]> asedeno.scripts.mit.edu Git - linux.git/log
linux.git
6 years agoBtrfs: fix error handling in btrfs_truncate()
Omar Sandoval [Tue, 22 May 2018 16:47:58 +0000 (09:47 -0700)]
Btrfs: fix error handling in btrfs_truncate()

Jun Wu at Facebook reported that an internal service was seeing a return
value of 1 from ftruncate() on Btrfs in some cases. This is coming from
the NEED_TRUNCATE_BLOCK return value from btrfs_truncate_inode_items().

btrfs_truncate() uses two variables for error handling, ret and err.
When btrfs_truncate_inode_items() returns non-zero, we set err to the
return value. However, NEED_TRUNCATE_BLOCK is not an error. Make sure we
only set err if ret is an error (i.e., negative).

To reproduce the issue: mount a filesystem with -o compress-force=zstd
and the following program will encounter return value of 1 from
ftruncate:

int main(void) {
        char buf[256] = { 0 };
        int ret;
        int fd;

        fd = open("test", O_CREAT | O_WRONLY | O_TRUNC, 0666);
        if (fd == -1) {
                perror("open");
                return EXIT_FAILURE;
        }

        if (write(fd, buf, sizeof(buf)) != sizeof(buf)) {
                perror("write");
                close(fd);
                return EXIT_FAILURE;
        }

        if (fsync(fd) == -1) {
                perror("fsync");
                close(fd);
                return EXIT_FAILURE;
        }

        ret = ftruncate(fd, 128);
        if (ret) {
                printf("ftruncate() returned %d\n", ret);
                close(fd);
                return EXIT_FAILURE;
        }

        close(fd);
        return EXIT_SUCCESS;
}

Fixes: ddfae63cc8e0 ("btrfs: move btrfs_truncate_block out of trans handle")
CC: stable@vger.kernel.org # 4.15+
Reported-by: Jun Wu <quark@fb.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix crash when trying to resume balance without the resume flag
Anand Jain [Thu, 17 May 2018 07:16:51 +0000 (15:16 +0800)]
btrfs: fix crash when trying to resume balance without the resume flag

We set the BTRFS_BALANCE_RESUME flag in the btrfs_recover_balance()
only, which isn't called during the remount. So when resuming from
the paused balance we hit the bug:

 kernel: kernel BUG at fs/btrfs/volumes.c:3890!
 ::
 kernel:  balance_kthread+0x51/0x60 [btrfs]
 kernel:  kthread+0x111/0x130
 ::
 kernel: RIP: btrfs_balance+0x12e1/0x1570 [btrfs] RSP: ffffba7d0090bde8

Reproducer:
  On a mounted filesystem:

  btrfs balance start --full-balance /btrfs
  btrfs balance pause /btrfs
  mount -o remount,ro /dev/sdb /btrfs
  mount -o remount,rw /dev/sdb /btrfs

To fix this set the BTRFS_BALANCE_RESUME flag in
btrfs_resume_balance_async().

CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Fix delalloc inodes invalidation during transaction abort
Nikolay Borisov [Fri, 27 Apr 2018 09:21:53 +0000 (12:21 +0300)]
btrfs: Fix delalloc inodes invalidation during transaction abort

When a transaction is aborted btrfs_cleanup_transaction is called to
cleanup all the various in-flight bits and pieces which migth be
active. One of those is delalloc inodes - inodes which have dirty
pages which haven't been persisted yet. Currently the process of
freeing such delalloc inodes in exceptional circumstances such as
transaction abort boiled down to calling btrfs_invalidate_inodes whose
sole job is to invalidate the dentries for all inodes related to a
root. This is in fact wrong and insufficient since such delalloc inodes
will likely have pending pages or ordered-extents and will be linked to
the sb->s_inode_list. This means that unmounting a btrfs instance with
an aborted transaction could potentially lead inodes/their pages
visible to the system long after their superblock has been freed. This
in turn leads to a "use-after-free" situation once page shrink is
triggered. This situation could be simulated by running generic/019
which would cause such inodes to be left hanging, followed by
generic/176 which causes memory pressure and page eviction which lead
to touching the freed super block instance. This situation is
additionally detected by the unmount code of VFS with the following
message:

"VFS: Busy inodes after unmount of Self-destruct in 5 seconds.  Have a nice day..."

Additionally btrfs hits WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
in free_fs_root for the same reason.

This patch aims to rectify the sitaution by doing the following:

1. Change btrfs_destroy_delalloc_inodes so that it calls
invalidate_inode_pages2 for every inode on the delalloc list, this
ensures that all the pages of the inode are released. This function
boils down to calling btrfs_releasepage. During test I observed cases
where inodes on the delalloc list were having an i_count of 0, so this
necessitates using igrab to be sure we are working on a non-freed inode.

2. Since calling btrfs_releasepage might queue delayed iputs move the
call out to btrfs_cleanup_transaction in btrfs_error_commit_super before
calling run_delayed_iputs for the last time. This is necessary to ensure
that delayed iputs are run.

Note: this patch is tagged for 4.14 stable but the fix applies to older
versions too but needs to be backported manually due to conflicts.

CC: stable@vger.kernel.org # 4.14.x: 2b8773313494: btrfs: Split btrfs_del_delalloc_inode into 2 functions
CC: stable@vger.kernel.org # 4.14.x
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment to igrab ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Split btrfs_del_delalloc_inode into 2 functions
Nikolay Borisov [Fri, 27 Apr 2018 09:21:51 +0000 (12:21 +0300)]
btrfs: Split btrfs_del_delalloc_inode into 2 functions

This is in preparation of fixing delalloc inodes leakage on transaction
abort. Also export the new function.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix reading stale metadata blocks after degraded raid1 mounts
Liu Bo [Tue, 15 May 2018 17:37:36 +0000 (01:37 +0800)]
btrfs: fix reading stale metadata blocks after degraded raid1 mounts

If a btree block, aka. extent buffer, is not available in the extent
buffer cache, it'll be read out from the disk instead, i.e.

btrfs_search_slot()
  read_block_for_search()  # hold parent and its lock, go to read child
    btrfs_release_path()
    read_tree_block()  # read child

Unfortunately, the parent lock got released before reading child, so
commit 5bdd3536cbbe ("Btrfs: Fix block generation verification race") had
used 0 as parent transid to read the child block.  It forces
read_tree_block() not to check if parent transid is different with the
generation id of the child that it reads out from disk.

A simple PoC is included in btrfs/124,

0. A two-disk raid1 btrfs,

1. Right after mkfs.btrfs, block A is allocated to be device tree's root.

2. Mount this filesystem and put it in use, after a while, device tree's
   root got COW but block A hasn't been allocated/overwritten yet.

3. Umount it and reload the btrfs module to remove both disks from the
   global @fs_devices list.

4. mount -odegraded dev1 and write some data, so now block A is allocated
   to be a leaf in checksum tree.  Note that only dev1 has the latest
   metadata of this filesystem.

5. Umount it and mount it again normally (with both disks), since raid1
   can pick up one disk by the writer task's pid, if btrfs_search_slot()
   needs to read block A, dev2 which does NOT have the latest metadata
   might be read for block A, then we got a stale block A.

6. As parent transid is not checked, block A is marked as uptodate and
   put into the extent buffer cache, so the future search won't bother
   to read disk again, which means it'll make changes on this stale
   one and make it dirty and flush it onto disk.

To avoid the problem, parent transid needs to be passed to
read_tree_block().

In order to get a valid parent transid, we need to hold the parent's
lock until finishing reading child.

This patch needs to be slightly adapted for stable kernels, the
&first_key parameter added to read_tree_block() is from 4.16+
(581c1760415c4). The fix is to replace 0 by 'gen'.

Fixes: 5bdd3536cbbe ("Btrfs: Fix block generation verification race")
CC: stable@vger.kernel.org # 4.4+
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: property: Set incompat flag if lzo/zstd compression is set
Misono Tomohiro [Tue, 15 May 2018 07:51:26 +0000 (16:51 +0900)]
btrfs: property: Set incompat flag if lzo/zstd compression is set

Incompat flag of LZO/ZSTD compression should be set at:

 1. mount time (-o compress/compress-force)
 2. when defrag is done
 3. when property is set

Currently 3. is missing and this commit adds this.

This could lead to a filesystem that uses ZSTD but is not marked as
such. If a kernel without a ZSTD support encounteres a ZSTD compressed
extent, it will handle that but this could be confusing to the user.

Typically the filesystem is mounted with the ZSTD option, but the
discrepancy can arise when a filesystem is never mounted with ZSTD and
then the property on some file is set (and some new extents are
written). A simple mount with -o compress=zstd will fix that up on an
unpatched kernel.

Same goes for LZO, but this has been around for a very long time
(2.6.37) so it's unlikely that a pre-LZO kernel would be used.

Fixes: 5c1aab1dd544 ("btrfs: Add zstd support")
CC: stable@vger.kernel.org # 4.14+
Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add user visible impact ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix duplicate extents after fsync of file with prealloc extents
Filipe Manana [Wed, 9 May 2018 15:01:46 +0000 (16:01 +0100)]
Btrfs: fix duplicate extents after fsync of file with prealloc extents

In commit 471d557afed1 ("Btrfs: fix loss of prealloc extents past i_size
after fsync log replay"), on fsync,  we started to always log all prealloc
extents beyond an inode's i_size in order to avoid losing them after a
power failure. However under some cases this can lead to the log replay
code to create duplicate extent items, with different lengths, in the
extent tree. That happens because, as of that commit, we can now log
extent items based on extent maps that are not on the "modified" list
of extent maps of the inode's extent map tree. Logging extent items based
on extent maps is used during the fast fsync path to save time and for
this to work reliably it requires that the extent maps are not merged
with other adjacent extent maps - having the extent maps in the list
of modified extents gives such guarantee.

Consider the following example, captured during a long run of fsstress,
which illustrates this problem.

We have inode 271, in the filesystem tree (root 5), for which all of the
following operations and discussion apply to.

A buffered write starts at offset 312391 with a length of 933471 bytes
(end offset at 1245862). At this point we have, for this inode, the
following extent maps with the their field values:

em A, start 0, orig_start 0, len 40960, block_start 18446744073709551613,
      block_len 0, orig_block_len 0
em B, start 40960, orig_start 40960, len 376832, block_start 1106399232,
      block_len 376832, orig_block_len 376832
em C, start 417792, orig_start 417792, len 782336, block_start
      18446744073709551613, block_len 0, orig_block_len 0
em D, start 1200128, orig_start 1200128, len 835584, block_start
      1106776064, block_len 835584, orig_block_len 835584
em E, start 2035712, orig_start 2035712, len 245760, block_start
      1107611648, block_len 245760, orig_block_len 245760

Extent map A corresponds to a hole and extent maps D and E correspond to
preallocated extents.

Extent map D ends where extent map E begins (1106776064 + 835584 =
1107611648), but these extent maps were not merged because they are in
the inode's list of modified extent maps.

An fsync against this inode is made, which triggers the fast path
(BTRFS_INODE_NEEDS_FULL_SYNC is not set). This fsync triggers writeback
of the data previously written using buffered IO, and when the respective
ordered extent finishes, btrfs_drop_extents() is called against the
(aligned) range 311296..1249279. This causes a split of extent map D at
btrfs_drop_extent_cache(), replacing extent map D with a new extent map
D', also added to the list of modified extents,  with the following
values:

em D', start 1249280, orig_start of 1200128,
       block_start 1106825216 (= 1106776064 + 1249280 - 1200128),
       orig_block_len 835584,
       block_len 786432 (835584 - (1249280 - 1200128))

Then, during the fast fsync, btrfs_log_changed_extents() is called and
extent maps D' and E are removed from the list of modified extents. The
flag EXTENT_FLAG_LOGGING is also set on them. After the extents are logged
clear_em_logging() is called on each of them, and that makes extent map E
to be merged with extent map D' (try_merge_map()), resulting in D' being
deleted and E adjusted to:

em E, start 1249280, orig_start 1200128, len 1032192,
      block_start 1106825216, block_len 1032192,
      orig_block_len 245760

A direct IO write at offset 1847296 and length of 360448 bytes (end offset
at 2207744) starts, and at that moment the following extent maps exist for
our inode:

em A, start 0, orig_start 0, len 40960, block_start 18446744073709551613,
      block_len 0, orig_block_len 0
em B, start 40960, orig_start 40960, len 270336, block_start 1106399232,
      block_len 270336, orig_block_len 376832
em C, start 311296, orig_start 311296, len 937984, block_start 1112842240,
      block_len 937984, orig_block_len 937984
em E (prealloc), start 1249280, orig_start 1200128, len 1032192,
      block_start 1106825216, block_len 1032192, orig_block_len 245760

The dio write results in drop_extent_cache() being called twice. The first
time for a range that starts at offset 1847296 and ends at offset 2035711
(length of 188416), which results in a double split of extent map E,
replacing it with two new extent maps:

em F, start 1249280, orig_start 1200128, block_start 1106825216,
      block_len 598016, orig_block_len 598016
em G, start 2035712, orig_start 1200128, block_start 1107611648,
      block_len 245760, orig_block_len 1032192

It also creates a new extent map that represents a part of the requested
IO (through create_io_em()):

em H, start 1847296, len 188416, block_start 1107423232, block_len 188416

The second call to drop_extent_cache() has a range with a start offset of
2035712 and end offset of 2207743 (length of 172032). This leads to
replacing extent map G with a new extent map I with the following values:

em I, start 2207744, orig_start 1200128, block_start 1107783680,
      block_len 73728, orig_block_len 1032192

It also creates a new extent map that represents the second part of the
requested IO (through create_io_em()):

em J, start 2035712, len 172032, block_start 1107611648, block_len 172032

The dio write set the inode's i_size to 2207744 bytes.

