sfrench/cifs-2.6.git
18 months agoBtrfs: add missing inode version, ctime and mtime updates when punching hole
Filipe Manana [Wed, 19 Jun 2019 12:05:50 +0000 (13:05 +0100)]
Btrfs: add missing inode version, ctime and mtime updates when punching hole

If the range for which we are punching a hole covers only part of a page,
we end up updating the inode item but we skip the update of the inode's
iversion, mtime and ctime. Fix that by ensuring we update those properties
of the inode.

A patch for fstests test case generic/059 that tests this as been sent
along with this fix.

Fixes: 2aaa66558172b0 ("Btrfs: add hole punching")
Fixes: e8c1c76e804b18 ("Btrfs: add missing inode update when punching hole")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agoBtrfs: fix fsync not persisting dentry deletions due to inode evictions
Filipe Manana [Wed, 19 Jun 2019 12:05:39 +0000 (13:05 +0100)]
Btrfs: fix fsync not persisting dentry deletions due to inode evictions

In order to avoid searches on a log tree when unlinking an inode, we check
if the inode being unlinked was logged in the current transaction, as well
as the inode of its parent directory. When any of the inodes are logged,
we proceed to delete directory items and inode reference items from the
log, to ensure that if a subsequent fsync of only the inode being unlinked
or only of the parent directory when the other is not fsync'ed as well,
does not result in the entry still existing after a power failure.

That check however is not reliable when one of the inodes involved (the
one being unlinked or its parent directory's inode) is evicted, since the
logged_trans field is transient, that is, it is not stored on disk, so it
is lost when the inode is evicted and loaded into memory again (which is
set to zero on load). As a consequence the checks currently being done by
btrfs_del_dir_entries_in_log() and btrfs_del_inode_ref_in_log() always
return true if the inode was evicted before, regardless of the inode
having been logged or not before (and in the current transaction), this
results in the dentry being unlinked still existing after a log replay
if after the unlink operation only one of the inodes involved is fsync'ed.

Example:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ mkdir /mnt/dir
  $ touch /mnt/dir/foo
  $ xfs_io -c fsync /mnt/dir/foo

  # Keep an open file descriptor on our directory while we evict inodes.
  # We just want to evict the file's inode, the directory's inode must not
  # be evicted.
  $ ( cd /mnt/dir; while true; do :; done ) &
  $ pid=$!

  # Wait a bit to give time to background process to chdir to our test
  # directory.
  $ sleep 0.5

  # Trigger eviction of the file's inode.
  $ echo 2 > /proc/sys/vm/drop_caches

  # Unlink our file and fsync the parent directory. After a power failure
  # we don't expect to see the file anymore, since we fsync'ed the parent
  # directory.
  $ rm -f $SCRATCH_MNT/dir/foo
  $ xfs_io -c fsync /mnt/dir

  <power failure>

  $ mount /dev/sdb /mnt
  $ ls /mnt/dir
  foo
  $
   --> file still there, unlink not persisted despite explicit fsync on dir

Fix this by checking if the inode has the full_sync bit set in its runtime
flags as well, since that bit is set everytime an inode is loaded from
disk, or for other less common cases such as after a shrinking truncate
or failure to allocate extent maps for holes, and gets cleared after the
first fsync. Also consider the inode as possibly logged only if it was
last modified in the current transaction (besides having the full_fsync
flag set).

Fixes: 3a5f1d458ad161 ("Btrfs: Optimize btree walking while logging inodes")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Use btrfs_get_io_geometry appropriately
Nikolay Borisov [Mon, 3 Jun 2019 09:05:05 +0000 (12:05 +0300)]
btrfs: Use btrfs_get_io_geometry appropriately

Presently btrfs_map_block is used not only to do everything necessary to
map a bio to the underlying allocation profile but it's also used to
identify how much data could be written based on btrfs' stripe logic
without actually submitting anything. This is achieved by passing NULL
for 'bbio_ret' parameter.

This patch refactors all callers that require just the mapping length
by switching them to using btrfs_io_geometry instead of calling
btrfs_map_block with a special NULL value for 'bbio_ret'. No functional
change.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Introduce btrfs_io_geometry infrastructure
Nikolay Borisov [Mon, 3 Jun 2019 09:05:03 +0000 (12:05 +0300)]
btrfs: Introduce btrfs_io_geometry infrastructure

Add a structure that holds various parameters for IO calculations and a
helper that fills the values. This will help further refactoring and
reduction of functions that in some way open-coded the calculations.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: improve messages when updating feature flags
David Sterba [Thu, 13 Jun 2019 15:55:03 +0000 (17:55 +0200)]
btrfs: improve messages when updating feature flags

Currently the messages printed after setting an incompat feature are
cryptis, we can easily make it better as the textual description is
passed to the helpers. Old:

  setting 128 feature flag

updated:

  setting incompat feature flag for RAID56 (0x80)

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: shut up bogus -Wmaybe-uninitialized warning
Arnd Bergmann [Mon, 17 Jun 2019 11:07:28 +0000 (13:07 +0200)]
btrfs: shut up bogus -Wmaybe-uninitialized warning

gcc sometimes can't determine whether a variable has been initialized
when both the initialization and the use are conditional:

fs/btrfs/props.c: In function 'inherit_props':
fs/btrfs/props.c:389:4: error: 'num_bytes' may be used uninitialized in this function [-Werror=maybe-uninitialized]
    btrfs_block_rsv_release(fs_info, trans->block_rsv,

This code is fine. Unfortunately, I cannot think of a good way to
rephrase it in a way that makes gcc understand this, so I add a bogus
initialization the way one should not.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ gcc 8 and 9 don't emit the warning ]
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agoBtrfs: prevent send failures and crashes due to concurrent relocation
Filipe Manana [Mon, 22 Apr 2019 15:44:09 +0000 (16:44 +0100)]
Btrfs: prevent send failures and crashes due to concurrent relocation

Send always operates on read-only trees and always expected that while it
is in progress, nothing changes in those trees. Due to that expectation
and the fact that send is a read-only operation, it operates on commit
roots and does not hold transaction handles. However relocation can COW
nodes and leafs from read-only trees, which can cause unexpected failures
and crashes (hitting BUG_ONs). while send using a node/leaf, it gets
COWed, the transaction used to COW it is committed, a new transaction
starts, the extent previously used for that node/leaf gets allocated,
possibly for another tree, and the respective extent buffer' content
changes while send is still using it. When this happens send normally
fails with EIO being returned to user space and messages like the
following are found in dmesg/syslog:

  [ 3408.699121] BTRFS error (device sdc): parent transid verify failed on 58703872 wanted 250 found 253
  [ 3441.523123] BTRFS error (device sdc): did not find backref in send_root. inode=63211, offset=0, disk_byte=5222825984 found extent=5222825984

Other times, less often, we hit a BUG_ON() because an extent buffer that
send is using used to be a node, and while send is still using it, it
got COWed and got reused as a leaf while send is still using, producing
the following trace:

 [ 3478.466280] ------------[ cut here ]------------
 [ 3478.466282] kernel BUG at fs/btrfs/ctree.c:1806!
 [ 3478.466965] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC PTI
 [ 3478.467635] CPU: 0 PID: 2165 Comm: btrfs Not tainted 5.0.0-btrfs-next-46 #1
 [ 3478.468311] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.2-0-gf9626ccb91-prebuilt.qemu-project.org 04/01/2014
 [ 3478.469681] RIP: 0010:read_node_slot+0x122/0x130 [btrfs]
 (...)
 [ 3478.471758] RSP: 0018:ffffa437826bfaa0 EFLAGS: 00010246
 [ 3478.472457] RAX: ffff961416ed7000 RBX: 000000000000003d RCX: 0000000000000002
 [ 3478.473151] RDX: 000000000000003d RSI: ffff96141e387408 RDI: ffff961599b30000
 [ 3478.473837] RBP: ffffa437826bfb8e R08: 0000000000000001 R09: ffffa437826bfb8e
 [ 3478.474515] R10: ffffa437826bfa70 R11: 0000000000000000 R12: ffff9614385c8708
 [ 3478.475186] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
 [ 3478.475840] FS:  00007f8e0e9cc8c0(0000) GS:ffff9615b6a00000(0000) knlGS:0000000000000000
 [ 3478.476489] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 [ 3478.477127] CR2: 00007f98b67a056e CR3: 0000000005df6005 CR4: 00000000003606f0
 [ 3478.477762] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 [ 3478.478385] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 [ 3478.479003] Call Trace:
 [ 3478.479600]  ? do_raw_spin_unlock+0x49/0xc0
 [ 3478.480202]  tree_advance+0x173/0x1d0 [btrfs]
 [ 3478.480810]  btrfs_compare_trees+0x30c/0x690 [btrfs]
 [ 3478.481388]  ? process_extent+0x1280/0x1280 [btrfs]
 [ 3478.481954]  btrfs_ioctl_send+0x1037/0x1270 [btrfs]
 [ 3478.482510]  _btrfs_ioctl_send+0x80/0x110 [btrfs]
 [ 3478.483062]  btrfs_ioctl+0x13fe/0x3120 [btrfs]
 [ 3478.483581]  ? rq_clock_task+0x2e/0x60
 [ 3478.484086]  ? wake_up_new_task+0x1f3/0x370
 [ 3478.484582]  ? do_vfs_ioctl+0xa2/0x6f0
 [ 3478.485075]  ? btrfs_ioctl_get_supported_features+0x30/0x30 [btrfs]
 [ 3478.485552]  do_vfs_ioctl+0xa2/0x6f0
 [ 3478.486016]  ? __fget+0x113/0x200
 [ 3478.486467]  ksys_ioctl+0x70/0x80
 [ 3478.486911]  __x64_sys_ioctl+0x16/0x20
 [ 3478.487337]  do_syscall_64+0x60/0x1b0
 [ 3478.487751]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
 [ 3478.488159] RIP: 0033:0x7f8e0d7d4dd7
 (...)
 [ 3478.489349] RSP: 002b:00007ffcf6fb4908 EFLAGS: 00000202 ORIG_RAX: 0000000000000010
 [ 3478.489742] RAX: ffffffffffffffda RBX: 0000000000000105 RCX: 00007f8e0d7d4dd7
 [ 3478.490142] RDX: 00007ffcf6fb4990 RSI: 0000000040489426 RDI: 0000000000000005
 [ 3478.490548] RBP: 0000000000000005 R08: 00007f8e0d6f3700 R09: 00007f8e0d6f3700
 [ 3478.490953] R10: 00007f8e0d6f39d0 R11: 0000000000000202 R12: 0000000000000005
 [ 3478.491343] R13: 00005624e0780020 R14: 0000000000000000 R15: 0000000000000001
 (...)
 [ 3478.493352] ---[ end trace d5f537302be4f8c8 ]---

Another possibility, much less likely to happen, is that send will not
fail but the contents of the stream it produces may not be correct.

