kernel zlib_inflate")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Ilya Leoshkevich <iii@linux.ibm.com>
Cc: Mikhail Zaslonko <zaslonko@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Link: https://lkml.kernel.org/r/20201123191712.4882-1-rdunlap@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
3 years agokbuild: avoid split lines in .mod files
Masahiro Yamada [Thu, 3 Dec 2020 17:55:51 +0000 (02:55 +0900)]
kbuild: avoid split lines in .mod files

"xargs echo" is not a safe way to remove line breaks because the input
may exceed the command line limit and xargs may break it up into
multiple invocations of echo. This should never happen because
scripts/gen_autoksyms.sh expects all undefined symbols are placed in
the second line of .mod files.

One possible way is to replace "xargs echo" with
"sed ':x;N;$!bx;s/\n/ /g'" or something, but I rewrote the code by
using awk because it is more readable.

This issue was reported by Sami Tolvanen; in his Clang LTO patch set,
$(multi-used-m) is no longer an ELF object, but a thin archive that
contains LLVM bitcode files. llvm-nm prints out symbols for each
archive member separately, which results a lot of dupications, in some
places, beyond the system-defined limit.

This problem must be fixed irrespective of LTO, and we must ensure
zero possibility of having this issue.

Link: https://lkml.org/lkml/2020/12/1/1658
Reported-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
3 years agoRevert "mei: virtio: virtualization frontend driver"
Michael S. Tsirkin [Sat, 5 Dec 2020 19:38:46 +0000 (14:38 -0500)]
Revert "mei: virtio: virtualization frontend driver"

This reverts commit d162219c655c8cf8003128a13840d6c1e183fb80.

The device uses a VIRTIO device ID out of a not-for-production range.
Releasing Linux using an ID out of this range will make it conflict with
development setups. An official request to reserve an ID for an MEI
device is yet to be submitted to the virtio TC, thus there's no chance
it will be reserved and fixed in time before the next release.

Once requested it usually takes 2-3 weeks to land in the spec, which
means the device can be supported with the official ID in the next Linux
version if contributors act quickly.

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Cc: Tomas Winkler <tomas.winkler@intel.com>
Cc: Alexander Usyskin <alexander.usyskin@intel.com>
Cc: Wang Yu <yu1.wang@intel.com>
Cc: Liu Shuo <shuo.a.liu@intel.com>
Link: https://lore.kernel.org/r/20201205193625.469773-1-mst@redhat.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
3 years agox86/sev-es: Use new for_each_insn_prefix() macro to loop over prefixes bytes
Masami Hiramatsu [Thu, 3 Dec 2020 04:51:01 +0000 (13:51 +0900)]
x86/sev-es: Use new for_each_insn_prefix() macro to loop over prefixes bytes

Since insn.prefixes.nbytes can be bigger than the size of
insn.prefixes.bytes[] when a prefix is repeated, the proper
check must be:

  insn.prefixes.bytes[i] != 0 and i < 4

instead of using insn.prefixes.nbytes. Use the new
for_each_insn_prefix() macro which does it correctly.

Debugged by Kees Cook <keescook@chromium.org>.

 [ bp: Massage commit message. ]

Fixes: 25189d08e516 ("x86/sev-es: Add support for handling IOIO exceptions")
Reported-by: syzbot+9b64b619f10f19d19a7c@syzkaller.appspotmail.com
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/160697106089.3146288.2052422845039649176.stgit@devnote2
3 years agox86/insn-eval: Use new for_each_insn_prefix() macro to loop over prefixes bytes
Masami Hiramatsu [Thu, 3 Dec 2020 04:50:50 +0000 (13:50 +0900)]
x86/insn-eval: Use new for_each_insn_prefix() macro to loop over prefixes bytes

Since insn.prefixes.nbytes can be bigger than the size of
insn.prefixes.bytes[] when a prefix is repeated, the proper check must
be

  insn.prefixes.bytes[i] != 0 and i < 4

instead of using insn.prefixes.nbytes. Use the new
for_each_insn_prefix() macro which does it correctly.