After the dio write the inode has the following extent maps:

em A, start 0, orig_start 0, len 40960, block_start 18446744073709551613,
      block_len 0, orig_block_len 0
em B, start 40960, orig_start 40960, len 270336, block_start 1106399232,
      block_len 270336, orig_block_len 376832
em C, start 311296, orig_start 311296, len 937984, block_start 1112842240,
      block_len 937984, orig_block_len 937984
em F, start 1249280, orig_start 1200128, len 598016,
      block_start 1106825216, block_len 598016, orig_block_len 598016
em H, start 1847296, orig_start 1200128, len 188416,
      block_start 1107423232, block_len 188416, orig_block_len 835584
em J, start 2035712, orig_start 2035712, len 172032,
      block_start 1107611648, block_len 172032, orig_block_len 245760
em I, start 2207744, orig_start 1200128, len 73728,
      block_start 1107783680, block_len 73728, orig_block_len 1032192

Now do some change to the file, like adding a xattr for example and then
fsync it again. This triggers a fast fsync path, and as of commit
471d557afed1 ("Btrfs: fix loss of prealloc extents past i_size after fsync
log replay"), we use the extent map I to log a file extent item because
it's a prealloc extent and it starts at an offset matching the inode's
i_size. However when we log it, we create a file extent item with a value
for the disk byte location that is wrong, as can be seen from the
following output of "btrfs inspect-internal dump-tree":

 item 1 key (271 EXTENT_DATA 2207744) itemoff 3782 itemsize 53
     generation 22 type 2 (prealloc)
     prealloc data disk byte 1106776064 nr 1032192
     prealloc data offset 1007616 nr 73728

Here the disk byte value corresponds to calculation based on some fields
from the extent map I:

  1106776064 = block_start (1107783680) - 1007616 (extent_offset)
  extent_offset = 2207744 (start) - 1200128 (orig_start) = 1007616

The disk byte value of 1106776064 clashes with disk byte values of the
file extent items at offsets 1249280 and 1847296 in the fs tree:

        item 6 key (271 EXTENT_DATA 1249280) itemoff 3568 itemsize 53
                generation 20 type 2 (prealloc)
                prealloc data disk byte 1106776064 nr 835584
                prealloc data offset 49152 nr 598016
        item 7 key (271 EXTENT_DATA 1847296) itemoff 3515 itemsize 53
                generation 20 type 1 (regular)
                extent data disk byte 1106776064 nr 835584
                extent data offset 647168 nr 188416 ram 835584
                extent compression 0 (none)
        item 8 key (271 EXTENT_DATA 2035712) itemoff 3462 itemsize 53
                generation 20 type 1 (regular)
                extent data disk byte 1107611648 nr 245760
                extent data offset 0 nr 172032 ram 245760
                extent compression 0 (none)
        item 9 key (271 EXTENT_DATA 2207744) itemoff 3409 itemsize 53
                generation 20 type 2 (prealloc)
                prealloc data disk byte 1107611648 nr 245760
                prealloc data offset 172032 nr 73728

Instead of the disk byte value of 1106776064, the value of 1107611648
should have been logged. Also the data offset value should have been
172032 and not 1007616.
After a log replay we end up getting two extent items in the extent tree
with different lengths, one of 835584, which is correct and existed
before the log replay, and another one of 1032192 which is wrong and is
based on the logged file extent item:

 item 12 key (1106776064 EXTENT_ITEM 835584) itemoff 3406 itemsize 53
    refs 2 gen 15 flags DATA
    extent data backref root 5 objectid 271 offset 1200128 count 2
 item 13 key (1106776064 EXTENT_ITEM 1032192) itemoff 3353 itemsize 53
    refs 1 gen 22 flags DATA
    extent data backref root 5 objectid 271 offset 1200128 count 1

Obviously this leads to many problems and a filesystem check reports many
errors:

 (...)
 checking extents
 Extent back ref already exists for 1106776064 parent 0 root 5 owner 271 offset 1200128 num_refs 1
 extent item 1106776064 has multiple extent items
 ref mismatch on [1106776064 835584] extent item 2, found 3
 Incorrect local backref count on 1106776064 root 5 owner 271 offset 1200128 found 2 wanted 1 back 0x55b1d0ad7680
 Backref 1106776064 root 5 owner 271 offset 1200128 num_refs 0 not found in extent tree
 Incorrect local backref count on 1106776064 root 5 owner 271 offset 1200128 found 1 wanted 0 back 0x55b1d0ad4e70
 Backref bytes do not match extent backref, bytenr=1106776064, ref bytes=835584, backref bytes=1032192
 backpointer mismatch on [1106776064 835584]
 checking free space cache
 block group 1103101952 has wrong amount of free space
 failed to load free space cache for block group 1103101952
 checking fs roots
 (...)

So fix this by logging the prealloc extents beyond the inode's i_size
based on searches in the subvolume tree instead of the extent maps.

Fixes: 471d557afed1 ("Btrfs: fix loss of prealloc extents past i_size after fsync log replay")
CC: stable@vger.kernel.org # 4.14+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix xattr loss after power failure
Filipe Manana [Fri, 11 May 2018 15:42:42 +0000 (16:42 +0100)]
Btrfs: fix xattr loss after power failure

If a file has xattrs, we fsync it, to ensure we clear the flags
BTRFS_INODE_NEEDS_FULL_SYNC and BTRFS_INODE_COPY_EVERYTHING from its
inode, the current transaction commits and then we fsync it (without
either of those bits being set in its inode), we end up not logging
all its xattrs. This results in deleting all xattrs when replying the
log after a power failure.

Trivial reproducer

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

  $ touch /mnt/foobar
  $ setfattr -n user.xa -v qwerty /mnt/foobar
  $ xfs_io -c "fsync" /mnt/foobar

  $ sync

  $ xfs_io -c "pwrite -S 0xab 0 64K" /mnt/foobar
  $ xfs_io -c "fsync" /mnt/foobar
  <power failure>

  $ mount /dev/sdb /mnt
  $ getfattr --absolute-names --dump /mnt/foobar
  <empty output>
  $

So fix this by making sure all xattrs are logged if we log a file's inode
item and neither the flags BTRFS_INODE_NEEDS_FULL_SYNC nor
BTRFS_INODE_COPY_EVERYTHING were set in the inode.

Fixes: 36283bf777d9 ("Btrfs: fix fsync xattr loss in the fast fsync path")
Cc: <stable@vger.kernel.org> # 4.2+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: send, fix invalid access to commit roots due to concurrent snapshotting
Robbie Ko [Mon, 14 May 2018 02:51:34 +0000 (10:51 +0800)]
Btrfs: send, fix invalid access to commit roots due to concurrent snapshotting

[BUG]
btrfs incremental send BUG happens when creating a snapshot of snapshot
that is being used by send.

[REASON]
The problem can happen if while we are doing a send one of the snapshots
used (parent or send) is snapshotted, because snapshoting implies COWing
the root of the source subvolume/snapshot.

1. When doing an incremental send, the send process will get the commit
   roots from the parent and send snapshots, and add references to them
   through extent_buffer_get().

2. When a snapshot/subvolume is snapshotted, its root node is COWed
   (transaction.c:create_pending_snapshot()).

3. COWing releases the space used by the node immediately, through:

   __btrfs_cow_block()
   --btrfs_free_tree_block()
   ----btrfs_add_free_space(bytenr of node)

4. Because send doesn't hold a transaction open, it's possible that
   the transaction used to create the snapshot commits, switches the
   commit root and the old space used by the previous root node gets
   assigned to some other node allocation. Allocation of a new node will
   use the existing extent buffer found in memory, which we previously
   got a reference through extent_buffer_get(), and allow the extent
   buffer's content (pages) to be modified:

   btrfs_alloc_tree_block
   --btrfs_reserve_extent
   ----find_free_extent (get bytenr of old node)
   --btrfs_init_new_buffer (use bytenr of old node)
   ----btrfs_find_create_tree_block
   ------alloc_extent_buffer
   --------find_extent_buffer (get old node)

5. So send can access invalid memory content and have unpredictable
   behaviour.

[FIX]
So we fix the problem by copying the commit roots of the send and
parent snapshots and use those copies.

CallTrace looks like this:
 ------------[ cut here ]------------
 kernel BUG at fs/btrfs/ctree.c:1861!
 invalid opcode: 0000 [#1] SMP
 CPU: 6 PID: 24235 Comm: btrfs Tainted: P           O 3.10.105 #23721
 ffff88046652d680 ti: ffff88041b720000 task.ti: ffff88041b720000
 RIP: 0010:[<ffffffffa08dd0e8>] read_node_slot+0x108/0x110 [btrfs]
 RSP: 0018:ffff88041b723b68  EFLAGS: 00010246
 RAX: ffff88043ca6b000 RBX: ffff88041b723c50 RCX: ffff880000000000
 RDX: 000000000000004c RSI: ffff880314b133f8 RDI: ffff880458b24000
 RBP: 0000000000000000 R08: 0000000000000001 R09: ffff88041b723c66
 R10: 0000000000000001 R11: 0000000000001000 R12: ffff8803f3e48890
 R13: ffff8803f3e48880 R14: ffff880466351800 R15: 0000000000000001
 FS:  00007f8c321dc8c0(0000) GS:ffff88047fcc0000(0000)
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 R2: 00007efd1006d000 CR3: 0000000213a24000 CR4: 00000000003407e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Stack:
 ffff88041b723c50 ffff8803f3e48880 ffff8803f3e48890 ffff8803f3e48880
 ffff880466351800 0000000000000001 ffffffffa08dd9d7 ffff88041b723c50
 ffff8803f3e48880 ffff88041b723c66 ffffffffa08dde85 a9ff88042d2c4400
 Call Trace:
 [<ffffffffa08dd9d7>] ? tree_move_down.isra.33+0x27/0x50 [btrfs]
 [<ffffffffa08dde85>] ? tree_advance+0xb5/0xc0 [btrfs]
 [<ffffffffa08e83d4>] ? btrfs_compare_trees+0x2d4/0x760 [btrfs]
 [<ffffffffa0982050>] ? finish_inode_if_needed+0x870/0x870 [btrfs]
 [<ffffffffa09841ea>] ? btrfs_ioctl_send+0xeda/0x1050 [btrfs]
 [<ffffffffa094bd3d>] ? btrfs_ioctl+0x1e3d/0x33f0 [btrfs]
 [<ffffffff81111133>] ? handle_pte_fault+0x373/0x990
 [<ffffffff8153a096>] ? atomic_notifier_call_chain+0x16/0x20
 [<ffffffff81063256>] ? set_task_cpu+0xb6/0x1d0
 [<ffffffff811122c3>] ? handle_mm_fault+0x143/0x2a0
 [<ffffffff81539cc0>] ? __do_page_fault+0x1d0/0x500
 [<ffffffff81062f07>] ? check_preempt_curr+0x57/0x90
 [<ffffffff8115075a>] ? do_vfs_ioctl+0x4aa/0x990
 [<ffffffff81034f83>] ? do_fork+0x113/0x3b0
 [<ffffffff812dd7d7>] ? trace_hardirqs_off_thunk+0x3a/0x6c
 [<ffffffff81150cc8>] ? SyS_ioctl+0x88/0xa0
 [<ffffffff8153e422>] ? system_call_fastpath+0x16/0x1b
 ---[ end trace 29576629ee80b2e1 ]---

Fixes: 7069830a9e38 ("Btrfs: add btrfs_compare_trees function")
CC: stable@vger.kernel.org # 3.6+
Signed-off-by: Robbie Ko <robbieko@synology.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: send, fix missing truncate for inode with prealloc extent past eof
Filipe Manana [Mon, 30 Apr 2018 18:05:07 +0000 (19:05 +0100)]
Btrfs: send, fix missing truncate for inode with prealloc extent past eof

An incremental send operation can miss a truncate operation when an inode
has an increased size in the send snapshot and a prealloc extent beyond
its size.