To avoid this, do not allow send and relocation (balance) to run in
parallel. In the long term the goal is to allow for both to be able to
run concurrently without any problems, but that will take a significant
effort in development and testing.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: document BTRFS_MAX_MIRRORS
David Sterba [Fri, 31 May 2019 15:22:59 +0000 (17:22 +0200)]
btrfs: document BTRFS_MAX_MIRRORS

The real meaning of that constant is not clear from the context due to
the target device inclusion.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use mask for RAID56 profiles
David Sterba [Fri, 31 May 2019 14:54:26 +0000 (16:54 +0200)]
btrfs: use mask for RAID56 profiles

We don't need to enumerate the profiles, use the mask for consistency.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: add mask for all RAID1 types
David Sterba [Fri, 31 May 2019 13:39:31 +0000 (15:39 +0200)]
btrfs: add mask for all RAID1 types

Preparatory patch for additional RAID1 profiles with more copies. The
mask will contain 3-copy and 4-copy, most of the checks for plain RAID1
work the same for the other profiles.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: qgroup: Don't hold qgroup_ioctl_lock in btrfs_qgroup_inherit()
Qu Wenruo [Thu, 13 Jun 2019 09:31:24 +0000 (17:31 +0800)]
btrfs: qgroup: Don't hold qgroup_ioctl_lock in btrfs_qgroup_inherit()

[BUG]
Lockdep will report the following circular locking dependency:

  WARNING: possible circular locking dependency detected
  5.2.0-rc2-custom #24 Tainted: G           O
  ------------------------------------------------------
  btrfs/8631 is trying to acquire lock:
  000000002536438c (&fs_info->qgroup_ioctl_lock#2){+.+.}, at: btrfs_qgroup_inherit+0x40/0x620 [btrfs]

  but task is already holding lock:
  000000003d52cc23 (&fs_info->tree_log_mutex){+.+.}, at: create_pending_snapshot+0x8b6/0xe60 [btrfs]

  which lock already depends on the new lock.

  the existing dependency chain (in reverse order) is:

  -> #2 (&fs_info->tree_log_mutex){+.+.}:
         __mutex_lock+0x76/0x940
         mutex_lock_nested+0x1b/0x20
         btrfs_commit_transaction+0x475/0xa00 [btrfs]
         btrfs_commit_super+0x71/0x80 [btrfs]
         close_ctree+0x2bd/0x320 [btrfs]
         btrfs_put_super+0x15/0x20 [btrfs]
         generic_shutdown_super+0x72/0x110
         kill_anon_super+0x18/0x30
         btrfs_kill_super+0x16/0xa0 [btrfs]
         deactivate_locked_super+0x3a/0x80
         deactivate_super+0x51/0x60
         cleanup_mnt+0x3f/0x80
         __cleanup_mnt+0x12/0x20
         task_work_run+0x94/0xb0
         exit_to_usermode_loop+0xd8/0xe0
         do_syscall_64+0x210/0x240
         entry_SYSCALL_64_after_hwframe+0x49/0xbe

  -> #1 (&fs_info->reloc_mutex){+.+.}:
         __mutex_lock+0x76/0x940
         mutex_lock_nested+0x1b/0x20
         btrfs_commit_transaction+0x40d/0xa00 [btrfs]
         btrfs_quota_enable+0x2da/0x730 [btrfs]
         btrfs_ioctl+0x2691/0x2b40 [btrfs]
         do_vfs_ioctl+0xa9/0x6d0
         ksys_ioctl+0x67/0x90
         __x64_sys_ioctl+0x1a/0x20
         do_syscall_64+0x65/0x240
         entry_SYSCALL_64_after_hwframe+0x49/0xbe

  -> #0 (&fs_info->qgroup_ioctl_lock#2){+.+.}:
         lock_acquire+0xa7/0x190
         __mutex_lock+0x76/0x940
         mutex_lock_nested+0x1b/0x20
         btrfs_qgroup_inherit+0x40/0x620 [btrfs]
         create_pending_snapshot+0x9d7/0xe60 [btrfs]
         create_pending_snapshots+0x94/0xb0 [btrfs]
         btrfs_commit_transaction+0x415/0xa00 [btrfs]
         btrfs_mksubvol+0x496/0x4e0 [btrfs]
         btrfs_ioctl_snap_create_transid+0x174/0x180 [btrfs]
         btrfs_ioctl_snap_create_v2+0x11c/0x180 [btrfs]
         btrfs_ioctl+0xa90/0x2b40 [btrfs]
         do_vfs_ioctl+0xa9/0x6d0
         ksys_ioctl+0x67/0x90
         __x64_sys_ioctl+0x1a/0x20
         do_syscall_64+0x65/0x240
         entry_SYSCALL_64_after_hwframe+0x49/0xbe

  other info that might help us debug this:

  Chain exists of:
    &fs_info->qgroup_ioctl_lock#2 --> &fs_info->reloc_mutex --> &fs_info->tree_log_mutex

   Possible unsafe locking scenario:

         CPU0                    CPU1
         ----                    ----
    lock(&fs_info->tree_log_mutex);
                                 lock(&fs_info->reloc_mutex);
                                 lock(&fs_info->tree_log_mutex);
    lock(&fs_info->qgroup_ioctl_lock#2);

   *** DEADLOCK ***

  6 locks held by btrfs/8631:
   #0: 00000000ed8f23f6 (sb_writers#12){.+.+}, at: mnt_want_write_file+0x28/0x60
   #1: 000000009fb1597a (&type->i_mutex_dir_key#10/1){+.+.}, at: btrfs_mksubvol+0x70/0x4e0 [btrfs]
   #2: 0000000088c5ad88 (&fs_info->subvol_sem){++++}, at: btrfs_mksubvol+0x128/0x4e0 [btrfs]
   #3: 000000009606fc3e (sb_internal#2){.+.+}, at: start_transaction+0x37a/0x520 [btrfs]
   #4: 00000000f82bbdf5 (&fs_info->reloc_mutex){+.+.}, at: btrfs_commit_transaction+0x40d/0xa00 [btrfs]
   #5: 000000003d52cc23 (&fs_info->tree_log_mutex){+.+.}, at: create_pending_snapshot+0x8b6/0xe60 [btrfs]

[CAUSE]
Due to the delayed subvolume creation, we need to call
btrfs_qgroup_inherit() inside commit transaction code, with a lot of
other mutex hold.
This hell of lock chain can lead to above problem.

[FIX]
On the other hand, we don't really need to hold qgroup_ioctl_lock if
we're in the context of create_pending_snapshot().
As in that context, we're the only one being able to modify qgroup.

All other qgroup functions which needs qgroup_ioctl_lock are either
holding a transaction handle, or will start a new transaction:
  Functions will start a new transaction():
  * btrfs_quota_enable()
  * btrfs_quota_disable()
  Functions hold a transaction handler:
  * btrfs_add_qgroup_relation()
  * btrfs_del_qgroup_relation()
  * btrfs_create_qgroup()
  * btrfs_remove_qgroup()
  * btrfs_limit_qgroup()
  * btrfs_qgroup_inherit() call inside create_subvol()

So we have a higher level protection provided by transaction, thus we
don't need to always hold qgroup_ioctl_lock in btrfs_qgroup_inherit().

Only the btrfs_qgroup_inherit() call in create_subvol() needs to hold
qgroup_ioctl_lock, while the btrfs_qgroup_inherit() call in
create_pending_snapshot() is already protected by transaction.

So the fix is to detect the context by checking
trans->transaction->state.
If we're at TRANS_STATE_COMMIT_DOING, then we're in commit transaction
context and no need to get the mutex.

Reported-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: correctly validate compression type
Johannes Thumshirn [Thu, 6 Jun 2019 10:07:15 +0000 (12:07 +0200)]
btrfs: correctly validate compression type

Nikolay reported the following KASAN splat when running btrfs/048:

[ 1843.470920] ==================================================================
[ 1843.471971] BUG: KASAN: slab-out-of-bounds in strncmp+0x66/0xb0
[ 1843.472775] Read of size 1 at addr ffff888111e369e2 by task btrfs/3979

[ 1843.473904] CPU: 3 PID: 3979 Comm: btrfs Not tainted 5.2.0-rc3-default #536
[ 1843.475009] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
[ 1843.476322] Call Trace:
[ 1843.476674]  dump_stack+0x7c/0xbb
[ 1843.477132]  ? strncmp+0x66/0xb0
[ 1843.477587]  print_address_description+0x114/0x320
[ 1843.478256]  ? strncmp+0x66/0xb0
[ 1843.478740]  ? strncmp+0x66/0xb0
[ 1843.479185]  __kasan_report+0x14e/0x192
[ 1843.479759]  ? strncmp+0x66/0xb0
[ 1843.480209]  kasan_report+0xe/0x20
[ 1843.480679]  strncmp+0x66/0xb0
[ 1843.481105]  prop_compression_validate+0x24/0x70
[ 1843.481798]  btrfs_xattr_handler_set_prop+0x65/0x160
[ 1843.482509]  __vfs_setxattr+0x71/0x90
[ 1843.483012]  __vfs_setxattr_noperm+0x84/0x130
[ 1843.483606]  vfs_setxattr+0xac/0xb0
[ 1843.484085]  setxattr+0x18c/0x230
[ 1843.484546]  ? vfs_setxattr+0xb0/0xb0
[ 1843.485048]  ? __mod_node_page_state+0x1f/0xa0
[ 1843.485672]  ? _raw_spin_unlock+0x24/0x40
[ 1843.486233]  ? __handle_mm_fault+0x988/0x1290
[ 1843.486823]  ? lock_acquire+0xb4/0x1e0
[ 1843.487330]  ? lock_acquire+0xb4/0x1e0
[ 1843.487842]  ? mnt_want_write_file+0x3c/0x80
[ 1843.488442]  ? debug_lockdep_rcu_enabled+0x22/0x40
[ 1843.489089]  ? rcu_sync_lockdep_assert+0xe/0x70
[ 1843.489707]  ? __sb_start_write+0x158/0x200
[ 1843.490278]  ? mnt_want_write_file+0x3c/0x80
[ 1843.490855]  ? __mnt_want_write+0x98/0xe0
[ 1843.491397]  __x64_sys_fsetxattr+0xba/0xe0
[ 1843.492201]  ? trace_hardirqs_off_thunk+0x1a/0x1c
[ 1843.493201]  do_syscall_64+0x6c/0x230
[ 1843.493988]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[ 1843.495041] RIP: 0033:0x7fa7a8a7707a
[ 1843.495819] Code: 48 8b 0d 21 de 2b 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 49 89 ca b8 be 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d ee dd 2b 00 f7 d8 64 89 01 48
[ 1843.499203] RSP: 002b:00007ffcb73bca38 EFLAGS: 00000202 ORIG_RAX: 00000000000000be
[ 1843.500210] RAX: ffffffffffffffda RBX: 00007ffcb73bda9d RCX: 00007fa7a8a7707a
[ 1843.501170] RDX: 00007ffcb73bda9d RSI: 00000000006dc050 RDI: 0000000000000003
[ 1843.502152] RBP: 00000000006dc050 R08: 0000000000000000 R09: 0000000000000000
[ 1843.503109] R10: 0000000000000002 R11: 0000000000000202 R12: 00007ffcb73bda91
[ 1843.504055] R13: 0000000000000003 R14: 00007ffcb73bda82 R15: ffffffffffffffff