Debugged by Kees Cook <keescook@chromium.org>.

 [ bp: Massage commit message. ]

Fixes: 32d0b95300db ("x86/insn-eval: Add utility functions to get segment selector")
Reported-by: syzbot+9b64b619f10f19d19a7c@syzkaller.appspotmail.com
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/160697104969.3146288.16329307586428270032.stgit@devnote2
3 years agox86/uprobes: Do not use prefixes.nbytes when looping over prefixes.bytes
Masami Hiramatsu [Thu, 3 Dec 2020 04:50:37 +0000 (13:50 +0900)]
x86/uprobes: Do not use prefixes.nbytes when looping over prefixes.bytes

Since insn.prefixes.nbytes can be bigger than the size of
insn.prefixes.bytes[] when a prefix is repeated, the proper check must
be

  insn.prefixes.bytes[i] != 0 and i < 4

instead of using insn.prefixes.nbytes.

Introduce a for_each_insn_prefix() macro for this purpose. Debugged by
Kees Cook <keescook@chromium.org>.

 [ bp: Massage commit message, sync with the respective header in tools/
   and drop "we". ]

Fixes: 2b1444983508 ("uprobes, mm, x86: Add the ability to install and remove uprobes breakpoints")
Reported-by: syzbot+9b64b619f10f19d19a7c@syzkaller.appspotmail.com
Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/160697103739.3146288.7437620795200799020.stgit@devnote2
3 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Sun, 6 Dec 2020 00:16:34 +0000 (16:16 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:
 "A fix for 'RETRIGEN' handling in Atmel touch controllers that was
  causing lost interrupts on systems using edge-triggered interrupts, a
  quirk for i8042 driver, and a couple more fixes."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: atmel_mxt_ts - fix lost interrupts
  Input: xpad - support Ardwiino Controllers
  Input: i8042 - add ByteSpeed touchpad to noloop table
  Input: i8042 - fix error return code in i8042_setup_aux()
  Input: soc_button_array - add missing include

3 years agoMerge branch 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sat, 5 Dec 2020 23:27:22 +0000 (15:27 -0800)]
Merge branch 'i2c/for-current' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Some more I2C driver updates. IMX updates are a tad bigger, but not
  exceptionally big"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: mlxbf: Fix the return check of devm_ioremap and ioremap
  i2c: mlxbf: select CONFIG_I2C_SLAVE
  i2c: imx: Don't generate STOP condition if arbitration has been lost
  i2c: imx: Check for I2SR_IAL after every byte
  i2c: imx: Fix reset of I2SR_IAL flag
  i2c: qcom: Fix IRQ error misassignement
  i2c: qup: Fix error return code in qup_i2c_bam_schedule_desc()

3 years agoMerge tag 'block-5.10-2020-12-05' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 5 Dec 2020 22:45:30 +0000 (14:45 -0800)]
Merge tag 'block-5.10-2020-12-05' of git://git.kernel.dk/linux-block

Pull block fix from Jens Axboe:
 "Single fix for an issue with chunk_sectors and stacked devices"

* tag 'block-5.10-2020-12-05' of git://git.kernel.dk/linux-block:
  block: use gcd() to fix chunk_sectors limit stacking

3 years agoMerge tag 'io_uring-5.10-2020-12-05' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 5 Dec 2020 22:39:59 +0000 (14:39 -0800)]
Merge tag 'io_uring-5.10-2020-12-05' of git://git.kernel.dk/linux-block

Pull io_uring fix from Jens Axboe:
 "Just a small fix this time, for an issue with 32-bit compat apps and
  buffer selection with recvmsg"

* tag 'io_uring-5.10-2020-12-05' of git://git.kernel.dk/linux-block:
  io_uring: fix recvmsg setup with compat buf-select

3 years agoMerge tag 'powerpc-5.10-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sat, 5 Dec 2020 19:16:21 +0000 (11:16 -0800)]
Merge tag 'powerpc-5.10-5' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "Some more powerpc fixes for 5.10:

   - Three commits fixing possible missed TLB invalidations for
     multi-threaded processes when CPUs are hotplugged in and out.