Consider the following scenario where a necessary truncate operation is
missing in the incremental send stream:

1) In the parent snapshot an inode has a size of 1282957 bytes and it has
   no prealloc extents beyond its size;

2) In the the send snapshot it has a size of 5738496 bytes and has a new
   extent at offsets 1884160 (length of 106496 bytes) and a prealloc
   extent beyond eof at offset 6729728 (and a length of 339968 bytes);

3) When processing the prealloc extent, at offset 6729728, we end up at
   send.c:send_write_or_clone() and set the @len variable to a value of
   18446744073708560384 because @offset plus the original @len value is
   larger then the inode's size (6729728 + 339968 > 5738496). We then
   call send_extent_data(), with that @offset and @len, which in turn
   calls send_write(), and then the later calls fill_read_buf(). Because
   the offset passed to fill_read_buf() is greater then inode's i_size,
   this function returns 0 immediately, which makes send_write() and
   send_extent_data() do nothing and return immediately as well. When
   we get back to send.c:send_write_or_clone() we adjust the value
   of sctx->cur_inode_next_write_offset to @offset plus @len, which
   corresponds to 6729728 + 18446744073708560384 = 5738496, which is
   precisely the the size of the inode in the send snapshot;

4) Later when at send.c:finish_inode_if_needed() we determine that
   we don't need to issue a truncate operation because the value of
   sctx->cur_inode_next_write_offset corresponds to the inode's new
   size, 5738496 bytes. This is wrong because the last write operation
   that was issued started at offset 1884160 with a length of 106496
   bytes, so the correct value for sctx->cur_inode_next_write_offset
   should be 1990656 (1884160 + 106496), so that a truncate operation
   with a value of 5738496 bytes would have been sent to insert a
   trailing hole at the destination.

So fix the issue by making send.c:send_write_or_clone() not attempt
to send write or clone operations for extents that start beyond the
inode's size, since such attempts do nothing but waste time by
calling helper functions and allocating path structures, and send
currently has no fallocate command in order to create prealloc extents
at the destination (either beyond a file's eof or not).

The issue was found running the test btrfs/007 from fstests using a seed
value of 1524346151 for fsstress.

Reported-by: Gu, Jinxiang <gujx@cn.fujitsu.com>
Fixes: ffa7c4296e93 ("Btrfs: send, do not issue unnecessary truncate operations")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Take trans lock before access running trans in check_delayed_ref
ethanwu [Sun, 29 Apr 2018 07:59:42 +0000 (15:59 +0800)]
btrfs: Take trans lock before access running trans in check_delayed_ref

In preivous patch:
Btrfs: kill trans in run_delalloc_nocow and btrfs_cross_ref_exist
We avoid starting btrfs transaction and get this information from
fs_info->running_transaction directly.

When accessing running_transaction in check_delayed_ref, there's a
chance that current transaction will be freed by commit transaction
after the NULL pointer check of running_transaction is passed.

After looking all the other places using fs_info->running_transaction,
they are either protected by trans_lock or holding the transactions.

Fix this by using trans_lock and increasing the use_count.

Fixes: e4c3b2dcd144 ("Btrfs: kill trans in run_delalloc_nocow and btrfs_cross_ref_exist")
CC: stable@vger.kernel.org # 4.14+
Signed-off-by: ethanwu <ethanwu@synology.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Fix wrong first_key parameter in replace_path
Qu Wenruo [Mon, 23 Apr 2018 09:32:04 +0000 (17:32 +0800)]
btrfs: Fix wrong first_key parameter in replace_path

Commit 581c1760415c ("btrfs: Validate child tree block's level and first
key") introduced new @first_key parameter for read_tree_block(), however
caller in replace_path() is parasing wrong key to read_tree_block().

It should use parameter @first_key other than @key.

Normally it won't expose problem as @key is normally initialzied to the
same value of @first_key we expect.
However in relocation recovery case, @key can be set to (0, 0, 0), and
since no valid key in relocation tree can be (0, 0, 0), it will cause
read_tree_block() to return -EUCLEAN and interrupt relocation recovery.

Fix it by setting @first_key correctly.

Fixes: 581c1760415c ("btrfs: Validate child tree block's level and first key")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: print-tree: debugging output enhancement
Qu Wenruo [Wed, 11 Apr 2018 09:08:12 +0000 (17:08 +0800)]
btrfs: print-tree: debugging output enhancement

This patch enhances the following things:

- tree block header
  * add generation and owner output for node and leaf
- node pointer generation output
- allow btrfs_print_tree() to not follow nodes
  * just like btrfs-progs

Please note that, although function btrfs_print_tree() is not called by
anyone right now, it's still a pretty useful function to debug kernel.
So that function is still kept for later use.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Fix race condition between delayed refs and blockgroup removal
Nikolay Borisov [Wed, 18 Apr 2018 06:41:54 +0000 (09:41 +0300)]
btrfs: Fix race condition between delayed refs and blockgroup removal

When the delayed refs for a head are all run, eventually
cleanup_ref_head is called which (in case of deletion) obtains a
reference for the relevant btrfs_space_info struct by querying the bg
for the range. This is problematic because when the last extent of a
bg is deleted a race window emerges between removal of that bg and the
subsequent invocation of cleanup_ref_head. This can result in cache being null
and either a null pointer dereference or assertion failure.

task: ffff8d04d31ed080 task.stack: ffff9e5dc10cc000
RIP: 0010:assfail.constprop.78+0x18/0x1a [btrfs]
RSP: 0018:ffff9e5dc10cfbe8 EFLAGS: 00010292
RAX: 0000000000000044 RBX: 0000000000000000 RCX: 0000000000000000
RDX: ffff8d04ffc1f868 RSI: ffff8d04ffc178c8 RDI: ffff8d04ffc178c8
RBP: ffff8d04d29e5ea0 R08: 00000000000001f0 R09: 0000000000000001
R10: ffff9e5dc0507d58 R11: 0000000000000001 R12: ffff8d04d29e5ea0
R13: ffff8d04d29e5f08 R14: ffff8d04efe29b40 R15: ffff8d04efe203e0
FS:  00007fbf58ead500(0000) GS:ffff8d04ffc00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fe6c6975648 CR3: 0000000013b2a000 CR4: 00000000000006f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 __btrfs_run_delayed_refs+0x10e7/0x12c0 [btrfs]
 btrfs_run_delayed_refs+0x68/0x250 [btrfs]
 btrfs_should_end_transaction+0x42/0x60 [btrfs]
 btrfs_truncate_inode_items+0xaac/0xfc0 [btrfs]
 btrfs_evict_inode+0x4c6/0x5c0 [btrfs]
 evict+0xc6/0x190
 do_unlinkat+0x19c/0x300
 do_syscall_64+0x74/0x140
 entry_SYSCALL_64_after_hwframe+0x3d/0xa2
RIP: 0033:0x7fbf589c57a7

To fix this, introduce a new flag "is_system" to head_ref structs,
which is populated at insertion time. This allows to decouple the
querying for the spaceinfo from querying the possibly deleted bg.

Fixes: d7eae3403f46 ("Btrfs: rework delayed ref total_bytes_pinned accounting")
CC: stable@vger.kernel.org # 4.14+
Suggested-by: Omar Sandoval <osandov@osandov.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix unaligned access in readdir
David Sterba [Mon, 16 Apr 2018 19:10:14 +0000 (21:10 +0200)]
btrfs: fix unaligned access in readdir

The last update to readdir introduced a temporary buffer to store the
emitted readdir data, but as there are file names of variable length,
there's a lot of unaligned access.

This was observed on a sparc64 machine:

  Kernel unaligned access at TPC[102f3080] btrfs_real_readdir+0x51c/0x718 [btrfs]

Fixes: 23b5ec74943 ("btrfs: fix readdir deadlock with pagefault")
CC: stable@vger.kernel.org # 4.14+
Reported-and-tested-by: René Rebe <rene@exactcode.com>
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Fix wrong btrfs_delalloc_release_extents parameter
Qu Wenruo [Tue, 17 Apr 2018 10:43:58 +0000 (18:43 +0800)]
btrfs: Fix wrong btrfs_delalloc_release_extents parameter

Commit 43b18595d660 ("btrfs: qgroup: Use separate meta reservation type
for delalloc") merged into mainline is not the latest version submitted
to mail list in Dec 2017.

It has a fatal wrong @qgroup_free parameter, which results increasing
qgroup metadata pertrans reserved space, and causing a lot of early EDQUOT.

Fix it by applying the correct diff on top of current branch.

Fixes: 43b18595d660 ("btrfs: qgroup: Use separate meta reservation type for delalloc")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: delayed-inode: Remove wrong qgroup meta reservation calls
Qu Wenruo [Tue, 17 Apr 2018 08:52:45 +0000 (16:52 +0800)]
btrfs: delayed-inode: Remove wrong qgroup meta reservation calls

Commit 4f5427ccce5d ("btrfs: delayed-inode: Use new qgroup meta rsv for
delayed inode and item") merged into mainline was not latest version
submitted to the mail list in Dec 2017.

Which lacks the following fixes:

1) Remove btrfs_qgroup_convert_reserved_meta() call in
   btrfs_delayed_item_release_metadata()
2) Remove btrfs_qgroup_reserve_meta_prealloc() call in
   btrfs_delayed_inode_reserve_metadata()

Those fixes will resolve unexpected EDQUOT problems.

Fixes: 4f5427ccce5d ("btrfs: delayed-inode: Use new qgroup meta rsv for delayed inode and item")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Use independent and accurate per inode qgroup rsv
Qu Wenruo [Thu, 21 Dec 2017 05:42:04 +0000 (13:42 +0800)]
btrfs: qgroup: Use independent and accurate per inode qgroup rsv

Unlike reservation calculation used in inode rsv for metadata, qgroup
doesn't really need to care about things like csum size or extent usage
for the whole tree COW.

Qgroups care more about net change of the extent usage.
That's to say, if we're going to insert one file extent, it will mostly
find its place in COWed tree block, leaving no change in extent usage.
Or causing a leaf split, resulting in one new net extent and increasing
qgroup number by nodesize.
Or in an even more rare case, increase the tree level, increasing qgroup
number by 2 * nodesize.

So here instead of using the complicated calculation for extent
allocator, which cares more about accuracy and no error, qgroup doesn't
need that over-estimated reservation.

This patch will maintain 2 new members in btrfs_block_rsv structure for
qgroup, using much smaller calculation for qgroup rsv, reducing false
EDQUOT.

Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
6 years agobtrfs: qgroup: Commit transaction in advance to reduce early EDQUOT
Qu Wenruo [Fri, 22 Dec 2017 08:06:39 +0000 (16:06 +0800)]
btrfs: qgroup: Commit transaction in advance to reduce early EDQUOT

Unlike previous method that tries to commit transaction inside
qgroup_reserve(), this time we will try to commit transaction using
fs_info->transaction_kthread to avoid nested transaction and no need to
worry about locking context.

Since it's an asynchronous function call and we won't wait for
transaction commit, unlike previous method, we must call it before we
hit the qgroup limit.