[ 1843.505268] Allocated by task 3979:
[ 1843.505771]  save_stack+0x19/0x80
[ 1843.506211]  __kasan_kmalloc.constprop.5+0xa0/0xd0
[ 1843.506836]  setxattr+0xeb/0x230
[ 1843.507264]  __x64_sys_fsetxattr+0xba/0xe0
[ 1843.507886]  do_syscall_64+0x6c/0x230
[ 1843.508429]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

[ 1843.509558] Freed by task 0:
[ 1843.510188] (stack is not available)

[ 1843.511309] The buggy address belongs to the object at ffff888111e369e0
                which belongs to the cache kmalloc-8 of size 8
[ 1843.514095] The buggy address is located 2 bytes inside of
                8-byte region [ffff888111e369e0ffff888111e369e8)
[ 1843.516524] The buggy address belongs to the page:
[ 1843.517561] page:ffff88813f478d80 refcount:1 mapcount:0 mapping:ffff88811940c300 index:0xffff888111e373b8 compound_mapcount: 0
[ 1843.519993] flags: 0x4404000010200(slab|head)
[ 1843.520951] raw: 0004404000010200 ffff88813f48b008 ffff888119403d50 ffff88811940c300
[ 1843.522616] raw: ffff888111e373b8 000000000016000f 00000001ffffffff 0000000000000000
[ 1843.524281] page dumped because: kasan: bad access detected

[ 1843.525936] Memory state around the buggy address:
[ 1843.526975]  ffff888111e36880: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 1843.528479]  ffff888111e36900: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 1843.530138] >ffff888111e36980: fc fc fc fc fc fc fc fc fc fc fc fc 02 fc fc fc
[ 1843.531877]                                                        ^
[ 1843.533287]  ffff888111e36a00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 1843.534874]  ffff888111e36a80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 1843.536468] ==================================================================

This is caused by supplying a too short compression value ('lz') in the
test-case and comparing it to 'lzo' with strncmp() and a length of 3.
strncmp() read past the 'lz' when looking for the 'o' and thus caused an
out-of-bounds read.

Introduce a new check 'btrfs_compress_is_valid_type()' which not only
checks the user-supplied value against known compression types, but also
employs checks for too short values.

Reported-by: Nikolay Borisov <nborisov@suse.com>
Fixes: 272e5326c783 ("btrfs: prop: fix vanished compression property after failed set")
CC: stable@vger.kernel.org # 5.1+
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agoBtrfs: fix data loss after inode eviction, renaming it, and fsync it
Filipe Manana [Fri, 7 Jun 2019 10:25:24 +0000 (11:25 +0100)]
Btrfs: fix data loss after inode eviction, renaming it, and fsync it

When we log an inode, regardless of logging it completely or only that it
exists, we always update it as logged (logged_trans and last_log_commit
fields of the inode are updated). This is generally fine and avoids future
attempts to log it from having to do repeated work that brings no value.

However, if we write data to a file, then evict its inode after all the
dealloc was flushed (and ordered extents completed), rename the file and
fsync it, we end up not logging the new extents, since the rename may
result in logging that the inode exists in case the parent directory was
logged before. The following reproducer shows and explains how this can
happen:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt

  $ mkdir /mnt/dir
  $ touch /mnt/dir/foo
  $ touch /mnt/dir/bar

  # Do a direct IO write instead of a buffered write because with a
  # buffered write we would need to make sure dealloc gets flushed and
  # complete before we do the inode eviction later, and we can not do that
  # from user space with call to things such as sync(2) since that results
  # in a transaction commit as well.
  $ xfs_io -d -c "pwrite -S 0xd3 0 4K" /mnt/dir/bar

  # Keep the directory dir in use while we evict inodes. We want our file
  # bar's inode to be evicted but we don't want our directory's inode to
  # be evicted (if it were evicted too, we would not be able to reproduce
  # the issue since the first fsync below, of file foo, would result in a
  # transaction commit.
  $ ( cd /mnt/dir; while true; do :; done ) &
  $ pid=$!

  # Wait a bit to give time for the background process to chdir.
  $ sleep 0.1

  # Evict all inodes, except the inode for the directory dir because it is
  # currently in use by our background process.
  $ echo 2 > /proc/sys/vm/drop_caches

  # fsync file foo, which ends up persisting information about the parent
  # directory because it is a new inode.
  $ xfs_io -c fsync /mnt/dir/foo

  # Rename bar, this results in logging that this inode exists (inode item,
  # names, xattrs) because the parent directory is in the log.
  $ mv /mnt/dir/bar /mnt/dir/baz

  # Now fsync baz, which ends up doing absolutely nothing because of the
  # rename operation which logged that the inode exists only.
  $ xfs_io -c fsync /mnt/dir/baz

  <power failure>

  $ mount /dev/sdb /mnt
  $ od -t x1 -A d /mnt/dir/baz
  0000000

    --> Empty file, data we wrote is missing.

Fix this by not updating last_sub_trans of an inode when we are logging
only that it exists and the inode was not yet logged since it was loaded
from disk (full_sync bit set), this is enough to make btrfs_inode_in_log()
return false for this scenario and make us log the inode. The logged_trans
of the inode is still always setsince that alone is used to track if names
need to be deleted as part of unlink operations.

Fixes: 257c62e1bce03e ("Btrfs: avoid tree log commit when there are no changes")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: raid56: clear incompat block group flags after removing the last one
David Sterba [Tue, 11 Jun 2019 12:31:01 +0000 (14:31 +0200)]
btrfs: raid56: clear incompat block group flags after removing the last one

The incompat bit for RAID56 is set either at mount time or automatically
when the profile is used by balance. The part where the bit is removed
is missing and can be unexpected or undesired when an older kernel is
needed.

This patch will drop the incompat bit after this command, assuming
that RAID5 profile is not used by system or metadata:

 $ btrfs balance start -dconvert=raid5 /mnt
 $ btrfs balance start -dconvert=raid1 /mnt

This will print "clearing 128 feature flag" to the system log.

The patch is safe for backporting to older kernels.

Reported-by: Hugo Mills <hugo@carfax.org.uk>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: switch extent_buffer write_locks from atomic to int
David Sterba [Thu, 2 May 2019 14:53:47 +0000 (16:53 +0200)]
btrfs: switch extent_buffer write_locks from atomic to int

The write_locks is either 0 or 1 and always updated under the lock,
so we don't need the atomic_t semantics.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: switch extent_buffer spinning_writers from atomic to int
David Sterba [Thu, 2 May 2019 14:51:53 +0000 (16:51 +0200)]
btrfs: switch extent_buffer spinning_writers from atomic to int

The spinning_writers is either 0 or 1 and always updated under the lock,
so we don't need the atomic_t semantics.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: switch extent_buffer blocking_writers from atomic to int
David Sterba [Thu, 2 May 2019 14:47:23 +0000 (16:47 +0200)]
btrfs: switch extent_buffer blocking_writers from atomic to int

The blocking_writers is either 0 or 1 and always updated under the lock,
so we don't need the atomic_t semantics.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: assert delayed ref lock in btrfs_find_delayed_ref_head
David Sterba [Wed, 27 Mar 2019 15:19:55 +0000 (16:19 +0100)]
btrfs: assert delayed ref lock in btrfs_find_delayed_ref_head

Turn the comment about required lock into an assertion.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: tests: add locks around add_extent_mapping
David Sterba [Wed, 3 Apr 2019 17:32:56 +0000 (19:32 +0200)]
btrfs: tests: add locks around add_extent_mapping

There are no concerns about locking during the selftests so the locks
are not necessary, but following patches will add lockdep assertions to
add_extent_mapping so this is needed in tests too.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Document __etree_search
Nikolay Borisov [Wed, 5 Jun 2019 11:50:04 +0000 (14:50 +0300)]
btrfs: Document __etree_search

The function has a lot of return values and specific conventions making
it cumbersome to understand what's returned. Have a go at documenting
its parameters and return values.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Don't trim returned range based on input value in find_first_clear_extent_bit
Nikolay Borisov [Mon, 3 Jun 2019 10:06:02 +0000 (13:06 +0300)]
btrfs: Don't trim returned range based on input value in find_first_clear_extent_bit

Currently find_first_clear_extent_bit always returns a range whose
starting value is >= passed 'start'. This implicit trimming behavior is
somewhat subtle and an implementation detail.

Instead, this patch modifies the function such that now it always
returns the range which contains passed 'start' and has the given bits
unset. This range could either be due to presence of existing records
which contains 'start' but have the bits unset or because there are no
records that contain the given starting offset.

This patch also adds test cases which cover find_first_clear_extent_bit
since they were missing up until now.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: trim: make reserved device area adjustments more explicit
Nikolay Borisov [Mon, 3 Jun 2019 10:06:01 +0000 (13:06 +0300)]
btrfs: trim: make reserved device area adjustments more explicit

Currently the first megabyte on a device housing a btrfs filesystem is
exempt from allocation and trimming. Currently this is not a problem
since 'start' is set to 1M at the beginning of btrfs_trim_free_extents
and find_first_clear_extent_bit always returns a range that is >= start.

However, in a follow up patch find_first_clear_extent_bit will be
changed such that it will return a range containing 'start' and this
range may very well be 0...>=1M so 'start'.