   - A fix for a host crash triggerable by host userspace (qemu) in KVM
     on Power9.

   - A fix for a host crash in machine check handling when running HPT
     guests on a HPT host.

   - One commit fixing potential missed TLB invalidations when using the
     hash MMU on Power9 or later.

   - A regression fix for machines with CPUs on node 0 but no memory.

  Thanks to Aneesh Kumar K.V, Cédric Le Goater, Greg Kurz, Milan
  Mohanty, Milton Miller, Nicholas Piggin, Paul Mackerras, and Srikar
  Dronamraju"

* tag 'powerpc-5.10-5' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/64s/powernv: Fix memory corruption when saving SLB entries on MCE
  KVM: PPC: Book3S HV: XIVE: Fix vCPU id sanity check
  powerpc/numa: Fix a regression on memoryless node 0
  powerpc/64s: Trim offlined CPUs from mm_cpumasks
  kernel/cpu: add arch override for clear_tasks_mm_cpumask() mm handling
  powerpc/64s/pseries: Fix hash tlbiel_all_isa300 for guest kernels
  powerpc/64s: Fix hash ISA v3.0 TLBIEL instruction generation

3 years agoMerge tag '5.10-rc6-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sat, 5 Dec 2020 19:09:07 +0000 (11:09 -0800)]
Merge tag '5.10-rc6-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fixes from Steve French:
 "Three smb3 fixes (two for stable) fixing

   - a null pointer issue in a DFS error path

   - a problem with excessive padding when mounted with "idsfromsid"
     causing owner fields to get corrupted

   - a more recent problem with compounded reparse point query found in
     testing to the Linux kernel server"

* tag '5.10-rc6-smb3-fixes-part2' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: refactor create_sd_buf() and and avoid corrupting the buffer
  cifs: add NULL check for ses->tcon_ipc
  smb3: set COMPOUND_FID to FileID field of subsequent compound request

3 years agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 5 Dec 2020 18:59:21 +0000 (10:59 -0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Four small fixes in two drivers.

  The mpt3sas fixes are all problems with timeout under unusual
  conditions, and the storvsc is a missed incoming packet validation
  and a missed error return"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: mpt3sas: Increase IOCInit request timeout to 30s
  scsi: mpt3sas: Fix ioctl timeout
  scsi: storvsc: Validate length of incoming packet in storvsc_on_channel_callback()
  scsi: storvsc: Fix error return in storvsc_probe()

3 years agoMerge tag 'for-5.10/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 5 Dec 2020 18:51:25 +0000 (10:51 -0800)]
Merge tag 'for-5.10/dm-fixes-2' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull fix for device mapper fixes from Mike Snitzer:
 "Apologies for the glaring bug I introduced with my previous pull
  request!

  Fix incorrect branching at top of blk_max_size_offset()"

* tag 'for-5.10/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  block: fix incorrect branching in blk_max_size_offset()

3 years agoi2c: mlxbf: Fix the return check of devm_ioremap and ioremap
Wang Xiaojun [Thu, 3 Dec 2020 01:46:47 +0000 (20:46 -0500)]
i2c: mlxbf: Fix the return check of devm_ioremap and ioremap

devm_ioremap and ioremap may return NULL which cannot be checked
by IS_ERR.