So this patch will use the ratio and size of qgroup meta_pertrans
reservation as indicator to check if we should trigger a transaction
commit.  (meta_prealloc won't be cleaned in transaction committ, it's
useless anyway)

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Only check first key for committed tree blocks
Qu Wenruo [Thu, 12 Apr 2018 22:32:47 +0000 (06:32 +0800)]
btrfs: Only check first key for committed tree blocks

When looping btrfs/074 with many cpus (>= 8), it's possible to trigger
kernel warning due to first key verification:

[ 4239.523446] WARNING: CPU: 5 PID: 2381 at fs/btrfs/disk-io.c:460 btree_read_extent_buffer_pages+0x1ad/0x210
[ 4239.523830] Modules linked in:
[ 4239.524630] RIP: 0010:btree_read_extent_buffer_pages+0x1ad/0x210
[ 4239.527101] Call Trace:
[ 4239.527251]  read_tree_block+0x42/0x70
[ 4239.527434]  read_node_slot+0xd2/0x110
[ 4239.527632]  push_leaf_right+0xad/0x1b0
[ 4239.527809]  split_leaf+0x4ea/0x700
[ 4239.527988]  ? leaf_space_used+0xbc/0xe0
[ 4239.528192]  ? btrfs_set_lock_blocking_rw+0x99/0xb0
[ 4239.528416]  btrfs_search_slot+0x8cc/0xa40
[ 4239.528605]  btrfs_insert_empty_items+0x71/0xc0
[ 4239.528798]  __btrfs_run_delayed_refs+0xa98/0x1680
[ 4239.529013]  btrfs_run_delayed_refs+0x10b/0x1b0
[ 4239.529205]  btrfs_commit_transaction+0x33/0xaf0
[ 4239.529445]  ? start_transaction+0xa8/0x4f0
[ 4239.529630]  btrfs_alloc_data_chunk_ondemand+0x1b0/0x4e0
[ 4239.529833]  btrfs_check_data_free_space+0x54/0xa0
[ 4239.530045]  btrfs_delalloc_reserve_space+0x25/0x70
[ 4239.531907]  btrfs_direct_IO+0x233/0x3d0
[ 4239.532098]  generic_file_direct_write+0xcb/0x170
[ 4239.532296]  btrfs_file_write_iter+0x2bb/0x5f4
[ 4239.532491]  aio_write+0xe2/0x180
[ 4239.532669]  ? lock_acquire+0xac/0x1e0
[ 4239.532839]  ? __might_fault+0x3e/0x90
[ 4239.533032]  do_io_submit+0x594/0x860
[ 4239.533223]  ? do_io_submit+0x594/0x860
[ 4239.533398]  SyS_io_submit+0x10/0x20
[ 4239.533560]  ? SyS_io_submit+0x10/0x20
[ 4239.533729]  do_syscall_64+0x75/0x1d0
[ 4239.533979]  entry_SYSCALL_64_after_hwframe+0x42/0xb7
[ 4239.534182] RIP: 0033:0x7f8519741697

The problem here is, at btree_read_extent_buffer_pages() we don't have
acquired read/write lock on that extent buffer, only basic info like
level/bytenr is reliable.

So race condition leads to such false alert.

However in current call site, it's impossible to acquire proper lock
without race window.
To fix the problem, we only verify first key for committed tree blocks
(whose generation is no larger than fs_info->last_trans_committed), so
the content of such tree blocks will not change and there is no need to
get read/write lock.

Reported-by: Nikolay Borisov <nborisov@suse.com>
Fixes: 581c1760415c ("btrfs: Validate child tree block's level and first key")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: add SPDX header to Kconfig
David Sterba [Tue, 3 Apr 2018 17:23:33 +0000 (19:23 +0200)]
btrfs: add SPDX header to Kconfig

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: replace GPL boilerplate by SPDX -- sources
David Sterba [Tue, 3 Apr 2018 17:23:33 +0000 (19:23 +0200)]
btrfs: replace GPL boilerplate by SPDX -- sources

Remove GPL boilerplate text (long, short, one-line) and keep the rest,
ie. personal, company or original source copyright statements. Add the
SPDX header.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: replace GPL boilerplate by SPDX -- headers
David Sterba [Tue, 3 Apr 2018 17:16:55 +0000 (19:16 +0200)]
btrfs: replace GPL boilerplate by SPDX -- headers

Remove GPL boilerplate text (long, short, one-line) and keep the rest,
ie. personal, company or original source copyright statements. Add the
SPDX header.

Unify the include protection macros to match the file names.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix loss of prealloc extents past i_size after fsync log replay
Filipe Manana [Thu, 5 Apr 2018 21:55:12 +0000 (22:55 +0100)]
Btrfs: fix loss of prealloc extents past i_size after fsync log replay

Currently if we allocate extents beyond an inode's i_size (through the
fallocate system call) and then fsync the file, we log the extents but
after a power failure we replay them and then immediately drop them.
This behaviour happens since about 2009, commit c71bf099abdd ("Btrfs:
Avoid orphan inodes cleanup while replaying log"), because it marks
the inode as an orphan instead of dropping any extents beyond i_size
before replaying logged extents, so after the log replay, and while
the mount operation is still ongoing, we find the inode marked as an
orphan and then perform a truncation (drop extents beyond the inode's
i_size). Because the processing of orphan inodes is still done
right after replaying the log and before the mount operation finishes,
the intention of that commit does not make any sense (at least as
of today). However reverting that behaviour is not enough, because
we can not simply discard all extents beyond i_size and then replay
logged extents, because we risk dropping extents beyond i_size created
in past transactions, for example:

  add prealloc extent beyond i_size
  fsync - clears the flag BTRFS_INODE_NEEDS_FULL_SYNC from the inode
  transaction commit
  add another prealloc extent beyond i_size
  fsync - triggers the fast fsync path
  power failure

In that scenario, we would drop the first extent and then replay the
second one. To fix this just make sure that all prealloc extents
beyond i_size are logged, and if we find too many (which is far from
a common case), fallback to a full transaction commit (like we do when
logging regular extents in the fast fsync path).

Trivial reproducer:

 $ mkfs.btrfs -f /dev/sdb
 $ mount /dev/sdb /mnt
 $ xfs_io -f -c "pwrite -S 0xab 0 256K" /mnt/foo
 $ sync
 $ xfs_io -c "falloc -k 256K 1M" /mnt/foo
 $ xfs_io -c "fsync" /mnt/foo
 <power failure>

 # mount to replay log
 $ mount /dev/sdb /mnt
 # at this point the file only has one extent, at offset 0, size 256K

A test case for fstests follows soon, covering multiple scenarios that
involve adding prealloc extents with previous shrinking truncates and
without such truncates.

Fixes: c71bf099abdd ("Btrfs: Avoid orphan inodes cleanup while replaying log")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: clean up resources during umount after trans is aborted
Liu Bo [Fri, 30 Mar 2018 22:11:56 +0000 (06:11 +0800)]
Btrfs: clean up resources during umount after trans is aborted

Currently if some fatal errors occur, like all IO get -EIO, resources
would be cleaned up when
a) transaction is being committed or
b) BTRFS_FS_STATE_ERROR is set

However, in some rare cases, resources may be left alone after transaction
gets aborted and umount may run into some ASSERT(), e.g.
ASSERT(list_empty(&block_group->dirty_list));

For case a), in btrfs_commit_transaciton(), there're several places at the
beginning where we just call btrfs_end_transaction() without cleaning up
resources.  For case b), it is possible that the trans handle doesn't have
any dirty stuff, then only trans hanlde is marked as aborted while
BTRFS_FS_STATE_ERROR is not set, so resources remain in memory.

This makes btrfs also check BTRFS_FS_STATE_TRANS_ABORTED to make sure that
all resources won't stay in memory after umount.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Fix possible softlock on single core machines
Nikolay Borisov [Thu, 5 Apr 2018 07:40:15 +0000 (10:40 +0300)]
btrfs: Fix possible softlock on single core machines

do_chunk_alloc implements a loop checking whether there is a pending
chunk allocation and if so causes the caller do loop. Generally this
loop is executed only once, however testing with btrfs/072 on a single
core vm machines uncovered an extreme case where the system could loop
indefinitely. This is due to a missing cond_resched when loop which
doesn't give a chance to the previous chunk allocator finish its job.

The fix is to simply add the missing cond_resched.

Fixes: 6d74119f1a3e ("Btrfs: avoid taking the chunk_mutex in do_chunk_alloc")
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: bail out on error during replay_dir_deletes
Liu Bo [Mon, 2 Apr 2018 17:59:48 +0000 (01:59 +0800)]
Btrfs: bail out on error during replay_dir_deletes

If errors were returned by btrfs_next_leaf(), replay_dir_deletes needs
to bail out, otherwise @ret would be forced to be 0 after 'break;' and
the caller won't be aware of it.

Fixes: e02119d5a7b4 ("Btrfs: Add a write ahead tree log to optimize synchronous operations")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix NULL pointer dereference in log_dir_items
Liu Bo [Mon, 2 Apr 2018 17:59:47 +0000 (01:59 +0800)]
Btrfs: fix NULL pointer dereference in log_dir_items

0, 1 and <0 can be returned by btrfs_next_leaf(), and when <0 is
returned, path->nodes[0] could be NULL, log_dir_items lacks such a
check for <0 and we may run into a null pointer dereference panic.

Fixes: e02119d5a7b4 ("Btrfs: Add a write ahead tree log to optimize synchronous operations")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: lift errors from add_extent_changeset to the callers
David Sterba [Thu, 1 Mar 2018 16:56:34 +0000 (17:56 +0100)]
btrfs: lift errors from add_extent_changeset to the callers

The missing error handling in add_extent_changeset was hidden, so make
it at least visible in the callers.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: print error messages when failing to read trees
Liu Bo [Wed, 28 Mar 2018 22:11:45 +0000 (06:11 +0800)]
Btrfs: print error messages when failing to read trees

When mount fails to read trees like fs tree, checksum tree, extent
tree, etc, there is not enough information about where went wrong.

With this, messages like

"BTRFS warning (device sdf): failed to read root (objectid=7): -5"

would help us a bit.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: user proper type for btrfs_mask_flags flags
David Sterba [Mon, 26 Mar 2018 16:29:41 +0000 (18:29 +0200)]
btrfs: user proper type for btrfs_mask_flags flags

All users pass a local unsigned int and not the __uXX types that are
supposed to be used for userspace interfaces.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: split dev-replace locking helpers for read and write
David Sterba [Sat, 24 Mar 2018 01:11:38 +0000 (02:11 +0100)]
btrfs: split dev-replace locking helpers for read and write

The current calls are unclear in what way btrfs_dev_replace_lock takes
the locks, so drop the argument, split the helpers and use similar
naming as for read and write locks.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove stale comments about fs_mutex
David Sterba [Tue, 27 Mar 2018 18:08:28 +0000 (20:08 +0200)]
btrfs: remove stale comments about fs_mutex

The fs_mutex has been killed in 2008, a213501153fd66e2 ("Btrfs: Replace
the big fs_mutex with a collection of other locks"), still remembered in
some comments.

We don't have any extra needs for locking in the ACL handlers.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use RCU in btrfs_show_devname for device list traversal
David Sterba [Fri, 16 Mar 2018 02:27:02 +0000 (03:27 +0100)]
btrfs: use RCU in btrfs_show_devname for device list traversal

The show_devname callback is used to print device name in
/proc/self/mounts, we need to traverse the device list consistently and
read the name that's copied to a seq buffer so we don't need further
locking.

If the first device is being deleted at the same time, the RCU will
allow us to read the device name, though it will become stale right
after the RCU protection ends. This is unavoidable and the user can
expect that the device will disappear from the filesystem's list at some
point.

The device_list_mutex was pretty heavy as it is used eg. for writing
superblock and a few other IO related contexts. This can stall any
application that reads the proc file for no reason.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: update barrier in should_cow_block
David Sterba [Fri, 16 Mar 2018 01:39:40 +0000 (02:39 +0100)]
btrfs: update barrier in should_cow_block

Once there was a simple int force_cow that was used with the plain
barriers, and then converted to a bit, so we should use the appropriate
barrier helper.

Other variables in the complex if condition do not depend on a barrier,
so we should be fine in case the atomic barrier becomes a no-op.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use lockdep_assert_held for mutexes
David Sterba [Fri, 16 Mar 2018 01:21:22 +0000 (02:21 +0100)]
btrfs: use lockdep_assert_held for mutexes

Using lockdep_assert_held is preferred, replace mutex_is_locked.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use lockdep_assert_held for spinlocks
David Sterba [Fri, 16 Mar 2018 01:21:22 +0000 (02:21 +0100)]
btrfs: use lockdep_assert_held for spinlocks

Using lockdep_assert_held is preferred, replace assert_spin_locked.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Validate child tree block's level and first key
Qu Wenruo [Thu, 29 Mar 2018 01:08:11 +0000 (09:08 +0800)]
btrfs: Validate child tree block's level and first key

We have several reports about node pointer points to incorrect child
tree blocks, which could have even wrong owner and level but still with
valid generation and checksum.

Although btrfs check could handle it and print error message like:
leaf parent key incorrect 60670574592

Kernel doesn't have enough check on this type of corruption correctly.
At least add such check to read_tree_block() and btrfs_read_buffer(),
where we need two new parameters @level and @first_key to verify the
child tree block.

The new @level check is mandatory and all call sites are already
modified to extract expected level from its call chain.

While @first_key is optional, the following call sites are skipping such
check:
1) Root node/leaf
   As ROOT_ITEM doesn't contain the first key, skip @first_key check.
2) Direct backref
   Only parent bytenr and level is known and we need to resolve the key
   all by ourselves, skip @first_key check.