Future proof the sole user of find_first_clear_extent_bit by setting
'start' after the function is called. No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use file:line format for assertion report
David Sterba [Thu, 30 May 2019 16:43:53 +0000 (18:43 +0200)]
btrfs: use file:line format for assertion report

The filename:line format is commonly understood by editors and can be
copy&pasted more easily than the current format.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: remove assumption about csum type form btrfs_print_data_csum_error()
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:58 +0000 (16:58 +0200)]
btrfs: remove assumption about csum type form btrfs_print_data_csum_error()

btrfs_print_data_csum_error() still assumed checksums to be 32 bit in
size.  Make it size agnostic.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: directly call into crypto framework for checksumming
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:57 +0000 (16:58 +0200)]
btrfs: directly call into crypto framework for checksumming

Currently btrfs_csum_data() relied on the crc32c() wrapper around the
crypto framework for calculating the CRCs.

As we have our own crypto_shash structure in the fs_info now, we can
directly call into the crypto framework without going trough the wrapper.

This way we can even remove the btrfs_csum_data() and btrfs_csum_final()
wrappers.

The module dependency on crc32c is preserved via MODULE_SOFTDEP("pre:
crc32c"), which was previously provided by LIBCRC32C config option doing
the same.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: add boilerplate code for directly including the crypto framework
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:56 +0000 (16:58 +0200)]
btrfs: add boilerplate code for directly including the crypto framework

Add boilerplate code for directly including the crypto framework.  This
helps us flipping the switch for new algorithms.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Simplify btrfs_check_super_csum() and get rid of size assumptions
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:55 +0000 (16:58 +0200)]
btrfs: Simplify btrfs_check_super_csum() and get rid of size assumptions

Now that we have already checked for a valid checksum type before
calling btrfs_check_super_csum(), it can be simplified even further.

While at it get rid of the implicit size assumption of the resulting
checksum as well.

This is a preparation for changing all checksum functionality to use the
crypto layer later.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: check for supported superblock checksum type before checksum validation
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:54 +0000 (16:58 +0200)]
btrfs: check for supported superblock checksum type before checksum validation

Now that we have factorerd out the superblock checksum type validation,
we can check for supported superblock checksum types before doing the
actual validation of the superblock read from disk.

This leads the path to further simplifications of
btrfs_check_super_csum() later on.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: add common checksum type validation
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:53 +0000 (16:58 +0200)]
btrfs: add common checksum type validation

Currently btrfs is only supporting CRC32C as checksumming algorithm. As
this is about to change provide a function to validate the checksum type
in the superblock against all possible algorithms.

This makes adding new algorithms easier as there are fewer places to
adjust when adding new algorithms.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: format checksums according to type for printing
Johannes Thumshirn [Mon, 3 Jun 2019 14:58:52 +0000 (16:58 +0200)]
btrfs: format checksums according to type for printing

Add a small helper for btrfs_print_data_csum_error() which formats the
checksum according to it's type for pretty printing.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ shorten macro name ]
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: don't assume compressed_bio sums to be 4 bytes
Johannes Thumshirn [Wed, 22 May 2019 08:19:02 +0000 (10:19 +0200)]
btrfs: don't assume compressed_bio sums to be 4 bytes

BTRFS has the implicit assumption that a checksum in compressed_bio is 4
bytes. While this is true for CRC32C, it is not for any other checksum.

Change the data type to be a byte array and adjust loop index calculation
accordingly.

Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: don't assume ordered sums to be 4 bytes
Johannes Thumshirn [Wed, 22 May 2019 08:19:01 +0000 (10:19 +0200)]
btrfs: don't assume ordered sums to be 4 bytes

BTRFS has the implicit assumption that a checksum in btrfs_orderd_sums
is 4 bytes. While this is true for CRC32C, it is not for any other
checksum.

Change the data type to be a byte array and adjust loop index
calculation accordingly.

This includes moving the adjustment of 'index' by 'ins_size' in
btrfs_csum_file_blocks() before dividing 'ins_size' by the checksum
size, because before this patch the 'sums' member of 'struct
btrfs_ordered_sum' was 4 Bytes in size and afterwards it is only one
byte.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use btrfs_crc32c{,_final}() in for free space cache
Johannes Thumshirn [Wed, 22 May 2019 08:19:00 +0000 (10:19 +0200)]
btrfs: use btrfs_crc32c{,_final}() in for free space cache

The CRC checksum in the free space cache is not dependant on the super
block's csum_type field but always a CRC32C.

So use btrfs_crc32c() and btrfs_crc32c_final() instead of
btrfs_csum_data() and btrfs_csum_final() for computing these checksums.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: resurrect btrfs_crc32c()
Johannes Thumshirn [Wed, 22 May 2019 08:18:59 +0000 (10:18 +0200)]
btrfs: resurrect btrfs_crc32c()

Commit 9678c54388b6 ("btrfs: Remove custom crc32c init code") removed
the btrfs_crc32c() function, because it was a duplicate of the crc32c()
library function we already have in the kernel.

Resurrect it as a shim wrapper over crc32c() to make following
transformations of the checksumming code in btrfs easier.

Also provide a btrfs_crc32_final() to ease following transformations.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use btrfs_csum_data() instead of directly calling crc32c
Johannes Thumshirn [Wed, 22 May 2019 08:18:58 +0000 (10:18 +0200)]
btrfs: use btrfs_csum_data() instead of directly calling crc32c

btrfsic_test_for_metadata() directly calls the crc32c() library function
for calculating the CRC32C checksum, but then uses btrfs_csum_final() to
invert the result.

To ease further refactoring and development around checksumming in BTRFS
convert to calling btrfs_csum_data(), which is a wrapper around
crc32c().

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Flush before reflinking any extent to prevent NOCOW write falling back to...
Qu Wenruo [Wed, 8 May 2019 10:49:58 +0000 (18:49 +0800)]
btrfs: Flush before reflinking any extent to prevent NOCOW write falling back to COW without data reservation

[BUG]
The following script can cause unexpected fsync failure:

  #!/bin/bash

  dev=/dev/test/test
  mnt=/mnt/btrfs

  mkfs.btrfs -f $dev -b 512M > /dev/null
  mount $dev $mnt -o nospace_cache

  # Prealloc one extent
  xfs_io -f -c "falloc 8k 64m" $mnt/file1
  # Fill the remaining data space
  xfs_io -f -c "pwrite 0 -b 4k 512M" $mnt/padding
  sync

  # Write into the prealloc extent
  xfs_io -c "pwrite 1m 16m" $mnt/file1

  # Reflink then fsync, fsync would fail due to ENOSPC
  xfs_io -c "reflink $mnt/file1 8k 0 4k" -c "fsync" $mnt/file1
  umount $dev

The fsync fails with ENOSPC, and the last page of the buffered write is
lost.

[CAUSE]
This is caused by:
- Btrfs' back reference only has extent level granularity
  So write into shared extent must be COWed even only part of the extent
  is shared.

So for above script we have:
- fallocate
  Create a preallocated extent where we can do NOCOW write.

- fill all the remaining data and unallocated space

- buffered write into preallocated space
  As we have not enough space available for data and the extent is not
  shared (yet) we fall into NOCOW mode.

- reflink
  Now part of the large preallocated extent is shared, later write
  into that extent must be COWed.

- fsync triggers writeback
  But now the extent is shared and therefore we must fallback into COW
  mode, which fails with ENOSPC since there's not enough space to
  allocate data extents.

[WORKAROUND]
The workaround is to ensure any buffered write in the related extents
(not just the reflink source range) get flushed before reflink/dedupe,
so that NOCOW writes succeed that happened before reflinking succeed.

The workaround is expensive, we could do it better by only flushing
NOCOW range, but that needs extra accounting for NOCOW range.
For now, fix the possible data loss first.

Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Return EAGAIN if we can't start no snpashot write in check_can_nocow
Nikolay Borisov [Tue, 7 May 2019 07:23:46 +0000 (10:23 +0300)]
btrfs: Return EAGAIN if we can't start no snpashot write in check_can_nocow

The first thing code does in check_can_nocow is trying to block
concurrent snapshots. If this fails (due to snpashot already being in
progress) the function returns ENOSPC which makes no sense. Instead
return EAGAIN. Despite this return value not being propagated to callers
it's good practice to return the closest in terms of semantics error
code. 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>
18 months agobtrfs: Add comments on locking of several device-related fields
Nikolay Borisov [Thu, 9 May 2019 15:11:11 +0000 (18:11 +0300)]
btrfs: Add comments on locking of several device-related fields

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Always use a cached extent_state in btrfs_lock_and_flush_ordered_range
Nikolay Borisov [Tue, 7 May 2019 07:19:24 +0000 (10:19 +0300)]
btrfs: Always use a cached extent_state in btrfs_lock_and_flush_ordered_range

In case no cached_state argument is passed to
btrfs_lock_and_flush_ordered_range use one locally in the function. This
optimises the case when an ordered extent is found since the unlock
function will be able to unlock that state directly without searching
for it again.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Use newly introduced btrfs_lock_and_flush_ordered_range
Nikolay Borisov [Tue, 7 May 2019 07:19:23 +0000 (10:19 +0300)]
btrfs: Use newly introduced btrfs_lock_and_flush_ordered_range

There several functions which open code
btrfs_lock_and_flush_ordered_range, just replace them with a call to the
function. No functional changes.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: add new helper btrfs_lock_and_flush_ordered_range
Nikolay Borisov [Tue, 7 May 2019 07:19:22 +0000 (10:19 +0300)]
btrfs: add new helper btrfs_lock_and_flush_ordered_range

There is a certain idiom used in multiple places in btrfs' codebase,
dealing with flushing an ordered range. Factor this in a separate
function that can be reused. Future patches will replace the existing
code with that function.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: remove the incorrect comment on RO fs when btrfs_run_delalloc_range() fails
Qu Wenruo [Thu, 9 May 2019 07:31:50 +0000 (15:31 +0800)]
btrfs: remove the incorrect comment on RO fs when btrfs_run_delalloc_range() fails

At the context of btrfs_run_delalloc_range(), we haven't started/joined
a transaction, thus even something went wrong, we can't and won't abort
transaction, thus no way to make the fs RO.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: extent-tree: Add trace events for space info numbers update
Qu Wenruo [Mon, 29 Apr 2019 06:03:33 +0000 (14:03 +0800)]
btrfs: extent-tree: Add trace events for space info numbers update

Add trace event for update_bytes_pinned() and update_bytes_may_use() to
detect underflow better.