Signed-off-by: Wang Xiaojun <wangxiaojun11@huawei.com>
Reported-by: Hulk Robot <hulkci@huawei.com>
Acked-by: Khalil Blaiech <kblaiech@nvidia.com>
Signed-off-by: Wolfram Sang <wsa@kernel.org>
3 years agoi2c: mlxbf: select CONFIG_I2C_SLAVE
Arnd Bergmann [Thu, 3 Dec 2020 22:32:50 +0000 (23:32 +0100)]
i2c: mlxbf: select CONFIG_I2C_SLAVE

If this is not enabled, the interfaces used in this driver do not work:

drivers/i2c/busses/i2c-mlxbf.c:1888:3: error: implicit declaration of function 'i2c_slave_event' [-Werror,-Wimplicit-function-declaration]
                i2c_slave_event(slave, I2C_SLAVE_WRITE_REQUESTED, &value);
                ^
drivers/i2c/busses/i2c-mlxbf.c:1888:26: error: use of undeclared identifier 'I2C_SLAVE_WRITE_REQUESTED'
                i2c_slave_event(slave, I2C_SLAVE_WRITE_REQUESTED, &value);
                                       ^
drivers/i2c/busses/i2c-mlxbf.c:1890:32: error: use of undeclared identifier 'I2C_SLAVE_WRITE_RECEIVED'
                ret = i2c_slave_event(slave, I2C_SLAVE_WRITE_RECEIVED,
                                             ^
drivers/i2c/busses/i2c-mlxbf.c:1892:26: error: use of undeclared identifier 'I2C_SLAVE_STOP'
                i2c_slave_event(slave, I2C_SLAVE_STOP, &value);
                                       ^

Fixes: b5b5b32081cd ("i2c: mlxbf: I2C SMBus driver for Mellanox BlueField SoC")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Khalil Blaiech <kblaiech@nvidia.com>
Signed-off-by: Wolfram Sang <wsa@kernel.org>
3 years agoblock: fix incorrect branching in blk_max_size_offset()
Mike Snitzer [Fri, 4 Dec 2020 22:21:03 +0000 (17:21 -0500)]
block: fix incorrect branching in blk_max_size_offset()

If non-zero 'chunk_sectors' is passed in to blk_max_size_offset() that
override will be incorrectly ignored.

Old blk_max_size_offset() branching, prior to commit 3ee16db390b4,
must be used only if passed 'chunk_sectors' override is zero.

Fixes: 3ee16db390b4 ("dm: fix IO splitting")
Cc: stable@vger.kernel.org # 5.9
Reported-by: John Dorminy <jdorminy@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
3 years agoMerge tag 'for-5.10/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device...
Linus Torvalds [Fri, 4 Dec 2020 21:28:39 +0000 (13:28 -0800)]
Merge tag 'for-5.10/dm-fixes' of git://git./linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - Fix DM's bio splitting changes that were made during v5.9. This
   restores splitting in terms of varied per-target ti->max_io_len
   rather than use block core's single stacked 'chunk_sectors' limit.

 - Like DM crypt, update DM integrity to not use crypto drivers that
   have CRYPTO_ALG_ALLOCATES_MEMORY set.

 - Fix DM writecache target's argument parsing and status display.

 - Remove needless BUG() from dm writecache's persistent_memory_claim()

 - Remove old gcc workaround in DM cache target's block_div() for ARM
   link errors now that gcc >= 4.9 is required.

 - Fix RCU locking in dm_blk_report_zones and dm_dax_zero_page_range.

 - Remove old, and now frowned upon, BUG_ON(in_interrupt()) in
   dm_table_event().

 - Remove invalid sparse annotations from dm_prepare_ioctl() and
   dm_unprepare_ioctl().