Another note of this verification is, it needs extra info from nodeptr
or ROOT_ITEM, so it can't fit into current tree-checker framework, which
is limited to node/leaf boundary.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: tests/qgroup: Fix wrong tree backref level
Qu Wenruo [Tue, 27 Mar 2018 12:44:18 +0000 (20:44 +0800)]
btrfs: tests/qgroup: Fix wrong tree backref level

The extent tree of the test fs is like the following:

 BTRFS info (device (null)): leaf 16327509003777336587 total ptrs 1 free space 3919
  item 0 key (4096 168 4096) itemoff 3944 itemsize 51
          extent refs 1 gen 1 flags 2
          tree block key (68719476736 0 0) level 1
                                           ^^^^^^^
          ref#0: tree block backref root 5

And it's using an empty tree for fs tree, so there is no way that its
level can be 1.

For REAL (created by mkfs) fs tree backref with no skinny metadata, the
result should look like:

 item 3 key (30408704 EXTENT_ITEM 4096) itemoff 3845 itemsize 51
         refs 1 gen 4 flags TREE_BLOCK
         tree block key (256 INODE_ITEM 0) level 0
                                           ^^^^^^^
         tree block backref root 5

Fix the level to 0, so it won't break later tree level checker.

Fixes: faa2dbf004e8 ("Btrfs: add sanity tests for new qgroup accounting code")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix copy_items() return value when logging an inode
Filipe Manana [Mon, 26 Mar 2018 22:59:12 +0000 (23:59 +0100)]
Btrfs: fix copy_items() return value when logging an inode

When logging an inode, at tree-log.c:copy_items(), if we call
btrfs_next_leaf() at the loop which checks for the need to log holes, we
need to make sure copy_items() returns the value 1 to its caller and
not 0 (on success). This is because the path the caller passed was
released and is now different from what is was before, and the caller
expects a return value of 0 to mean both success and that the path
has not changed, while a return value of 1 means both success and
signals the caller that it can not reuse the path, it has to perform
another tree search.

Even though this is a case that should not be triggered on normal
circumstances or very rare at least, its consequences can be very
unpredictable (especially when replaying a log tree).

Fixes: 16e7549f045d ("Btrfs: incompatible format change to remove hole extents")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: fix fsync after hole punching when using no-holes feature
Filipe Manana [Mon, 26 Mar 2018 22:59:00 +0000 (23:59 +0100)]
Btrfs: fix fsync after hole punching when using no-holes feature

When we have the no-holes mode enabled and fsync a file after punching a
hole in it, we can end up not logging the whole hole range in the log tree.
This happens if the file has extent items that span more than one leaf and
we punch a hole that covers a range that starts in a leaf but does not go
beyond the offset of the first extent in the next leaf.

Example:

  $ mkfs.btrfs -f -O no-holes -n 65536 /dev/sdb
  $ mount /dev/sdb /mnt
  $ for ((i = 0; i <= 831; i++)); do
offset=$((i * 2 * 256 * 1024))
xfs_io -f -c "pwrite -S 0xab -b 256K $offset 256K" \
/mnt/foobar >/dev/null
    done
  $ sync

  # We now have 2 leafs in our filesystem fs tree, the first leaf has an
  # item corresponding the extent at file offset 216530944 and the second
  # leaf has a first item corresponding to the extent at offset 217055232.
  # Now we punch a hole that partially covers the range of the extent at
  # offset 216530944 but does go beyond the offset 217055232.

  $ xfs_io -c "fpunch $((216530944 + 128 * 1024 - 4000)) 256K" /mnt/foobar
  $ xfs_io -c "fsync" /mnt/foobar

  <power fail>

  # mount to replay the log
  $ mount /dev/sdb /mnt

  # Before this patch, only the subrange [216658016216662016[ (length of
  # 4000 bytes) was logged, leaving an incorrect file layout after log
  # replay.

Fix this by checking if there is a hole between the last extent item that
we processed and the first extent item in the next leaf, and if there is
one, log an explicit hole extent item.

Fixes: 16e7549f045d ("Btrfs: incompatible format change to remove hole extents")
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: use helper to set ulist aux from a qgroup
David Sterba [Tue, 27 Mar 2018 17:04:50 +0000 (19:04 +0200)]
btrfs: use helper to set ulist aux from a qgroup

We have a nice helper to do proper casting of a qgroup to a ulist aux
value. And several places that could make use of it.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoRevert "btrfs: qgroups: Retry after commit on getting EDQUOT"
Qu Wenruo [Tue, 12 Dec 2017 07:34:36 +0000 (15:34 +0800)]
Revert "btrfs: qgroups: Retry after commit on getting EDQUOT"

This reverts commit 48a89bc4f2ceab87bc858a8eb189636b09c846a7.

The idea to commit transaction and free some space after hitting qgroup
limit is good, although the problem is it can easily cause deadlocks.

One deadlock example is caused by trying to flush data while still
holding it:

Call Trace:
 __schedule+0x49d/0x10f0
 schedule+0xc6/0x290
 schedule_timeout+0x187/0x1c0
 wait_for_completion+0x204/0x3a0
 btrfs_wait_ordered_extents+0xa40/0xaf0 [btrfs]
 qgroup_reserve+0x913/0xa10 [btrfs]
 btrfs_qgroup_reserve_data+0x3ef/0x580 [btrfs]
 btrfs_check_data_free_space+0x96/0xd0 [btrfs]
 __btrfs_buffered_write+0x3ac/0xd40 [btrfs]
 btrfs_file_write_iter+0x62a/0xba0 [btrfs]
 __vfs_write+0x320/0x430
 vfs_write+0x107/0x270
 SyS_write+0xbf/0x150
 do_syscall_64+0x1b0/0x3d0
 entry_SYSCALL64_slow_path+0x25/0x25

Another can be caused by trying to commit one transaction while nesting
with trans handle held by ourselves:

btrfs_start_transaction()
|- btrfs_qgroup_reserve_meta_pertrans()
   |- qgroup_reserve()
      |- btrfs_join_transaction()
      |- btrfs_commit_transaction()

The retry is causing more problems than exppected when limit is enabled.
At least a graceful EDQUOT is way better than deadlock.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Update trace events for metadata reservation
Qu Wenruo [Tue, 12 Dec 2017 07:34:35 +0000 (15:34 +0800)]
btrfs: qgroup: Update trace events for metadata reservation

Now trace_qgroup_meta_reserve() will have extra type parameter.

And introduce two new trace events:

1) trace_qgroup_meta_free_all_pertrans()
   For btrfs_qgroup_free_meta_all_pertrans()

2) trace_qgroup_meta_convert()
   For btrfs_qgroup_convert_reserved_meta()

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Use root::qgroup_meta_rsv_* to record qgroup meta reserved space
Qu Wenruo [Tue, 12 Dec 2017 07:34:34 +0000 (15:34 +0800)]
btrfs: qgroup: Use root::qgroup_meta_rsv_* to record qgroup meta reserved space

For quota disabled->enable case, it's possible that at reservation time
quota was not enabled so no bytes were really reserved, while at release
time, quota was enabled so we will try to release some bytes we didn't
really own.

Such situation can cause metadata reserveation underflow, for both types,
also less possible for per-trans type since quota enable will commit
transaction.

To address this, record qgroup meta reserved bytes into
root::qgroup_meta_rsv_pertrans and ::prealloc.
So at releasing time we won't free any bytes we didn't reserve.

For DATA, it's already handled by io_tree, so nothing needs to be done
there.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: delayed-inode: Use new qgroup meta rsv for delayed inode and item
Qu Wenruo [Tue, 12 Dec 2017 07:34:33 +0000 (15:34 +0800)]
btrfs: delayed-inode: Use new qgroup meta rsv for delayed inode and item

Quite similar for delalloc, some modification to delayed-inode and
delayed-item reservation.  Also needs extra parameter for release case
to distinguish normal release and error release.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Use separate meta reservation type for delalloc
Qu Wenruo [Tue, 12 Dec 2017 07:34:32 +0000 (15:34 +0800)]
btrfs: qgroup: Use separate meta reservation type for delalloc

Before this patch, btrfs qgroup is mixing per-transcation meta rsv with
preallocated meta rsv, making it quite easy to underflow qgroup meta
reservation.

Since we have the new qgroup meta rsv types, apply it to delalloc
reservation.

Now for delalloc, most of its reserved space will use META_PREALLOC qgroup
rsv type.

And for callers reducing outstanding extent like btrfs_finish_ordered_io(),
they will convert corresponding META_PREALLOC reservation to
META_PERTRANS.

This is mainly due to the fact that current qgroup numbers will only be
updated in btrfs_commit_transaction(), that's to say if we don't keep
such placeholder reservation, we can exceed qgroup limitation.

And for callers freeing outstanding extent in error handler, we will
just free META_PREALLOC bytes.

This behavior makes callers of btrfs_qgroup_release_meta() or
btrfs_qgroup_convert_meta() to be aware of which type they are.
So in this patch, btrfs_delalloc_release_metadata() and its callers get
an extra parameter to info qgroup to do correct meta convert/release.

The good news is, even we use the wrong type (convert or free), it won't
cause obvious bug, as prealloc type is always in good shape, and the
type only affects how per-trans meta is increased or not.

So the worst case will be at most metadata limitation can be sometimes
exceeded (no convert at all) or metadata limitation is reached too soon
(no free at all).

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Introduce function to convert META_PREALLOC into META_PERTRANS
Qu Wenruo [Tue, 12 Dec 2017 07:34:31 +0000 (15:34 +0800)]
btrfs: qgroup: Introduce function to convert META_PREALLOC into META_PERTRANS

For meta_prealloc reservation users, after btrfs_join_transaction()
caller will modify tree so part (or even all) meta_prealloc reservation
should be converted to meta_pertrans until transaction commit time.

This patch introduces a new function,
btrfs_qgroup_convert_reserved_meta() to do this for META_PREALLOC
reservation user.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Don't use root->qgroup_meta_rsv for qgroup
Qu Wenruo [Tue, 12 Dec 2017 07:34:30 +0000 (15:34 +0800)]
btrfs: qgroup: Don't use root->qgroup_meta_rsv for qgroup

Since qgroup has seperate metadata reservation types now, we can
completely get rid of the old root->qgroup_meta_rsv, which mostly acts
as current META_PERTRANS reservation type.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Split meta rsv type into meta_prealloc and meta_pertrans
Qu Wenruo [Tue, 12 Dec 2017 07:34:29 +0000 (15:34 +0800)]
btrfs: qgroup: Split meta rsv type into meta_prealloc and meta_pertrans

Btrfs uses 2 different methods to reseve metadata qgroup space.

1) Reserve at btrfs_start_transaction() time
   This is quite straightforward, caller will use the trans handler
   allocated to modify b-trees.

   In this case, reserved metadata should be kept until qgroup numbers
   are updated.

2) Reserve by using block_rsv first, and later btrfs_join_transaction()
   This is more complicated, caller will reserve space using block_rsv
   first, and then later call btrfs_join_transaction() to get a trans
   handle.

   In this case, before we modify trees, the reserved space can be
   modified on demand, and after btrfs_join_transaction(), such reserved
   space should also be kept until qgroup numbers are updated.

Since these two types behave differently, split the original "META"
reservation type into 2 sub-types:

  META_PERTRANS:
    For above case 1)

  META_PREALLOC:
    For reservations that happened before btrfs_join_transaction() of
    case 2)

NOTE: This patch will only convert existing qgroup meta reservation
callers according to its situation, not ensuring all callers are at
correct timing.
Such fix will be added in later patches.

Signed-off-by: Qu Wenruo <wqu@suse.com>
[ update comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Cleanup the remaining old reservation counters
Qu Wenruo [Tue, 12 Dec 2017 07:34:28 +0000 (15:34 +0800)]
btrfs: qgroup: Cleanup the remaining old reservation counters

So qgroup is switched to new separate types reservation system.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Update trace events to use new separate rsv types
Qu Wenruo [Tue, 12 Dec 2017 07:34:27 +0000 (15:34 +0800)]
btrfs: qgroup: Update trace events to use new separate rsv types

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Fix wrong qgroup reservation update for relationship modification
Qu Wenruo [Tue, 12 Dec 2017 07:34:26 +0000 (15:34 +0800)]
btrfs: qgroup: Fix wrong qgroup reservation update for relationship modification

When modifying qgroup relationship, for qgroup which only owns exclusive
extents, we will go through quick update path.

In this path, we will add/subtract exclusive and reference number for
parent qgroup, since the source (child) qgroup only has exclusive
extents, destination (parent) qgroup will also own or lose those extents
exclusively.

The same should be the same for reservation, since later reservation
adding/releasing will also affect parent qgroup, without the reservation
carried from child, parent will underflow reservation or have dead
reservation which will never be freed.

However original code doesn't do the same thing for reservation.
It handles qgroup reservation quite differently:

It removes qgroup reservation, as it's allocating space from the
reserved qgroup for relationship adding.
But does nothing for qgroup reservation if we're removing a qgroup
relationship.