The output would be something like (only showing data part):

  ## Buffered write start, 16K total ##
  2255.954 xfs_io/860 btrfs:update_bytes_may_use:(nil)U: type=DATA old=0 diff=4096
  2257.169 sudo/860 btrfs:update_bytes_may_use:(nil)U: type=DATA old=4096 diff=4096
  2257.346 sudo/860 btrfs:update_bytes_may_use:(nil)U: type=DATA old=8192 diff=4096
  2257.542 sudo/860 btrfs:update_bytes_may_use:(nil)U: type=DATA old=12288 diff=4096

  ## Delalloc start ##
  3727.853 kworker/u8:3-e/700 btrfs:update_bytes_may_use:(nil)U: type=DATA old=16384 diff=-16384

  ## Space cache update ##
  3733.132 sudo/862 btrfs:update_bytes_may_use:(nil)U: type=DATA old=0 diff=65536
  3733.169 sudo/862 btrfs:update_bytes_may_use:(nil)U: type=DATA old=65536 diff=-65536
  3739.868 sudo/862 btrfs:update_bytes_may_use:(nil)U: type=DATA old=0 diff=65536
  3739.891 sudo/862 btrfs:update_bytes_may_use:(nil)U: type=DATA old=65536 diff=-65536

These two trace events will allow bcc tool to probe btrfs_space_info
changes and detect underflow with more details (e.g. backtrace for each
update).

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: extent-tree: Add lockdep assert when updating space info
Qu Wenruo [Mon, 29 Apr 2019 06:03:32 +0000 (14:03 +0800)]
btrfs: extent-tree: Add lockdep assert when updating space info

Just add a safe net for btrfs_space_info member updating.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: read number of data stripes from map only once
David Sterba [Fri, 17 May 2019 09:43:45 +0000 (11:43 +0200)]
btrfs: read number of data stripes from map only once

There are several places that call nr_data_stripes, but this value does
not change.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: constify map parameter for nr_parity_stripes and nr_data_stripes
David Sterba [Fri, 17 May 2019 09:43:43 +0000 (11:43 +0200)]
btrfs: constify map parameter for nr_parity_stripes and nr_data_stripes

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: refactor helper for bg flags to name conversion
David Sterba [Fri, 17 May 2019 09:43:41 +0000 (11:43 +0200)]
btrfs: refactor helper for bg flags to name conversion

The helper lacks the btrfs_ prefix and the parameter is the raw
blockgroup type, so none of the callers has to do the flags -> index
conversion.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: factor out devs_max setting in __btrfs_alloc_chunk
David Sterba [Fri, 17 May 2019 09:43:38 +0000 (11:43 +0200)]
btrfs: factor out devs_max setting in __btrfs_alloc_chunk

Merge the repeated code before the if-else block.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use u8 for raid_array members
David Sterba [Fri, 17 May 2019 09:43:36 +0000 (11:43 +0200)]
btrfs: use u8 for raid_array members

The raid_attr table is now 7 * 56 = 392 bytes long, consisting of just
small numbers so we don't have to use ints. New size is 7 * 32 = 224,
saving 3 cachelines.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: factor out helper for counting data stripes
David Sterba [Fri, 17 May 2019 09:43:34 +0000 (11:43 +0200)]
btrfs: factor out helper for counting data stripes

Factor the sequence of ifs to a helper, the 'data stripes' here means
the number of stripes without redundancy and parity.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use raid_attr table for btrfs_bg_type_to_factor
David Sterba [Fri, 17 May 2019 09:43:31 +0000 (11:43 +0200)]
btrfs: use raid_attr table for btrfs_bg_type_to_factor

The factor is the number of copies.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use raid_attr table to find profiles for integrity lowering
David Sterba [Fri, 17 May 2019 09:43:29 +0000 (11:43 +0200)]
btrfs: use raid_attr table to find profiles for integrity lowering

Replace open coded list of the profiles by selecting them from the
raid_attr table. The criteria are now more explicit, we need profiles
that have more than 1 copy of the data or can reconstruct the data with
a missing device.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use raid_attr to get allowed profiles for balance conversion
David Sterba [Fri, 17 May 2019 09:43:27 +0000 (11:43 +0200)]
btrfs: use raid_attr to get allowed profiles for balance conversion

Iterate over the table and gather all allowed profiles for a given
number of devices, instead of open coding.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use raid_attr in btrfs_chunk_max_errors
David Sterba [Fri, 17 May 2019 09:43:22 +0000 (11:43 +0200)]
btrfs: use raid_attr in btrfs_chunk_max_errors

The number of tolerated failures is stored in the raid_attr table, use
it.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: use raid_attr table in get_profile_num_devs
David Sterba [Fri, 17 May 2019 09:43:20 +0000 (11:43 +0200)]
btrfs: use raid_attr table in get_profile_num_devs

The dev_max constraints are defined in the raid_attr table, use it
instead of open-coding it.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: remove mapping tree structures indirection
David Sterba [Fri, 17 May 2019 09:43:17 +0000 (11:43 +0200)]
btrfs: remove mapping tree structures indirection

fs_info::mapping_tree is the physical<->logical mapping tree and uses
the same underlying structure as extents, but is embedded to another
structure. There are no other members and this indirection is useless.
No functional change.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: raid56: allow the exact minimum number of devices for balance convert
David Sterba [Fri, 17 May 2019 09:43:15 +0000 (11:43 +0200)]
btrfs: raid56: allow the exact minimum number of devices for balance convert

The minimum number of devices for RAID5 is 2, though this is only a bit
expensive RAID1, and for RAID6 it's 3, which is a triple copy that works
only 3 devices.

mkfs.btrfs allows that and mounting such filesystem also works, so the
conversion via balance filters is inconsistent with the others and we
should not prevent it.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: fix minimum number of chunk errors for DUP
David Sterba [Fri, 17 May 2019 09:43:13 +0000 (11:43 +0200)]
btrfs: fix minimum number of chunk errors for DUP

The list of profiles in btrfs_chunk_max_errors lists DUP as a profile
DUP able to tolerate 1 device missing. Though this profile is special
with 2 copies, it still needs the device, unlike the others.

Looking at the history of changes, thre's no clear reason why DUP is
there, functions were refactored and blocks of code merged to one
helper.

d20983b40e828 Btrfs: fix writing data into the seed filesystem
  - factor code to a helper

de11cc12df173 Btrfs: don't pre-allocate btrfs bio
  - unrelated change, DUP still in the list with max errors 1

a236aed14ccb0 Btrfs: Deal with failed writes in mirrored configurations
  - introduced the max errors, leaves DUP and RAID1 in the same group

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agoBtrfs: remove unused variables in __btrfs_unlink_inode
Liu Bo [Tue, 11 Sep 2018 22:06:25 +0000 (06:06 +0800)]
Btrfs: remove unused variables in __btrfs_unlink_inode

This code was first introduced in 5f39d397dfbe ("Btrfs: Create
extent_buffer interface for large blocksizes") and the function was
named btrfs_unlink_trans. It later got renamed to __btrfs_unlink_inode
and finally commit 16cdcec736cd ("btrfs: implement delayed inode items
operation") changed the way inodes are deleted and obviated the need for
those two members.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ replace changelog by Nikolay's version ]
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Remove unused variable mode in btrfs_mount
Goldwyn Rodrigues [Fri, 5 Oct 2018 12:26:15 +0000 (07:26 -0500)]
btrfs: Remove unused variable mode in btrfs_mount

This is a leftover from 312c89fbca06 ("btrfs: cleanup btrfs_mount()
using btrfs_mount_root()"), the mode was used for opening devices that's
not done here anymore.

Reviewed-by: Nikolay Borisov <nborisov@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>
18 months agobtrfs: switch order of unlocks of space_info and bg in do_trimming()
Su Yue [Wed, 28 Nov 2018 03:21:12 +0000 (11:21 +0800)]
btrfs: switch order of unlocks of space_info and bg in do_trimming()

In function do_trimming(), block_group->lock should be unlocked first,
as the locks should be released in the reverse order. This does not
cause problems but should follow the best practices.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: tree-checker: Check if the file extent end overflows
Qu Wenruo [Fri, 3 May 2019 00:30:54 +0000 (08:30 +0800)]
btrfs: tree-checker: Check if the file extent end overflows

Under certain conditions, we could have strange file extent item in log
tree like:

  item 18 key (69599 108 397312) itemoff 15208 itemsize 53
extent data disk bytenr 0 nr 0
extent data offset 0 nr 18446744073709547520 ram 18446744073709547520

The num_bytes + ram_bytes overflow 64 bit type.

For num_bytes part, we can detect such overflow along with file offset
(key->offset), as file_offset + num_bytes should never go beyond u64.

For ram_bytes part, it's about the decompressed size of the extent, not
directly related to the size.
In theory it is OK to have a large value, and put extra limitation
on RAM bytes may cause unexpected false alerts.

So in tree-checker, we only check if the file offset and num bytes
overflow.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Remove redundant assignment of tgt_device->commit_total_bytes
Nikolay Borisov [Tue, 14 May 2019 10:54:45 +0000 (13:54 +0300)]
btrfs: Remove redundant assignment of tgt_device->commit_total_bytes

This is already done in btrfs_init_dev_replace_tgtdev which is the first
phase of device replace, called before doing scrub. During that time
exclusive lock is held. Additionally btrfs_fs_device::commit_total_bytes
is always set based on the size of the underlying block device which
shouldn't change once set. This makes the 2nd assignment of the variable
in the finishing phase redundant.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Explicitly reserve space for devreplace item
Nikolay Borisov [Tue, 14 May 2019 10:54:43 +0000 (13:54 +0300)]
btrfs: Explicitly reserve space for devreplace item

Part of device replace involves writing an item to the device root
containing information about pending replace operations. Currently space
for this item is not being explicitly reserved so this works thanks to
presence of global reserve. While not fatal it's not a good practice.
Let's be explicit about space requirement of device replace and reserve
space when starting the transaction.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Streamline replace sem unlock in btrfs_dev_replace_start
Nikolay Borisov [Tue, 14 May 2019 10:54:42 +0000 (13:54 +0300)]
btrfs: Streamline replace sem unlock in btrfs_dev_replace_start

There are only 2 branches which goto leave label with need_unlock set
to true. Essentially need_unlock is used as a substitute for directly
calling up_write. Since the branches needing this are only 2 and their
context is not that big it's more clear to just call up_write where
required. No functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Ensure btrfs_init_dev_replace_tgtdev sees up to date values
Nikolay Borisov [Tue, 14 May 2019 10:54:41 +0000 (13:54 +0300)]
btrfs: Ensure btrfs_init_dev_replace_tgtdev sees up to date values

btrfs_init_dev_replace_tgtdev reads certain values from the source
device (such as commit_total_bytes) which are updated during transaction
commit. Currently this function is called before committing any pending
transaction, leading to possibly reading outdated values.