* tag 'for-5.10/dm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm: remove invalid sparse __acquires and __releases annotations
  dm: fix double RCU unlock in dm_dax_zero_page_range() error path
  dm: fix IO splitting
  dm writecache: remove BUG() and fail gracefully instead
  dm table: Remove BUG_ON(in_interrupt())
  dm: fix bug with RCU locking in dm_blk_report_zones
  Revert "dm cache: fix arm link errors with inline"
  dm writecache: fix the maximum number of arguments
  dm writecache: advance the number of arguments when reporting max_age
  dm integrity: don't use drivers that have CRYPTO_ALG_ALLOCATES_MEMORY

3 years agodm: remove invalid sparse __acquires and __releases annotations
Mike Snitzer [Fri, 4 Dec 2020 20:25:18 +0000 (15:25 -0500)]
dm: remove invalid sparse __acquires and __releases annotations

Fixes sparse warnings:
drivers/md/dm.c:508:12: warning: context imbalance in 'dm_prepare_ioctl' - wrong count at exit
drivers/md/dm.c:543:13: warning: context imbalance in 'dm_unprepare_ioctl' - wrong count at exit

Fixes: 971888c46993f ("dm: hold DM table for duration of ioctl rather than use blkdev_get")
Cc: stable@vger.kernel.org
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
3 years agodm: fix double RCU unlock in dm_dax_zero_page_range() error path
Mike Snitzer [Fri, 4 Dec 2020 20:19:27 +0000 (15:19 -0500)]
dm: fix double RCU unlock in dm_dax_zero_page_range() error path

Remove redundant dm_put_live_table() in dm_dax_zero_page_range() error
path to fix sparse warning:
drivers/md/dm.c:1208:9: warning: context imbalance in 'dm_dax_zero_page_range' - unexpected unlock

Fixes: cdf6cdcd3b99a ("dm,dax: Add dax zero_page_range operation")
Cc: stable@vger.kernel.org
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
3 years agodm: fix IO splitting
Mike Snitzer [Mon, 30 Nov 2020 15:57:43 +0000 (10:57 -0500)]
dm: fix IO splitting

Commit 882ec4e609c1 ("dm table: stack 'chunk_sectors' limit to account
for target-specific splitting") caused a couple regressions:
1) Using lcm_not_zero() when stacking chunk_sectors was a bug because
   chunk_sectors must reflect the most limited of all devices in the
   IO stack.
2) DM targets that set max_io_len but that do _not_ provide an
   .iterate_devices method no longer had there IO split properly.

And commit 5091cdec56fa ("dm: change max_io_len() to use
blk_max_size_offset()") also caused a regression where DM no longer
supported varied (per target) IO splitting. The implication being the
potential for severely reduced performance for IO stacks that use a DM
target like dm-cache to hide performance limitations of a slower
device (e.g. one that requires 4K IO splitting).

Coming full circle: Fix all these issues by discontinuing stacking
chunk_sectors up using ti->max_io_len in dm_calculate_queue_limits(),
add optional chunk_sectors override argument to blk_max_size_offset()
and update DM's max_io_len() to pass ti->max_io_len to its
blk_max_size_offset() call.

Passing in an optional chunk_sectors override to blk_max_size_offset()
allows for code reuse of block's centralized calculation for max IO
size based on provided offset and split boundary.

Fixes: 882ec4e609c1 ("dm table: stack 'chunk_sectors' limit to account for target-specific splitting")
Fixes: 5091cdec56fa ("dm: change max_io_len() to use blk_max_size_offset()")
Cc: stable@vger.kernel.org
Reported-by: John Dorminy <jdorminy@redhat.com>
Reported-by: Bruce Johnston <bjohnsto@redhat.com>
Reported-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: John Dorminy <jdorminy@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'drm-fixes-2020-12-04' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 4 Dec 2020 17:25:22 +0000 (09:25 -0800)]
Merge tag 'drm-fixes-2020-12-04' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "This week's regular fixes.

  i915 has fixes for a few races, use-after-free, and gpu hangs. Tegra
  just has some minor fixes that I didn't see much point in hanging on
  to. The nouveau fix is for all pre-nv50 cards and was reported a few
  times. Otherwise it's just some amdgpu, and a few misc fixes.