According to the original code, it looks just like because we're adding
qgroup->rfer, the code assumes we're writing new data, so it's follows
the normal write routine, by reducing qgroup->reserved and adding
qgroup->rfer/excl.

This old behavior is wrong, and should be fixed to follow the same
excl/rfer behavior.

Just fix it by using the correct behavior described above.

Fixes: 31193213f1f9 ("Btrfs: qgroup: Introduce a may_use to account space_info->bytes_may_use.")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Make qgroup_reserve and its callers to use separate reservation type
Qu Wenruo [Tue, 12 Dec 2017 07:34:25 +0000 (15:34 +0800)]
btrfs: qgroup: Make qgroup_reserve and its callers to use separate reservation type

Since most callers of qgroup_reserve() are already defined by type,
converting qgroup_reserve() is quite an easy work.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Introduce helpers to update and access new qgroup rsv
Qu Wenruo [Tue, 12 Dec 2017 07:34:24 +0000 (15:34 +0800)]
btrfs: qgroup: Introduce helpers to update and access new qgroup rsv

Introduce helpers to:

1) Get total reserved space
   For limit calculation
2) Add/release reserved space for given type
   With underflow detection and warning
3) Add/release reserved space according to child qgroup

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: qgroup: Skeleton to support separate qgroup reservation type
Qu Wenruo [Tue, 12 Dec 2017 07:34:23 +0000 (15:34 +0800)]
btrfs: qgroup: Skeleton to support separate qgroup reservation type

Instead of single qgroup->reserved, use a new structure btrfs_qgroup_rsv
to store different types of reservation.

This patch only updates the header needed to compile.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: delete dead code in btrfs_orphan_add()
Omar Sandoval [Thu, 25 Jan 2018 23:56:17 +0000 (15:56 -0800)]
Btrfs: delete dead code in btrfs_orphan_add()

btrfs_orphan_add() has had this case commented out since it was first
introduced in commit d68fc57b7e32 ("Btrfs: Metadata reservation for
orphan inodes"). Most of the orphan cleanup code has been rewritten
since then, so it's safe to say that this code isn't needed.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
[ switch to bool ]
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: ctree.h: Fix wrong comment position about csum size
Misono, Tomohiro [Tue, 20 Mar 2018 06:47:06 +0000 (15:47 +0900)]
btrfs: ctree.h: Fix wrong comment position about csum size

Signed-off-by: Tomohiro Misono <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: defer adding raid type kobject until after chunk relocation
Jeff Mahoney [Tue, 20 Mar 2018 19:25:26 +0000 (15:25 -0400)]
btrfs: defer adding raid type kobject until after chunk relocation

Any time the first block group of a new type is created, we add a new
kobject to sysfs to hold the attributes for that type.  Kobject-internal
allocations always use GFP_KERNEL, making them prone to fs-reclaim races.
While it appears as if this can occur any time a block group is created,
the only times the first block group of a new type can be created in
memory is at mount and when we create the first new block group during
raid conversion.

This patch adds a new list to track pending kobject additions and then
handles them after we do chunk relocation.  Between relocating the
target chunk (or forcing allocation of a new chunk in the case of data)
and removing the old chunk, we're in a safe place for fs-reclaim to
occur.  We're holding the volume mutex, which is already held across
page faults, and the delete_unused_bgs_mutex, which will only stall
the cleaner thread.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove dead create_space_info calls
Jeff Mahoney [Tue, 20 Mar 2018 19:25:25 +0000 (15:25 -0400)]
btrfs: remove dead create_space_info calls

Since commit 2be12ef79 (btrfs: Separate space_info create/update), we've
separated out the creation and updating of the space info structures.
That commit was a straightforward refactoring of the two parts of
update_space_info, but we can go a step further.  Since commits
c59021f84 (Btrfs: fix OOPS of empty filesystem after balance) and
b742bb82f (Btrfs: Link block groups of different raid types), we know
that the space_info structures will be created at mount and there will
only ever be, at most, three of them.

This patch cleans out the create_space_info calls after __find_space_info
returns NULL since __find_space_info *can't* return NULL.

The initial cause for reviewing this was the kobject_add calls from
create_space_info occuring in sites where fs-reclaim wasn't allowed.  Now
we are certain they occur only early in the mount process and are safe.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: replace: cache rbio when rebuild data on missing device
Liu Bo [Thu, 22 Mar 2018 01:20:11 +0000 (09:20 +0800)]
Btrfs: replace: cache rbio when rebuild data on missing device

Rebuild on missing device is as same as recover, after it's done, rbio
has data which is consistent with on-disk data, so it can be cached to
avoid further reads.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: fix lockdep splat in btrfs_alloc_subvolume_writers
Jeff Mahoney [Fri, 16 Mar 2018 18:36:27 +0000 (14:36 -0400)]
btrfs: fix lockdep splat in btrfs_alloc_subvolume_writers

While running btrfs/011, I hit the following lockdep splat.

This is the important bit:
   pcpu_alloc+0x1ac/0x5e0
   __percpu_counter_init+0x4e/0xb0
   btrfs_init_fs_root+0x99/0x1c0 [btrfs]
   btrfs_get_fs_root.part.54+0x5b/0x150 [btrfs]
   resolve_indirect_refs+0x130/0x830 [btrfs]
   find_parent_nodes+0x69e/0xff0 [btrfs]
   btrfs_find_all_roots_safe+0xa0/0x110 [btrfs]
   btrfs_find_all_roots+0x50/0x70 [btrfs]
   btrfs_qgroup_prepare_account_extents+0x53/0x90 [btrfs]
   btrfs_commit_transaction+0x3ce/0x9b0 [btrfs]

The percpu_counter_init call in btrfs_alloc_subvolume_writers
uses GFP_KERNEL, which we can't do during transaction commit.

This switches it to GFP_NOFS.

========================================================
WARNING: possible irq lock inversion dependency detected
4.12.14-kvmsmall #8 Tainted: G        W
--------------------------------------------------------
kswapd0/50 just changed the state of lock:
 (&delayed_node->mutex){+.+.-.}, at: [<ffffffffc06994fa>] __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
but this lock took another, RECLAIM_FS-unsafe lock in the past:
 (pcpu_alloc_mutex){+.+.+.}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
Chain exists of:
  &delayed_node->mutex --> &found->groups_sem --> pcpu_alloc_mutex

 Possible interrupt unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(pcpu_alloc_mutex);
                               local_irq_disable();
                               lock(&delayed_node->mutex);
                               lock(&found->groups_sem);
  <Interrupt>
    lock(&delayed_node->mutex);

 *** DEADLOCK ***

2 locks held by kswapd0/50:
 #0:  (shrinker_rwsem){++++..}, at: [<ffffffff811dc11f>] shrink_slab+0x7f/0x5b0
 #1:  (&type->s_umount_key#30){+++++.}, at: [<ffffffff8126dec6>] trylock_super+0x16/0x50

the shortest dependencies between 2nd lock and 1st lock:
   -> (pcpu_alloc_mutex){+.+.+.} ops: 4904 {
      HARDIRQ-ON-W at:
                          __mutex_lock+0x4e/0x8c0
                          pcpu_alloc+0x1ac/0x5e0
                          alloc_kmem_cache_cpus.isra.70+0x25/0xa0
                          __do_tune_cpucache+0x2c/0x220
                          do_tune_cpucache+0x26/0xc0
                          enable_cpucache+0x6d/0xf0
                          kmem_cache_init_late+0x42/0x75
                          start_kernel+0x343/0x4cb
                          x86_64_start_kernel+0x127/0x134
                          secondary_startup_64+0xa5/0xb0
      SOFTIRQ-ON-W at:
                          __mutex_lock+0x4e/0x8c0
                          pcpu_alloc+0x1ac/0x5e0
                          alloc_kmem_cache_cpus.isra.70+0x25/0xa0
                          __do_tune_cpucache+0x2c/0x220
                          do_tune_cpucache+0x26/0xc0
                          enable_cpucache+0x6d/0xf0
                          kmem_cache_init_late+0x42/0x75
                          start_kernel+0x343/0x4cb
                          x86_64_start_kernel+0x127/0x134
                          secondary_startup_64+0xa5/0xb0
      RECLAIM_FS-ON-W at:
                             __kmalloc+0x47/0x310
                             pcpu_extend_area_map+0x2b/0xc0
                             pcpu_alloc+0x3ec/0x5e0
                             alloc_kmem_cache_cpus.isra.70+0x25/0xa0
                             __do_tune_cpucache+0x2c/0x220
                             do_tune_cpucache+0x26/0xc0
                             enable_cpucache+0x6d/0xf0
                             __kmem_cache_create+0x1bf/0x390
                             create_cache+0xba/0x1b0
                             kmem_cache_create+0x1f8/0x2b0
                             ksm_init+0x6f/0x19d
                             do_one_initcall+0x50/0x1b0
                             kernel_init_freeable+0x201/0x289
                             kernel_init+0xa/0x100
                             ret_from_fork+0x3a/0x50
      INITIAL USE at:
                         __mutex_lock+0x4e/0x8c0
                         pcpu_alloc+0x1ac/0x5e0
                         alloc_kmem_cache_cpus.isra.70+0x25/0xa0
                         setup_cpu_cache+0x2f/0x1f0
                         __kmem_cache_create+0x1bf/0x390
                         create_boot_cache+0x8b/0xb1
                         kmem_cache_init+0xa1/0x19e
                         start_kernel+0x270/0x4cb
                         x86_64_start_kernel+0x127/0x134
                         secondary_startup_64+0xa5/0xb0
    }
    ... key      at: [<ffffffff821d8e70>] pcpu_alloc_mutex+0x70/0xa0
    ... acquired at:
   pcpu_alloc+0x1ac/0x5e0
   __percpu_counter_init+0x4e/0xb0
   btrfs_init_fs_root+0x99/0x1c0 [btrfs]
   btrfs_get_fs_root.part.54+0x5b/0x150 [btrfs]
   resolve_indirect_refs+0x130/0x830 [btrfs]
   find_parent_nodes+0x69e/0xff0 [btrfs]
   btrfs_find_all_roots_safe+0xa0/0x110 [btrfs]
   btrfs_find_all_roots+0x50/0x70 [btrfs]
   btrfs_qgroup_prepare_account_extents+0x53/0x90 [btrfs]
   btrfs_commit_transaction+0x3ce/0x9b0 [btrfs]
   transaction_kthread+0x176/0x1b0 [btrfs]
   kthread+0x102/0x140
   ret_from_fork+0x3a/0x50