Fix this by moving the function below the transaction commit, at this
point the EXCL_OP bit it set hence once transaction is complete the
total size of the device cannot be changed (it's usually changed by
resize/remove ops which are blocked).

Fixes: 9e271ae27e44 ("Btrfs: kernel operation should come after user input has been verified")
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: dev-replace: Remove impossible WARN_ON
Nikolay Borisov [Tue, 14 May 2019 10:54:40 +0000 (13:54 +0300)]
btrfs: dev-replace: Remove impossible WARN_ON

This WARN_ON can never trigger because src_device cannot be null.
btrfs_find_device_by_devspec always returns either an error or a valid
pointer to the device. Just remove it.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Reduce critical section in btrfs_init_dev_replace_tgtdev
Nikolay Borisov [Tue, 14 May 2019 10:54:39 +0000 (13:54 +0300)]
btrfs: Reduce critical section in btrfs_init_dev_replace_tgtdev

There is no point in holding btrfs_fs_devices::device_list_mutex
while initialising fields of the not-yet-published device. Instead,
hold the mutex only when the newly initialised device is being
published. I think holding device_list_mutex here is redundant
altogether, because at this point BTRFS_FS_EXCL_OP is set which
prevents device removal/addition/balance/resize to occur.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: Don't opencode sync_blockdev in btrfs_init_dev_replace_tgtdev
Nikolay Borisov [Tue, 14 May 2019 10:54:38 +0000 (13:54 +0300)]
btrfs: Don't opencode sync_blockdev in btrfs_init_dev_replace_tgtdev

Using sync_blockdev makes it plain obvious what's happening. No
functional changes.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: fiemap: preallocate ulists for btrfs_check_shared
David Sterba [Wed, 15 May 2019 13:31:04 +0000 (15:31 +0200)]
btrfs: fiemap: preallocate ulists for btrfs_check_shared

btrfs_check_shared looks up parents of a given extent and uses ulists
for that. These are allocated and freed repeatedly. Preallocation in the
caller will avoid the overhead and also allow us to use the GFP_KERNEL
as it is happens before the extent locks are taken.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: detect fast implementation of crc32c on all architectures
David Sterba [Thu, 16 May 2019 11:39:59 +0000 (13:39 +0200)]
btrfs: detect fast implementation of crc32c on all architectures

Currently, there's only check for fast crc32c implementation on X86,
based on the CPU flags. This is used to decide if checksumming should be
offloaded to worker threads or can be calculated by the caller.

As there are more architectures that implement a faster version of
crc32c (ARM, SPARC, s390, MIPS, PowerPC), also there are specialized hw
cards.

The detection is based on driver name, all generic C implementations
contain 'generic', while the specialized versions do not. Alternatively
the priority could be used, but this is not currently provided by the
crypto API.

The flag is set per-filesystem at mount time and used for the offloading
decisions.

Signed-off-by: David Sterba <dsterba@suse.com>
18 months agobtrfs: extent-tree: Refactor add_pinned_bytes() to add|sub_pinned_bytes()
Qu Wenruo [Tue, 14 May 2019 23:33:48 +0000 (07:33 +0800)]
btrfs: extent-tree: Refactor add_pinned_bytes() to add|sub_pinned_bytes()

Instead of using @sign to determine whether we're adding or subtracting.
Even it only has 3 callers, it's still (and in fact already caused
problem in the past) confusing to use.

Refactor add_pinned_bytes() to add_pinned_bytes() and sub_pinned_bytes()
to explicitly show what we're doing.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
18 months agoLinux 5.2-rc7 v5.2-rc7
Linus Torvalds [Sun, 30 Jun 2019 03:25:36 +0000 (11:25 +0800)]
Linux 5.2-rc7

18 months agoMerge tag 'powerpc-5.2-7' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Sun, 30 Jun 2019 03:20:52 +0000 (11:20 +0800)]
Merge tag 'powerpc-5.2-7' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fix from Michael Ellerman:
 "One fix for a regression in my commit adding KUAP (Kernel User Access
  Prevention) on Radix, which incorrectly touched the AMR in the early
  machine check handler.

  Thanks to Nicholas Piggin"

* tag 'powerpc-5.2-7' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/64s/exception: Fix machine check early corrupting AMR

18 months agoMerge branch 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 30 Jun 2019 03:19:17 +0000 (11:19 +0800)]
Merge branch 'smp-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull SMP fixes from Thomas Gleixner:
 "Two small changes for the cpu hotplug code:

   - Prevent out of bounds access which actually might crash the machine
     caused by a missing bounds check in the fail injection code

   - Warn about unsupported migitation mode command line arguments to
     make people aware that they typoed the paramater. Not necessarily a
     fix but quite some people tripped over that"

* 'smp-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  cpu/hotplug: Fix out-of-bounds read when setting fail state
  cpu/speculation: Warn on unsupported mitigations= parameter

18 months agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Jun 2019 11:42:30 +0000 (19:42 +0800)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Ingo Molnar:
 "Misc fixes all over the place:

   - might_sleep() atomicity fix in the microcode loader

   - resctrl boundary condition fix

   - APIC arithmethics bug fix for frequencies >= 4.2 GHz

   - three 5-level paging crash fixes

   - two speculation fixes

   - a perf/stacktrace fix"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/unwind/orc: Fall back to using frame pointers for generated code
  perf/x86: Always store regs->ip in perf_callchain_kernel()
  x86/speculation: Allow guests to use SSBD even if host does not
  x86/mm: Handle physical-virtual alignment mismatch in phys_p4d_init()
  x86/boot/64: Add missing fixup_pointer() for next_early_pgt access
  x86/boot/64: Fix crash if kernel image crosses page table boundary
  x86/apic: Fix integer overflow on 10 bit left shift of cpu_khz
  x86/resctrl: Prevent possible overrun during bitmap operations
  x86/microcode: Fix the microcode load on CPU hotplug for real

18 months agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Jun 2019 11:39:17 +0000 (19:39 +0800)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "Various fixes, most of them related to bugs perf fuzzing found in the
  x86 code"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/x86/regs: Use PERF_REG_EXTENDED_MASK
  perf/x86: Remove pmu->pebs_no_xmm_regs
  perf/x86: Clean up PEBS_XMM_REGS
  perf/x86/regs: Check reserved bits
  perf/x86: Disable extended registers for non-supported PMUs
  perf/ioctl: Add check for the sample_period value
  perf/core: Fix perf_sample_regs_user() mm check

18 months agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Jun 2019 11:36:53 +0000 (19:36 +0800)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Ingo Molnar:
 "Diverse irqchip driver fixes"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3-its: Fix command queue pointer comparison bug
  irqchip/mips-gic: Use the correct local interrupt map registers
  irqchip/ti-sci-inta: Fix kernel crash if irq_create_fwspec_mapping fail
  irqchip/irq-csky-mpintc: Support auto irq deliver to all cpus

18 months agoMerge branch 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Jun 2019 11:32:09 +0000 (19:32 +0800)]
Merge branch 'efi-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull EFI fixes from Ingo Molnar:
 "Four fixes:
   - fix a kexec crash on arm64
   - fix a reboot crash on some Android platforms
   - future-proof the code for upcoming ACPI 6.2 changes
   - fix a build warning on x86"

* 'efi-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  efibc: Replace variable set function in notifier call
  x86/efi: fix a -Wtype-limits compilation warning
  efi/bgrt: Drop BGRT status field reserved bits check
  efi/memreserve: deal with memreserve entries in unmapped memory

18 months agoMerge tag 'pm-5.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Sat, 29 Jun 2019 11:29:45 +0000 (19:29 +0800)]
Merge tag 'pm-5.2-rc7' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fix from Rafael Wysocki:
 "Avoid skipping bus-level PCI power management during system resume for
  PCIe ports left in D0 during the preceding suspend transition on
  platforms where the power states of those ports can change out of the
  PCI layer's control"

* tag 'pm-5.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PCI: PM: Avoid skipping bus-level PM on platforms without ACPI

18 months agoMerge tag 'xarray-5.2-rc6' of git://git.infradead.org/users/willy/linux-dax
Linus Torvalds [Sat, 29 Jun 2019 09:14:57 +0000 (17:14 +0800)]
Merge tag 'xarray-5.2-rc6' of git://git.infradead.org/users/willy/linux-dax

Pull XArray fixes from Matthew Wilcox:

 - Account XArray nodes for the page cache to the appropriate cgroup
   (Johannes Weiner)

 - Fix idr_get_next() when called under the RCU lock (Matthew Wilcox)

 - Add a test for xa_insert() (Matthew Wilcox)

* tag 'xarray-5.2-rc6' of git://git.infradead.org/users/willy/linux-dax:
  XArray tests: Add check_insert
  idr: Fix idr_get_next race with idr_remove
  mm: fix page cache convergence regression

18 months agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 29 Jun 2019 09:11:01 +0000 (17:11 +0800)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "15 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  linux/kernel.h: fix overflow for DIV_ROUND_UP_ULL
  mm, swap: fix THP swap out
  fork,memcg: alloc_thread_stack_node needs to set tsk->stack
  MAINTAINERS: add CLANG/LLVM BUILD SUPPORT info
  mm/vmalloc.c: avoid bogus -Wmaybe-uninitialized warning
  mm/page_idle.c: fix oops because end_pfn is larger than max_pfn
  initramfs: fix populate_initrd_image() section mismatch
  mm/oom_kill.c: fix uninitialized oc->constraint
  mm: hugetlb: soft-offline: dissolve_free_huge_page() return zero on !PageHuge
  mm: soft-offline: return -EBUSY if set_hwpoison_free_buddy_page() fails
  signal: remove the wrong signal_pending() check in restore_user_sigmask()
  fs/binfmt_flat.c: make load_flat_shared_library() work
  mm/mempolicy.c: fix an incorrect rebind node in mpol_rebind_nodemask
  fs/proc/array.c: allow reporting eip/esp for all coredumping threads
  mm/dev_pfn: exclude MEMORY_DEVICE_PRIVATE while computing virtual address

18 months agoMerge tag 'arc-5.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
Linus Torvalds [Sat, 29 Jun 2019 09:05:58 +0000 (17:05 +0800)]
Merge tag 'arc-5.2-rc7' of git://git./linux/kernel/git/vgupta/arc

Pull ARC fixes from Vineet Gupta:

 - hsdk platform unifying apertures

 - build system CROSS_COMPILE prefix

* tag 'arc-5.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc:
  ARC: [plat-hsdk]: unify memory apertures configuration
  ARC: build: Try to guess CROSS_COMPILE with cc-cross-prefix

18 months agoMerge tag 'riscv-for-v5.2/fixes-rc7' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 29 Jun 2019 09:04:21 +0000 (17:04 +0800)]
Merge tag 'riscv-for-v5.2/fixes-rc7' of git://git./linux/kernel/git/riscv/linux

Pull RISC-V fixes from Paul Walmsley:
 "Minor RISC-V fixes and one defconfig update.