  Summary:

  amdgpu:
   - SMU11 manual fan fix
   - Renoir display clock fix
   - VCN3 dynamic powergating fix

  i915:
   - Program mocs:63 for cache eviction on gen9 (Chris)
   - Protect context lifetime with RCU (Chris)
   - Split the breadcrumb spinlock between global and contexts (Chris)
   - Retain default context state across shrinking (Venkata)
   - Limit frequency drop to RPe on parking (Chris)
   - Return earlier from intel_modeset_init() without display (Jani)
   - Defer initial modeset until after GGTT is initialized (Chris)

  nouveau:
   - pre-nv50 regression fix

  rockchip:
   - uninitialised LVDS property fix

  omap:
   - bridge fix

  panel:
   - race fix

  mxsfb:
   - fence sync fix
   - modifiers fix

  tegra:
   - idr init fix
   - sor fixes
   - output/of cleanup fix"

* tag 'drm-fixes-2020-12-04' of git://anongit.freedesktop.org/drm/drm: (22 commits)
  drm/amdgpu/vcn3.0: remove old DPG workaround
  drm/amdgpu/vcn3.0: stall DPG when WPTR/RPTR reset
  drm/amd/display: Init clock value by current vbios CLKs
  drm/amdgpu/pm/smu11: Fix fan set speed bug
  drm/i915/display: Defer initial modeset until after GGTT is initialised
  drm/i915/display: return earlier from intel_modeset_init() without display
  drm/i915/gt: Limit frequency drop to RPe on parking
  drm/i915/gt: Retain default context state across shrinking
  drm/i915/gt: Split the breadcrumb spinlock between global and contexts
  drm/i915/gt: Protect context lifetime with RCU
  drm/i915/gt: Program mocs:63 for cache eviction on gen9
  drm/omap: sdi: fix bridge enable/disable
  drm/panel: sony-acx565akm: Fix race condition in probe
  drm/rockchip: Avoid uninitialized use of endpoint id in LVDS
  drm/tegra: sor: Disable clocks on error in tegra_sor_init()
  drm/nouveau: make sure ret is initialized in nouveau_ttm_io_mem_reserve
  drm: mxsfb: Implement .format_mod_supported
  drm: mxsfb: fix fence synchronization
  drm/tegra: output: Do not put OF node twice
  drm/tegra: replace idr_init() by idr_init_base()
  ...

3 years agotty: Fix ->session locking
Jann Horn [Thu, 3 Dec 2020 01:25:05 +0000 (02:25 +0100)]
tty: Fix ->session locking

Currently, locking of ->session is very inconsistent; most places
protect it using the legacy tty mutex, but disassociate_ctty(),
__do_SAK(), tiocspgrp() and tiocgsid() don't.
Two of the writers hold the ctrl_lock (because they already need it for
->pgrp), but __proc_set_tty() doesn't do that yet.

On a PREEMPT=y system, an unprivileged user can theoretically abuse
this broken locking to read 4 bytes of freed memory via TIOCGSID if
tiocgsid() is preempted long enough at the right point. (Other things
might also go wrong, especially if root-only ioctls are involved; I'm
not sure about that.)

Change the locking on ->session such that:

 - tty_lock() is held by all writers: By making disassociate_ctty()
   hold it. This should be fine because the same lock can already be
   taken through the call to tty_vhangup_session().
   The tricky part is that we need to shorten the area covered by
   siglock to be able to take tty_lock() without ugly retry logic; as
   far as I can tell, this should be fine, since nothing in the
   signal_struct is touched in the `if (tty)` branch.
 - ctrl_lock is held by all writers: By changing __proc_set_tty() to
   hold the lock a little longer.
 - All readers that aren't holding tty_lock() hold ctrl_lock: By
   adding locking to tiocgsid() and __do_SAK(), and expanding the area
   covered by ctrl_lock in tiocspgrp().

Cc: stable@kernel.org
Signed-off-by: Jann Horn <jannh@google.com>
Reviewed-by: Jiri Slaby <jirislaby@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>