  -> (&fs_info->commit_root_sem){++++..} ops: 1566382 {
     HARDIRQ-ON-W at:
                        down_write+0x3e/0xa0
                        cache_block_group+0x287/0x420 [btrfs]
                        find_free_extent+0x106c/0x12d0 [btrfs]
                        btrfs_reserve_extent+0xd8/0x170 [btrfs]
                        cow_file_range.isra.66+0x133/0x470 [btrfs]
                        run_delalloc_range+0x121/0x410 [btrfs]
                        writepage_delalloc.isra.50+0xfe/0x180 [btrfs]
                        __extent_writepage+0x19a/0x360 [btrfs]
                        extent_write_cache_pages.constprop.56+0x249/0x3e0 [btrfs]
                        extent_writepages+0x4d/0x60 [btrfs]
                        do_writepages+0x1a/0x70
                        __filemap_fdatawrite_range+0xa7/0xe0
                        btrfs_rename+0x5ee/0xdb0 [btrfs]
                        vfs_rename+0x52a/0x7e0
                        SyS_rename+0x351/0x3b0
                        do_syscall_64+0x79/0x1e0
                        entry_SYSCALL_64_after_hwframe+0x42/0xb7
     HARDIRQ-ON-R at:
                        down_read+0x35/0x90
                        caching_thread+0x57/0x560 [btrfs]
                        normal_work_helper+0x1c0/0x5e0 [btrfs]
                        process_one_work+0x1e0/0x5c0
                        worker_thread+0x44/0x390
                        kthread+0x102/0x140
                        ret_from_fork+0x3a/0x50
     SOFTIRQ-ON-W at:
                        down_write+0x3e/0xa0
                        cache_block_group+0x287/0x420 [btrfs]
                        find_free_extent+0x106c/0x12d0 [btrfs]
                        btrfs_reserve_extent+0xd8/0x170 [btrfs]
                        cow_file_range.isra.66+0x133/0x470 [btrfs]
                        run_delalloc_range+0x121/0x410 [btrfs]
                        writepage_delalloc.isra.50+0xfe/0x180 [btrfs]
                        __extent_writepage+0x19a/0x360 [btrfs]
                        extent_write_cache_pages.constprop.56+0x249/0x3e0 [btrfs]
                        extent_writepages+0x4d/0x60 [btrfs]
                        do_writepages+0x1a/0x70
                        __filemap_fdatawrite_range+0xa7/0xe0
                        btrfs_rename+0x5ee/0xdb0 [btrfs]
                        vfs_rename+0x52a/0x7e0
                        SyS_rename+0x351/0x3b0
                        do_syscall_64+0x79/0x1e0
                        entry_SYSCALL_64_after_hwframe+0x42/0xb7
     SOFTIRQ-ON-R at:
                        down_read+0x35/0x90
                        caching_thread+0x57/0x560 [btrfs]
                        normal_work_helper+0x1c0/0x5e0 [btrfs]
                        process_one_work+0x1e0/0x5c0
                        worker_thread+0x44/0x390
                        kthread+0x102/0x140
                        ret_from_fork+0x3a/0x50
     INITIAL USE at:
                       down_write+0x3e/0xa0
                       cache_block_group+0x287/0x420 [btrfs]
                       find_free_extent+0x106c/0x12d0 [btrfs]
                       btrfs_reserve_extent+0xd8/0x170 [btrfs]
                       cow_file_range.isra.66+0x133/0x470 [btrfs]
                       run_delalloc_range+0x121/0x410 [btrfs]
                       writepage_delalloc.isra.50+0xfe/0x180 [btrfs]
                       __extent_writepage+0x19a/0x360 [btrfs]
                       extent_write_cache_pages.constprop.56+0x249/0x3e0 [btrfs]
                       extent_writepages+0x4d/0x60 [btrfs]
                       do_writepages+0x1a/0x70
                       __filemap_fdatawrite_range+0xa7/0xe0
                       btrfs_rename+0x5ee/0xdb0 [btrfs]
                       vfs_rename+0x52a/0x7e0
                       SyS_rename+0x351/0x3b0
                       do_syscall_64+0x79/0x1e0
                       entry_SYSCALL_64_after_hwframe+0x42/0xb7
   }
   ... key      at: [<ffffffffc0729578>] __key.61970+0x0/0xfffffffffff9aa88 [btrfs]
   ... acquired at:
   cache_block_group+0x287/0x420 [btrfs]
   find_free_extent+0x106c/0x12d0 [btrfs]
   btrfs_reserve_extent+0xd8/0x170 [btrfs]
   btrfs_alloc_tree_block+0x12f/0x4c0 [btrfs]
   btrfs_create_tree+0xbb/0x2a0 [btrfs]
   btrfs_create_uuid_tree+0x37/0x140 [btrfs]
   open_ctree+0x23c0/0x2660 [btrfs]
   btrfs_mount+0xd36/0xf90 [btrfs]
   mount_fs+0x3a/0x160
   vfs_kern_mount+0x66/0x150
   btrfs_mount+0x18c/0xf90 [btrfs]
   mount_fs+0x3a/0x160
   vfs_kern_mount+0x66/0x150
   do_mount+0x1c1/0xcc0
   SyS_mount+0x7e/0xd0
   do_syscall_64+0x79/0x1e0
   entry_SYSCALL_64_after_hwframe+0x42/0xb7

 -> (&found->groups_sem){++++..} ops: 2134587 {
    HARDIRQ-ON-W at:
                      down_write+0x3e/0xa0
                      __link_block_group+0x34/0x130 [btrfs]
                      btrfs_read_block_groups+0x33d/0x7b0 [btrfs]
                      open_ctree+0x2054/0x2660 [btrfs]
                      btrfs_mount+0xd36/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      btrfs_mount+0x18c/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      do_mount+0x1c1/0xcc0
                      SyS_mount+0x7e/0xd0
                      do_syscall_64+0x79/0x1e0
                      entry_SYSCALL_64_after_hwframe+0x42/0xb7
    HARDIRQ-ON-R at:
                      down_read+0x35/0x90
                      btrfs_calc_num_tolerated_disk_barrier_failures+0x113/0x1f0 [btrfs]
                      open_ctree+0x207b/0x2660 [btrfs]
                      btrfs_mount+0xd36/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      btrfs_mount+0x18c/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      do_mount+0x1c1/0xcc0
                      SyS_mount+0x7e/0xd0
                      do_syscall_64+0x79/0x1e0
                      entry_SYSCALL_64_after_hwframe+0x42/0xb7
    SOFTIRQ-ON-W at:
                      down_write+0x3e/0xa0
                      __link_block_group+0x34/0x130 [btrfs]
                      btrfs_read_block_groups+0x33d/0x7b0 [btrfs]
                      open_ctree+0x2054/0x2660 [btrfs]
                      btrfs_mount+0xd36/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      btrfs_mount+0x18c/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      do_mount+0x1c1/0xcc0
                      SyS_mount+0x7e/0xd0
                      do_syscall_64+0x79/0x1e0
                      entry_SYSCALL_64_after_hwframe+0x42/0xb7
    SOFTIRQ-ON-R at:
                      down_read+0x35/0x90
                      btrfs_calc_num_tolerated_disk_barrier_failures+0x113/0x1f0 [btrfs]
                      open_ctree+0x207b/0x2660 [btrfs]
                      btrfs_mount+0xd36/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      btrfs_mount+0x18c/0xf90 [btrfs]
                      mount_fs+0x3a/0x160
                      vfs_kern_mount+0x66/0x150
                      do_mount+0x1c1/0xcc0
                      SyS_mount+0x7e/0xd0
                      do_syscall_64+0x79/0x1e0
                      entry_SYSCALL_64_after_hwframe+0x42/0xb7
    INITIAL USE at:
                     down_write+0x3e/0xa0
                     __link_block_group+0x34/0x130 [btrfs]
                     btrfs_read_block_groups+0x33d/0x7b0 [btrfs]
                     open_ctree+0x2054/0x2660 [btrfs]
                     btrfs_mount+0xd36/0xf90 [btrfs]
                     mount_fs+0x3a/0x160
                     vfs_kern_mount+0x66/0x150
                     btrfs_mount+0x18c/0xf90 [btrfs]
                     mount_fs+0x3a/0x160
                     vfs_kern_mount+0x66/0x150
                     do_mount+0x1c1/0xcc0
                     SyS_mount+0x7e/0xd0
                     do_syscall_64+0x79/0x1e0
                     entry_SYSCALL_64_after_hwframe+0x42/0xb7
  }
  ... key      at: [<ffffffffc0729488>] __key.59101+0x0/0xfffffffffff9ab78 [btrfs]
  ... acquired at:
   find_free_extent+0xcb4/0x12d0 [btrfs]
   btrfs_reserve_extent+0xd8/0x170 [btrfs]
   btrfs_alloc_tree_block+0x12f/0x4c0 [btrfs]
   __btrfs_cow_block+0x110/0x5b0 [btrfs]
   btrfs_cow_block+0xd7/0x290 [btrfs]
   btrfs_search_slot+0x1f6/0x960 [btrfs]
   btrfs_lookup_inode+0x2a/0x90 [btrfs]
   __btrfs_update_delayed_inode+0x65/0x210 [btrfs]
   btrfs_commit_inode_delayed_inode+0x121/0x130 [btrfs]
   btrfs_evict_inode+0x3fe/0x6a0 [btrfs]
   evict+0xc4/0x190
   __dentry_kill+0xbf/0x170
   dput+0x2ae/0x2f0
   SyS_rename+0x2a6/0x3b0
   do_syscall_64+0x79/0x1e0
   entry_SYSCALL_64_after_hwframe+0x42/0xb7

-> (&delayed_node->mutex){+.+.-.} ops: 5580204 {
   HARDIRQ-ON-W at:
                    __mutex_lock+0x4e/0x8c0
                    btrfs_delayed_update_inode+0x46/0x6e0 [btrfs]
                    btrfs_update_inode+0x83/0x110 [btrfs]
                    btrfs_dirty_inode+0x62/0xe0 [btrfs]
                    touch_atime+0x8c/0xb0
                    do_generic_file_read+0x818/0xb10
                    __vfs_read+0xdc/0x150
                    vfs_read+0x8a/0x130
                    SyS_read+0x45/0xa0
                    do_syscall_64+0x79/0x1e0
                    entry_SYSCALL_64_after_hwframe+0x42/0xb7
   SOFTIRQ-ON-W at:
                    __mutex_lock+0x4e/0x8c0
                    btrfs_delayed_update_inode+0x46/0x6e0 [btrfs]
                    btrfs_update_inode+0x83/0x110 [btrfs]
                    btrfs_dirty_inode+0x62/0xe0 [btrfs]
                    touch_atime+0x8c/0xb0
                    do_generic_file_read+0x818/0xb10
                    __vfs_read+0xdc/0x150
                    vfs_read+0x8a/0x130
                    SyS_read+0x45/0xa0
                    do_syscall_64+0x79/0x1e0
                    entry_SYSCALL_64_after_hwframe+0x42/0xb7
   IN-RECLAIM_FS-W at:
                       __mutex_lock+0x4e/0x8c0
                       __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
                       btrfs_evict_inode+0x22c/0x6a0 [btrfs]
                       evict+0xc4/0x190
                       dispose_list+0x35/0x50
                       prune_icache_sb+0x42/0x50
                       super_cache_scan+0x139/0x190
                       shrink_slab+0x262/0x5b0
                       shrink_node+0x2eb/0x2f0
                       kswapd+0x2eb/0x890
                       kthread+0x102/0x140
                       ret_from_fork+0x3a/0x50
   INITIAL USE at:
                   __mutex_lock+0x4e/0x8c0
                   btrfs_delayed_update_inode+0x46/0x6e0 [btrfs]
                   btrfs_update_inode+0x83/0x110 [btrfs]
                   btrfs_dirty_inode+0x62/0xe0 [btrfs]
                   touch_atime+0x8c/0xb0
                   do_generic_file_read+0x818/0xb10
                   __vfs_read+0xdc/0x150
                   vfs_read+0x8a/0x130
                   SyS_read+0x45/0xa0
                   do_syscall_64+0x79/0x1e0
                   entry_SYSCALL_64_after_hwframe+0x42/0xb7
 }
 ... key      at: [<ffffffffc072d488>] __key.56935+0x0/0xfffffffffff96b78 [btrfs]
 ... acquired at:
   __lock_acquire+0x264/0x11c0
   lock_acquire+0xbd/0x1e0
   __mutex_lock+0x4e/0x8c0
   __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
   btrfs_evict_inode+0x22c/0x6a0 [btrfs]
   evict+0xc4/0x190
   dispose_list+0x35/0x50
   prune_icache_sb+0x42/0x50
   super_cache_scan+0x139/0x190
   shrink_slab+0x262/0x5b0
   shrink_node+0x2eb/0x2f0
   kswapd+0x2eb/0x890
   kthread+0x102/0x140
   ret_from_fork+0x3a/0x50

stack backtrace:
CPU: 1 PID: 50 Comm: kswapd0 Tainted: G        W        4.12.14-kvmsmall #8 SLE15 (unreleased)
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0x78/0xb7
 print_irq_inversion_bug.part.38+0x19f/0x1aa
 check_usage_forwards+0x102/0x120
 ? ret_from_fork+0x3a/0x50
 ? check_usage_backwards+0x110/0x110
 mark_lock+0x16c/0x270
 __lock_acquire+0x264/0x11c0
 ? pagevec_lookup_entries+0x1a/0x30
 ? truncate_inode_pages_range+0x2b3/0x7f0
 lock_acquire+0xbd/0x1e0
 ? __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
 __mutex_lock+0x4e/0x8c0
 ? __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
 ? __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
 ? btrfs_evict_inode+0x1f6/0x6a0 [btrfs]
 __btrfs_release_delayed_node+0x3a/0x1f0 [btrfs]
 btrfs_evict_inode+0x22c/0x6a0 [btrfs]
 evict+0xc4/0x190
 dispose_list+0x35/0x50
 prune_icache_sb+0x42/0x50
 super_cache_scan+0x139/0x190
 shrink_slab+0x262/0x5b0
 shrink_node+0x2eb/0x2f0
 kswapd+0x2eb/0x890
 kthread+0x102/0x140
 ? mem_cgroup_shrink_node+0x2c0/0x2c0
 ? kthread_create_on_node+0x40/0x40
 ret_from_fork+0x3a/0x50

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop optimal argument from find_live_mirror()
Anand Jain [Wed, 14 Mar 2018 08:29:13 +0000 (16:29 +0800)]
btrfs: drop optimal argument from find_live_mirror()

Drop optimal argument from the function find_live_mirror() as we can
deduce it in the function itself. Also rename optimal to
preferred_mirror.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop num argument from find_live_mirror()
Anand Jain [Wed, 14 Mar 2018 08:29:12 +0000 (16:29 +0800)]
btrfs: drop num argument from find_live_mirror()