  The fixes have no functional impact:

   - Fix some comment text in the memory management vmalloc_fault path.

   - Fix some warnings from the DT compiler in our newly-added DT files.

   - Change the newly-added DT bindings such that SoC IP blocks with
     external I/O are marked as "disabled" by default, then enable them
     explicitly in board DT files when the devices are used on the
     board. This aligns the bindings with existing upstream practice.

   - Add the MIT license as an option for a minor header file, at the
     request of one of the U-Boot maintainers.

  The RISC-V defconfig update builds the SiFive SPI driver and the
  MMC-SPI driver by default. The intention here is to make v5.2 more
  usable for testers and users with RISC-V hardware"

* tag 'riscv-for-v5.2/fixes-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux:
  riscv: mm: Fix code comment
  dt-bindings: clock: sifive: add MIT license as an option for the header file
  dt-bindings: riscv: resolve 'make dt_binding_check' warnings
  riscv: dts: Re-organize the DT nodes
  RISC-V: defconfig: enable MMC & SPI for RISC-V

18 months agoMerge tag 'nfs-for-5.2-4' of git://git.linux-nfs.org/projects/anna/linux-nfs
Linus Torvalds [Sat, 29 Jun 2019 09:02:22 +0000 (17:02 +0800)]
Merge tag 'nfs-for-5.2-4' of git://git.linux-nfs.org/projects/anna/linux-nfs

Pull two more NFS client fixes from Anna Schumaker:
 "These are both stable fixes.

  One to calculate the correct client message length in the case of
  partial transmissions. And the other to set the proper TCP timeout for
  flexfiles"

* tag 'nfs-for-5.2-4' of git://git.linux-nfs.org/projects/anna/linux-nfs:
  NFS/flexfiles: Use the correct TCP timeout for flexfiles I/O
  SUNRPC: Fix up calculation of client message length

18 months agoMerge tag 'ceph-for-5.2-rc7' of git://github.com/ceph/ceph-client
Linus Torvalds [Sat, 29 Jun 2019 09:01:02 +0000 (17:01 +0800)]
Merge tag 'ceph-for-5.2-rc7' of git://github.com/ceph/ceph-client

Pull ceph fix from Ilya Dryomov:
 "A small fix for a potential -rc1 regression from Jeff"

* tag 'ceph-for-5.2-rc7' of git://github.com/ceph/ceph-client:
  ceph: fix ceph_mdsc_build_path to not stop on first component

18 months agoMerge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Linus Torvalds [Sat, 29 Jun 2019 08:59:45 +0000 (16:59 +0800)]
Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fix from James Bottomley:
 "One simple fix for a driver use after free"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: vmw_pscsi: Fix use-after-free in pvscsi_queue_lck()

18 months agoMerge tag 'for-linus-20190628' of git://git.kernel.dk/linux-block
Linus Torvalds [Sat, 29 Jun 2019 08:58:35 +0000 (16:58 +0800)]
Merge tag 'for-linus-20190628' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "Just two small fixes.

  One from Paolo, fixing a silly mistake in BFQ. The other one is from
  me, ensuring that we have ->file cleared in the io_uring request a bit
  earlier. That avoids a use-before-free, if we encounter an error
  before ->file is assigned"

* tag 'for-linus-20190628' of git://git.kernel.dk/linux-block:
  block, bfq: fix operator in BFQQ_TOTALLY_SEEKY
  io_uring: ensure req->file is cleared on allocation

18 months agoMerge tag 'pinctrl-v5.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Sat, 29 Jun 2019 08:51:10 +0000 (16:51 +0800)]
Merge tag 'pinctrl-v5.2-3' of git://git./linux/kernel/git/linusw/linux-pinctrl

Pull pin control fixes from Linus Walleij:
 "Sorry to bomb in fixes this late. Maybe I can comfort you by saying it
  is only driver fixes, and mostly IRQ handling which is something GPIO
  and pin control drivers never get right. You think it works and then
  it doesn't.

  Summary:

   - Fix IRQ setup in the MCP23s08.

   - Fix pin setup on pins > 31 in the Ocelot driver.

   - Fix IRQs in the Mediatek driver"

* tag 'pinctrl-v5.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: mediatek: Update cur_mask in mask/mask ops
  pinctrl: mediatek: Ignore interrupts that are wake only during resume
  pinctrl: ocelot: fix pinmuxing for pins after 31
  pinctrl: ocelot: fix gpio direction for pins after 31
  pinctrl: mcp23s08: Fix add_data and irqchip_add_nested call order

18 months agolinux/kernel.h: fix overflow for DIV_ROUND_UP_ULL
Vinod Koul [Fri, 28 Jun 2019 19:07:21 +0000 (12:07 -0700)]
linux/kernel.h: fix overflow for DIV_ROUND_UP_ULL

DIV_ROUND_UP_ULL adds the two arguments and then invokes
DIV_ROUND_DOWN_ULL.  But on a 32bit system the addition of two 32 bit
values can overflow.  DIV_ROUND_DOWN_ULL does it correctly and stashes
the addition into a unsigned long long so cast the result to unsigned
long long here to avoid the overflow condition.

[akpm@linux-foundation.org: DIV_ROUND_UP_ULL must be an rval]
Link: http://lkml.kernel.org/r/20190625100518.30753-1-vkoul@kernel.org
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm, swap: fix THP swap out
Huang Ying [Fri, 28 Jun 2019 19:07:18 +0000 (12:07 -0700)]
mm, swap: fix THP swap out

0-Day test system reported some OOM regressions for several THP
(Transparent Huge Page) swap test cases.  These regressions are bisected
to 6861428921b5 ("block: always define BIO_MAX_PAGES as 256").  In the
commit, BIO_MAX_PAGES is set to 256 even when THP swap is enabled.  So the
bio_alloc(gfp_flags, 512) in get_swap_bio() may fail when swapping out
THP.  That causes the OOM.

As in the patch description of 6861428921b5 ("block: always define
BIO_MAX_PAGES as 256"), THP swap should use multi-page bvec to write THP
to swap space.  So the issue is fixed via doing that in get_swap_bio().

BTW: I remember I have checked the THP swap code when 6861428921b5
("block: always define BIO_MAX_PAGES as 256") was merged, and thought the
THP swap code needn't to be changed.  But apparently, I was wrong.  I
should have done this at that time.

Link: http://lkml.kernel.org/r/20190624075515.31040-1-ying.huang@intel.com
Fixes: 6861428921b5 ("block: always define BIO_MAX_PAGES as 256")
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agofork,memcg: alloc_thread_stack_node needs to set tsk->stack
Andrea Arcangeli [Fri, 28 Jun 2019 19:07:14 +0000 (12:07 -0700)]
fork,memcg: alloc_thread_stack_node needs to set tsk->stack

Commit 5eed6f1dff87 ("fork,memcg: fix crash in free_thread_stack on
memcg charge fail") corrected two instances, but there was a third
instance of this bug.

Without setting tsk->stack, if memcg_charge_kernel_stack fails, it'll
execute free_thread_stack() on a dangling pointer.

Enterprise kernels are compiled with VMAP_STACK=y so this isn't
critical, but custom VMAP_STACK=n builds should have some performance
advantage, with the drawback of risking to fail fork because compaction
didn't succeed.  So as long as VMAP_STACK=n is a supported option it's
worth fixing it upstream.

Link: http://lkml.kernel.org/r/20190619011450.28048-1-aarcange@redhat.com
Fixes: 9b6f7e163cd0 ("mm: rework memcg kernel stack accounting")
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: Roman Gushchin <guro@fb.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agoMAINTAINERS: add CLANG/LLVM BUILD SUPPORT info
Nick Desaulniers [Fri, 28 Jun 2019 19:07:12 +0000 (12:07 -0700)]
MAINTAINERS: add CLANG/LLVM BUILD SUPPORT info

Add keyword support so that our mailing list gets cc'ed for clang/llvm
patches.  We're pretty active on our mailing list so far as code review.
There are numerous Googlers like myself that are paid to support
building the Linux kernel with Clang and LLVM.

Link: http://lkml.kernel.org/r/20190620001907.255803-1-ndesaulniers@google.com
Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nathan Chancellor <natechancellor@gmail.com>
Cc: Joe Perches <joe@perches.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm/vmalloc.c: avoid bogus -Wmaybe-uninitialized warning
Arnd Bergmann [Fri, 28 Jun 2019 19:07:09 +0000 (12:07 -0700)]
mm/vmalloc.c: avoid bogus -Wmaybe-uninitialized warning

gcc gets confused in pcpu_get_vm_areas() because there are too many
branches that affect whether 'lva' was initialized before it gets used:

  mm/vmalloc.c: In function 'pcpu_get_vm_areas':
  mm/vmalloc.c:991:4: error: 'lva' may be used uninitialized in this function [-Werror=maybe-uninitialized]
      insert_vmap_area_augment(lva, &va->rb_node,
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
       &free_vmap_area_root, &free_vmap_area_list);
       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  mm/vmalloc.c:916:20: note: 'lva' was declared here
    struct vmap_area *lva;
                      ^~~

Add an intialization to NULL, and check whether this has changed before
the first use.

[akpm@linux-foundation.org: tweak comments]
Link: http://lkml.kernel.org/r/20190618092650.2943749-1-arnd@arndb.de
Fixes: 68ad4a330433 ("mm/vmalloc.c: keep track of free blocks for vmap allocation")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Joel Fernandes <joelaf@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm/page_idle.c: fix oops because end_pfn is larger than max_pfn
Colin Ian King [Fri, 28 Jun 2019 19:07:05 +0000 (12:07 -0700)]
mm/page_idle.c: fix oops because end_pfn is larger than max_pfn

Currently the calcuation of end_pfn can round up the pfn number to more
than the actual maximum number of pfns, causing an Oops.  Fix this by
ensuring end_pfn is never more than max_pfn.

This can be easily triggered when on systems where the end_pfn gets
rounded up to more than max_pfn using the idle-page stress-ng stress test:

sudo stress-ng --idle-page 0

  BUG: unable to handle kernel paging request at 00000000000020d8
  #PF error: [normal kernel read fault]
  PGD 0 P4D 0
  Oops: 0000 [#1] SMP PTI
  CPU: 1 PID: 11039 Comm: stress-ng-idle- Not tainted 5.0.0-5-generic #6-Ubuntu
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
  RIP: 0010:page_idle_get_page+0xc8/0x1a0
  Code: 0f b1 0a 75 7d 48 8b 03 48 89 c2 48 c1 e8 33 83 e0 07 48 c1 ea 36 48 8d 0c 40 4c 8d 24 88 49 c1 e4 07 4c 03 24 d5 00 89 c3 be <49> 8b 44 24 58 48 8d b8 80 a1 02 00 e8 07 d5 77 00 48 8b 53 08 48
  RSP: 0018:ffffafd7c672fde8 EFLAGS: 00010202
  RAX: 0000000000000005 RBX: ffffe36341fff700 RCX: 000000000000000f
  RDX: 0000000000000284 RSI: 0000000000000275 RDI: 0000000001fff700
  RBP: ffffafd7c672fe00 R08: ffffa0bc34056410 R09: 0000000000000276
  R10: ffffa0bc754e9b40 R11: ffffa0bc330f6400 R12: 0000000000002080
  R13: ffffe36341fff700 R14: 0000000000080000 R15: ffffa0bc330f6400
  FS: 00007f0ec1ea5740(0000) GS:ffffa0bc7db00000(0000) knlGS:0000000000000000
  CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00000000000020d8 CR3: 0000000077d68000 CR4: 00000000000006e0
  Call Trace:
    page_idle_bitmap_write+0x8c/0x140
    sysfs_kf_bin_write+0x5c/0x70
    kernfs_fop_write+0x12e/0x1b0
    __vfs_write+0x1b/0x40
    vfs_write+0xab/0x1b0
    ksys_write+0x55/0xc0
    __x64_sys_write+0x1a/0x20
    do_syscall_64+0x5a/0x110
    entry_SYSCALL_64_after_hwframe+0x44/0xa9

Link: http://lkml.kernel.org/r/20190618124352.28307-1-colin.king@canonical.com
Fixes: 33c3fc71c8cf ("mm: introduce idle page tracking")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agoinitramfs: fix populate_initrd_image() section mismatch
Geert Uytterhoeven [Fri, 28 Jun 2019 19:07:03 +0000 (12:07 -0700)]
initramfs: fix populate_initrd_image() section mismatch

With gcc-4.6.3:

    WARNING: vmlinux.o(.text.unlikely+0x140): Section mismatch in reference from the function populate_initrd_image() to the variable .init.ramfs.info:__initramfs_size
    The function populate_initrd_image() references
    the variable __init __initramfs_size.
    This is often because populate_initrd_image lacks a __init
    annotation or the annotation of __initramfs_size is wrong.

    WARNING: vmlinux.o(.text.unlikely+0x14c): Section mismatch in reference from the function populate_initrd_image() to the function .init.text:unpack_to_rootfs()
    The function populate_initrd_image() references
    the function __init unpack_to_rootfs().
    This is often because populate_initrd_image lacks a __init
    annotation or the annotation of unpack_to_rootfs is wrong.

    WARNING: vmlinux.o(.text.unlikely+0x198): Section mismatch in reference from the function populate_initrd_image() to the function .init.text:xwrite()
    The function populate_initrd_image() references
    the function __init xwrite().
    This is often because populate_initrd_image lacks a __init
    annotation or the annotation of xwrite is wrong.

Indeed, if the compiler decides not to inline populate_initrd_image(), a
warning is generated.

Fix this by adding the missing __init annotations.

Link: http://lkml.kernel.org/r/20190617074340.12779-1-geert@linux-m68k.org
Fixes: 7c184ecd262fe64f ("initramfs: factor out a helper to populate the initrd image")
Signed-off-by: Geert Uytterhoeven <geert@linux-m68k.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm/oom_kill.c: fix uninitialized oc->constraint
Yafang Shao [Fri, 28 Jun 2019 19:06:59 +0000 (12:06 -0700)]
mm/oom_kill.c: fix uninitialized oc->constraint

In dump_oom_summary() oc->constraint is used to show oom_constraint_text,
but it hasn't been set before.  So the value of it is always the default
value 0.  We should inititialize it before.

Bellow is the output when memcg oom occurs,

before this patch:
  oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null), cpuset=/,mems_allowed=0,oom_memcg=/foo,task_memcg=/foo,task=bash,pid=7997,uid=0

after this patch:
  oom-kill:constraint=CONSTRAINT_MEMCG,nodemask=(null), cpuset=/,mems_allowed=0,oom_memcg=/foo,task_memcg=/foo,task=bash,pid=13681,uid=0

Link: http://lkml.kernel.org/r/1560522038-15879-1-git-send-email-laoar.shao@gmail.com
Fixes: ef8444ea01d7 ("mm, oom: reorganize the oom report in dump_header")
Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Wind Yu <yuzhoujian@didichuxing.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm: hugetlb: soft-offline: dissolve_free_huge_page() return zero on !PageHuge
Naoya Horiguchi [Fri, 28 Jun 2019 19:06:56 +0000 (12:06 -0700)]
mm: hugetlb: soft-offline: dissolve_free_huge_page() return zero on !PageHuge

madvise(MADV_SOFT_OFFLINE) often returns -EBUSY when calling soft offline
for hugepages with overcommitting enabled.  That was caused by the
suboptimal code in current soft-offline code.  See the following part:

    ret = migrate_pages(&pagelist, new_page, NULL, MPOL_MF_MOVE_ALL,
                            MIGRATE_SYNC, MR_MEMORY_FAILURE);
    if (ret) {
            ...
    } else {
            /*
             * We set PG_hwpoison only when the migration source hugepage
             * was successfully dissolved, because otherwise hwpoisoned
             * hugepage remains on free hugepage list, then userspace will
             * find it as SIGBUS by allocation failure. That's not expected
             * in soft-offlining.
             */
            ret = dissolve_free_huge_page(page);
            if (!ret) {
                    if (set_hwpoison_free_buddy_page(page))
                            num_poisoned_pages_inc();
            }
    }
    return ret;

Here dissolve_free_huge_page() returns -EBUSY if the migration source page
was freed into buddy in migrate_pages(), but even in that case we actually
has a chance that set_hwpoison_free_buddy_page() succeeds.  So that means
current code gives up offlining too early now.

dissolve_free_huge_page() checks that a given hugepage is suitable for
dissolving, where we should return success for !PageHuge() case because
the given hugepage is considered as already dissolved.

This change also affects other callers of dissolve_free_huge_page(), which
are cleaned up together.

[n-horiguchi@ah.jp.nec.com: v3]
Link: http://lkml.kernel.org/r/1560761476-4651-3-git-send-email-n-horiguchi@ah.jp.nec.comLink:
Fixes: 6bc9b56433b76 ("mm: fix race on soft-offlining")
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Reported-by: Chen, Jerry T <jerry.t.chen@intel.com>
Tested-by: Chen, Jerry T <jerry.t.chen@intel.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Xishi Qiu <xishi.qiuxishi@alibaba-inc.com>
Cc: "Chen, Jerry T" <jerry.t.chen@intel.com>
Cc: "Zhuo, Qiuxu" <qiuxu.zhuo@intel.com>
Cc: <stable@vger.kernel.org> [4.19+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agomm: soft-offline: return -EBUSY if set_hwpoison_free_buddy_page() fails
Naoya Horiguchi [Fri, 28 Jun 2019 19:06:53 +0000 (12:06 -0700)]
mm: soft-offline: return -EBUSY if set_hwpoison_free_buddy_page() fails

The pass/fail of soft offline should be judged by checking whether the
raw error page was finally contained or not (i.e.  the result of
set_hwpoison_free_buddy_page()), but current code do not work like
that.  It might lead us to misjudge the test result when
set_hwpoison_free_buddy_page() fails.

Without this fix, there are cases where madvise(MADV_SOFT_OFFLINE) may
not offline the original page and will not return an error.

Link: http://lkml.kernel.org/r/1560154686-18497-2-git-send-email-n-horiguchi@ah.jp.nec.com
Signed-off-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Fixes: 6bc9b56433b76 ("mm: fix race on soft-offlining")
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Xishi Qiu <xishi.qiuxishi@alibaba-inc.com>
Cc: "Chen, Jerry T" <jerry.t.chen@intel.com>
Cc: "Zhuo, Qiuxu" <qiuxu.zhuo@intel.com>
Cc: <stable@vger.kernel.org> [4.19+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
18 months agosignal: remove the wrong signal_pending() check in restore_user_sigmask()
Oleg Nesterov [Fri, 28 Jun 2019 19:06:50 +0000 (12:06 -0700)]
signal: remove the wrong signal_pending() check in restore_user_sigmask()

This is the minimal fix for stable, I'll send cleanups later.

Commit 854a6ed56839 ("signal: Add restore_user_sigmask()") introduced
the visible change which breaks user-space: a signal temporary unblocked
by set_user_sigmask() can be delivered even if the caller returns
success or timeout.

Change restore_user_sigmask() to accept the additional "interrupted"
argument which should be used instead of signal_pending() check, and
update the callers.

Eric said:

: For clarity.  I don't think this is required by posix, or fundamentally to
: remove the races in select.  It is what linux has always done and we have
: applications who care so I agree this fix is needed.
:
: Further in any case where the semantic change that this patch rolls back
: (aka where allowing a signal to be delivered and the select like call to
: complete) would be advantage we can do as well if not better by using
: signalfd.
:
: Michael is there any chance we can get this guarantee of the linux
: implementation of pselect and friends clearly documented.  The guarantee
: that if the system call completes successfully we are guaranteed that no
: signal that is unblocked by using sigmask will be delivered?

Link: http://lkml.kernel.org/r/20190604134117.GA29963@redhat.com
Fixes: 854a6ed56839a40f6b5d02a2962f48841482eec4 ("signal: Add restore_user_sigmask()")
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Eric Wong <e@80x24.org>
Tested-by: Eric Wong <e@80x24.org>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Deepa Dinamani <deepa.kernel@gmail.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Jason Baron <jbaron@akamai.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: David Laight <David.Laight@ACULAB.COM>
Cc: <stable@vger.kernel.org> [5.0+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>