Obtain the stripes info from the map directly and so no need
to pass it as an argument.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Drop fs_info parameter from __btrfs_run_delayed_refs
Nikolay Borisov [Thu, 15 Mar 2018 14:00:27 +0000 (16:00 +0200)]
btrfs: Drop fs_info parameter from __btrfs_run_delayed_refs

It's provided by transaction handle.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Drop fs_info parameter from btrfs_finish_extent_commit
Nikolay Borisov [Thu, 15 Mar 2018 14:00:26 +0000 (16:00 +0200)]
btrfs: Drop fs_info parameter from btrfs_finish_extent_commit

It's provided by the transaction handle.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Drop fs_info parameter from btrfs_qgroup_account_extents
Nikolay Borisov [Thu, 15 Mar 2018 14:00:25 +0000 (16:00 +0200)]
btrfs: Drop fs_info parameter from btrfs_qgroup_account_extents

It's provided by the transaction handle.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from btrfs_run_delayed_refs
Nikolay Borisov [Thu, 15 Mar 2018 15:27:37 +0000 (17:27 +0200)]
btrfs: drop fs_info parameter from btrfs_run_delayed_refs

It's provided by the transaction handle.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused flush var in shrink_delalloc
Nikolay Borisov [Thu, 15 Mar 2018 12:36:29 +0000 (14:36 +0200)]
btrfs: Remove unused flush var in shrink_delalloc

Added by 08e007d2e577 ("Btrfs: improve the noflush reservation") and
made redundant by 17024ad0a0fd ("Btrfs: fix early ENOSPC due to
delalloc").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused extent_root var from caching_thread
Nikolay Borisov [Thu, 15 Mar 2018 12:36:28 +0000 (14:36 +0200)]
btrfs: Remove unused extent_root var from caching_thread

Added by b4570aa994b8 ("btrfs: fix compiling with CONFIG_BTRFS_DEBUG
enabled.") and obsoleted by 2ff7e61e0d30 ("btrfs: take an fs_info
directly when the root is not used otherwise").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove max_active var from open_ctree
Nikolay Borisov [Thu, 15 Mar 2018 12:36:26 +0000 (14:36 +0200)]
btrfs: remove max_active var from open_ctree

Introduced by 5cdc7ad337fb ("btrfs: Replace fs_info->workers with
btrfs_workqueue.") but obsoleted by 2a4581983f90 ("btrfs: factor
btrfs_init_workqueues() out of open_ctree()").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused root var from relink_file_extents
Nikolay Borisov [Thu, 15 Mar 2018 12:36:25 +0000 (14:36 +0200)]
btrfs: Remove unused root var from relink_file_extents

Added in 38c227d87c49 ("Btrfs: snapshot-aware defrag") but subsequently
made redundant by 0b246afa62b0 ("btrfs: root->fs_info cleanup, add
fs_info convenience variables").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused tot_len var from lzo_decompress
Nikolay Borisov [Thu, 15 Mar 2018 12:36:24 +0000 (14:36 +0200)]
btrfs: Remove unused tot_len var from lzo_decompress

Added already unused in a6fa6fae40ec ("btrfs: Add lzo compression
support").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused length var from scrub_handle_errored_block
Nikolay Borisov [Thu, 15 Mar 2018 12:36:23 +0000 (14:36 +0200)]
btrfs: Remove unused length var from scrub_handle_errored_block

Added in b5d67f64f9bc ("Btrfs: change scrub to support big blocks") but
rendered redundant by be50a8ddaae1 ("Btrfs: Simplify
scrub_setup_recheck_block()'s argument").

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Remove unused op_key var from add_delayed_refs
Nikolay Borisov [Thu, 15 Mar 2018 12:36:22 +0000 (14:36 +0200)]
btrfs: Remove unused op_key var from add_delayed_refs

Added as part of 86d5f9944252 ("btrfs: convert prelimary reference
tracking to use rbtrees") but never used. tmp_op_key essentially
subsumed that variable.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: volumes: Remove the meaningless condition of minimal nr_devs when allocating...
Qu Wenruo [Wed, 31 Jan 2018 05:56:15 +0000 (13:56 +0800)]
btrfs: volumes: Remove the meaningless condition of minimal nr_devs when allocating a chunk

When checking the minimal nr_devs, there is one dead and meaningless
condition:

if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

This condition is meaningless, @devs_increment has nothing to do with
@sub_stripes.

In fact, in btrfs_raid_array[], profile with sub_stripes larger than 1
(RAID10) already has the @devs_increment set to 2.
So no need to multiple it by @sub_stripes.

And above condition is also dead.
For RAID10, @devs_increment * @sub_stripes equals 4, which is also the
@devs_min of RAID10.
For other profiles, @sub_stripes is always 1, and since @ndevs is
rounded down to @devs_increment, the condition will always be true.

Remove the meaningless condition to make later reader wander less.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Document parameters of btrfs_reserve_extent
Nikolay Borisov [Tue, 13 Mar 2018 10:22:32 +0000 (12:22 +0200)]
btrfs: Document parameters of btrfs_reserve_extent

This function is the entry to the extent allocator and as such has
quite a number of parameters. Some of those have subtle effects on the
allocation algorithm. Document the parameters.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Handle error from btrfs_uuid_tree_rem call in _btrfs_ioctl_set_received_subvol
Nikolay Borisov [Mon, 12 Mar 2018 12:48:09 +0000 (14:48 +0200)]
btrfs: Handle error from btrfs_uuid_tree_rem call in _btrfs_ioctl_set_received_subvol

As with every function which deals with modifying the btree
btrfs_uuid_tree_rem can fail for any number of reasons (ie. EIO/ENOMEM).
Handle return error value from this function gracefully by aborting the
transaction.

Fixes: dd5f9615fc5c ("Btrfs: maintain subvolume items in the UUID tree")
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: Use sizeof directly instead of a constant variable
Nikolay Borisov [Tue, 13 Mar 2018 08:26:06 +0000 (10:26 +0200)]
btrfs: Use sizeof directly instead of a constant variable

The kernel would like to have all stack VLA usage removed[1].
Unfortunately using an integer constant variable as the size of an
array is still considered a VLA. Instead let's use directly sizeof(var)
which removes the VLA usage. Use the occasion to remove csum_size
altogether and use sizeof() also for the size passed to memcmp

[1]: https://lkml.org/lkml/2018/3/7/621

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: rename submit callbacks and drop double underscores
David Sterba [Thu, 8 Mar 2018 13:35:48 +0000 (14:35 +0100)]
btrfs: rename submit callbacks and drop double underscores

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove unused parameters from extent_submit_bio_done_t
David Sterba [Thu, 8 Mar 2018 12:47:33 +0000 (13:47 +0100)]
btrfs: remove unused parameters from extent_submit_bio_done_t

Remove parameters not used by any of the callbacks.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove unused parameters from extent_submit_bio_start_t
David Sterba [Thu, 8 Mar 2018 12:47:33 +0000 (13:47 +0100)]
btrfs: remove unused parameters from extent_submit_bio_start_t

Remove parameters not used by any of the callbacks.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: separate types for submit_bio_start and submit_bio_done
David Sterba [Fri, 23 Jun 2017 01:05:23 +0000 (03:05 +0200)]
btrfs: separate types for submit_bio_start and submit_bio_done

The callbacks make use of different parameters that are passed to the
other type unnecessarily. This patch adds separate types for each and
the unused parameters will be removed.

The type extent_submit_bio_hook_t keeps all parameters and can be used
where the start/done types are not appropriate.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: kill tree_mod_log_set_root_pointer helper
David Sterba [Mon, 5 Mar 2018 15:35:29 +0000 (16:35 +0100)]
btrfs: kill tree_mod_log_set_root_pointer helper

A useless wrapper around tree_mod_log_insert_root that hides missing
error handling. Move it to the callers.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: kill tree_mod_log_set_node_key helper
David Sterba [Mon, 5 Mar 2018 15:16:54 +0000 (16:16 +0100)]
btrfs: kill tree_mod_log_set_node_key helper

A trivial wrapper that can be simply opencoded and makes the GFP
allocation request more visible. The error handling is now moved to the
callers.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: kill trivial wrapper tree_mod_log_eb_move
David Sterba [Mon, 5 Mar 2018 14:47:39 +0000 (15:47 +0100)]
btrfs: kill trivial wrapper tree_mod_log_eb_move

The wrapper is effectively an alias for tree_mod_log_insert_move but
also hides the missing error handling. To make that more visible, lift
the BUG_ON to the callers.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove trivial locking wrappers of tree mod log
David Sterba [Mon, 5 Mar 2018 14:43:41 +0000 (15:43 +0100)]
btrfs: remove trivial locking wrappers of tree mod log

The wrappers are trivial and do not bring any extra value on top of the
plain locking primitives.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from __tree_mod_log_oldest_root
David Sterba [Mon, 5 Mar 2018 14:33:18 +0000 (15:33 +0100)]
btrfs: drop fs_info parameter from __tree_mod_log_oldest_root

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: embed tree_mod_move structure to tree_mod_elem
David Sterba [Mon, 5 Mar 2018 14:31:18 +0000 (15:31 +0100)]
btrfs: embed tree_mod_move structure to tree_mod_elem

The tree_mod_move is not used anywhere and can be embedded as anonymous
structure.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop unused fs_info parameter from tree_mod_log_eb_move
David Sterba [Mon, 5 Mar 2018 14:26:29 +0000 (15:26 +0100)]
btrfs: drop unused fs_info parameter from tree_mod_log_eb_move

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from tree_mod_log_free_eb
David Sterba [Mon, 5 Mar 2018 14:22:30 +0000 (15:22 +0100)]
btrfs: drop fs_info parameter from tree_mod_log_free_eb

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from tree_mod_log_free_eb
David Sterba [Mon, 5 Mar 2018 14:14:25 +0000 (15:14 +0100)]
btrfs: drop fs_info parameter from tree_mod_log_free_eb

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from tree_mod_log_insert_key
David Sterba [Mon, 5 Mar 2018 14:09:03 +0000 (15:09 +0100)]
btrfs: drop fs_info parameter from tree_mod_log_insert_key

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from tree_mod_log_insert_move
David Sterba [Mon, 5 Mar 2018 14:03:52 +0000 (15:03 +0100)]
btrfs: drop fs_info parameter from tree_mod_log_insert_move

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: drop fs_info parameter from tree_mod_log_set_node_key
David Sterba [Mon, 5 Mar 2018 14:00:37 +0000 (15:00 +0100)]
btrfs: drop fs_info parameter from tree_mod_log_set_node_key

It's provided by the extent_buffer.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: document more parameters of submit_extent_page
David Sterba [Mon, 12 Jun 2017 17:50:41 +0000 (19:50 +0200)]
btrfs: document more parameters of submit_extent_page

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: cleanup merging conditions in submit_extent_page
David Sterba [Mon, 12 Jun 2017 18:00:43 +0000 (20:00 +0200)]
btrfs: cleanup merging conditions in submit_extent_page

The merge call was factored out to a separate helper but it's a trivial
one and arguably we can opencode it and cache the value.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: remove redundant variable in __do_readpage
David Sterba [Tue, 6 Jun 2017 17:50:13 +0000 (19:50 +0200)]
btrfs: remove redundant variable in __do_readpage

The value of page_end is only stored to end, no other use.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agobtrfs: assume that bio_ret is always valid in submit_extent_page
David Sterba [Tue, 6 Jun 2017 17:22:55 +0000 (19:22 +0200)]
btrfs: assume that bio_ret is always valid in submit_extent_page

All callers pass a valid pointer so we can drop the redundant checks.
The call to submit_one_bio never happend and can be removed.

Signed-off-by: David Sterba <dsterba@suse.com>
6 years agoBtrfs: scrub: batch rebuild for raid56
Liu Bo [Wed, 7 Mar 2018 19:08:09 +0000 (12:08 -0700)]
Btrfs: scrub: batch rebuild for raid56

In case of raid56, writes and rebuilds always take BTRFS_STRIPE_LEN(64K)
as unit, however, scrub_extent() sets blocksize as unit, so rebuild
process may be triggered on every block on a same stripe.

A typical example would be that when we're replacing a disappeared disk,
all reads on the disks get -EIO, every block (size is 4K if blocksize is
4K) would go thru these,

scrub_handle_errored_block
  scrub_recheck_block # re-read pages one by one
  scrub_recheck_block # rebuild by calling raid56_parity_recover()
                        page by page

Although with raid56 stripe cache most of reads during rebuild can be
avoided, the parity recover calculation(xor or raid6 algorithms) needs to
be done $(BTRFS_STRIPE_LEN / blocksize) times.

This makes it smarter by doing raid56 scrub/replace on stripe length.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>