linux-2.6-microblaze.git
17 months agobtrfs: don't call btrfs_record_physical_zoned for failed append
Christoph Hellwig [Wed, 24 May 2023 15:03:05 +0000 (17:03 +0200)]
btrfs: don't call btrfs_record_physical_zoned for failed append

When a zoned append command fails there is no written address reported,
so don't try to record it.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: optimize out btrfs_is_zoned for !CONFIG_BLK_DEV_ZONED
Christoph Hellwig [Wed, 24 May 2023 15:03:04 +0000 (17:03 +0200)]
btrfs: optimize out btrfs_is_zoned for !CONFIG_BLK_DEV_ZONED

Add an IS_ENABLED check for CONFIG_BLK_DEV_ZONED in addition to the
run-time check for the zone size.  This will allow to make use of
compiler dead code elimination for code guarded by btrfs_is_zoned, and
for example provide just a dangling prototype for a function instead of
adding a stub.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: make btrfs_destroy_delayed_refs() return void
Filipe Manana [Fri, 2 Jun 2023 11:19:42 +0000 (12:19 +0100)]
btrfs: make btrfs_destroy_delayed_refs() return void

btrfs_destroy_delayed_refs() always returns 0 and its single caller does
not check its return value, as it also returns void, and so does the
callers' caller and so on. This is because we are in the transaction abort
path, where we have no way to deal with errors (we are in a critical
situation) and all cleanup of resources works in a best effort fashion.
So make btrfs_destroy_delayed_refs() return void.

Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove unnecessary prototype declarations at disk-io.c
Filipe Manana [Mon, 29 May 2023 15:17:06 +0000 (16:17 +0100)]
btrfs: remove unnecessary prototype declarations at disk-io.c

We have a few static functions at disk-io.c for which we have a forward
declaration of their prototype, but it's not needed because all those
functions are defined before they are called, so remove them.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use a single switch statement when initializing delayed ref head
Filipe Manana [Mon, 29 May 2023 15:17:05 +0000 (16:17 +0100)]
btrfs: use a single switch statement when initializing delayed ref head

At init_delayed_ref_head(), we are using two separate if statements to
check the delayed ref head action, and initializing 'must_insert_reserved'
to false twice, once when the variable is declared and once again in an
else branch.

Make this simpler and more straightforward by having a single switch
statement, also moving the comment about a drop action to the
corresponding switch case to make it more clear and eliminating the
duplicated initialization of 'must_insert_reserved' to false.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use bool type for delayed ref head fields that are used as booleans
Filipe Manana [Mon, 29 May 2023 15:17:04 +0000 (16:17 +0100)]
btrfs: use bool type for delayed ref head fields that are used as booleans

There's no point in have several fields defined as 1 bit unsigned int in
struct btrfs_delayed_ref_head, we can instead use a bool type, it makes
the code a bit more readable and it doesn't change the structure size.
So switch them to proper booleans.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: assert correct lock is held at btrfs_select_ref_head()
Filipe Manana [Mon, 29 May 2023 15:17:03 +0000 (16:17 +0100)]
btrfs: assert correct lock is held at btrfs_select_ref_head()

The function btrfs_select_ref_head() iterates over the red black tree of
delayed reference heads, which is protected by the spinlock in the delayed
refs root. The function doesn't take the lock, it's taken by its single
caller, btrfs_obtain_ref_head(), because it needs to call that function
and btrfs_delayed_ref_lock() in the same critical section (delimited by
that spinlock). So assert at btrfs_select_ref_head() that we are holding
the expected lock.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: get rid of label and goto at insert_delayed_ref()
Filipe Manana [Mon, 29 May 2023 15:17:02 +0000 (16:17 +0100)]
btrfs: get rid of label and goto at insert_delayed_ref()

At insert_delayed_ref() there's no point of having a label and goto in the
case we were able to insert the delayed ref head. We can just add the code
under label to the if statement's body and return immediately, and also
there is no need to track the return value in a variable, we can just
return a literal true or false value directly. So do those changes.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: make insert_delayed_ref() return a bool instead of an int
Filipe Manana [Mon, 29 May 2023 15:17:01 +0000 (16:17 +0100)]
btrfs: make insert_delayed_ref() return a bool instead of an int

insert_delayed_ref() can only return 0 or 1, to indicate if the given
delayed reference was added to the head reference or if it was merged
into an existing delayed ref, respectively. So just make it return a
boolean instead.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use a bool to track qgroup record insertion when adding ref head
Filipe Manana [Mon, 29 May 2023 15:17:00 +0000 (16:17 +0100)]
btrfs: use a bool to track qgroup record insertion when adding ref head

We are using an integer as a boolean to track the qgroup record insertion
status when adding a delayed reference head. Since all we need is a
boolean, switch the type from int to bool to make it more obvious.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove pointless in_tree field from struct btrfs_delayed_ref_node
Filipe Manana [Mon, 29 May 2023 15:16:59 +0000 (16:16 +0100)]
btrfs: remove pointless in_tree field from struct btrfs_delayed_ref_node

The 'in_tree' field is really not needed in struct btrfs_delayed_ref_node,
as we can check whether a reference is in the tree or not simply by
checking its red black tree node member with RB_EMPTY_NODE(), as when we
remove it from the tree we always call RB_CLEAR_NODE(). So remove that
field and use RB_EMPTY_NODE().

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove unused is_head field from struct btrfs_delayed_ref_node
Filipe Manana [Mon, 29 May 2023 15:16:58 +0000 (16:16 +0100)]
btrfs: remove unused is_head field from struct btrfs_delayed_ref_node

The 'is_head' field of struct btrfs_delayed_ref_node is no longer after
commit d278850eff30 ("btrfs: remove delayed_ref_node from ref_head"),
so remove it.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: reorder some members of struct btrfs_delayed_ref_head
Filipe Manana [Mon, 29 May 2023 15:16:57 +0000 (16:16 +0100)]
btrfs: reorder some members of struct btrfs_delayed_ref_head

Currently struct delayed_ref_head has its 'bytenr' and 'href_node' members
in different cache lines (even on a release, non-debug, kernel). This is
not optimal because when iterating the red black tree of delayed ref heads
for inserting a new delayed ref head (htree_insert()) we have to pull in 2
cache lines of delayed ref heads we find in a patch, one for the tree node
(struct rb_node) and another one for the 'bytenr' field. The same applies
when searching for an existing delayed ref head (find_ref_head()).
On a release (non-debug) kernel, the structure also has two 4 bytes holes,
which makes it 8 bytes longer than necessary. Its current layout is the
following:

  struct btrfs_delayed_ref_head {
          u64                        bytenr;               /*     0     8 */
          u64                        num_bytes;            /*     8     8 */
          refcount_t                 refs;                 /*    16     4 */

          /* XXX 4 bytes hole, try to pack */

          struct mutex               mutex;                /*    24    32 */
          spinlock_t                 lock;                 /*    56     4 */

          /* XXX 4 bytes hole, try to pack */

          /* --- cacheline 1 boundary (64 bytes) --- */
          struct rb_root_cached      ref_tree;             /*    64    16 */
          struct list_head           ref_add_list;         /*    80    16 */
          struct rb_node             href_node __attribute__((__aligned__(8))); /*    96    24 */
          struct btrfs_delayed_extent_op * extent_op;      /*   120     8 */
          /* --- cacheline 2 boundary (128 bytes) --- */
          int                        total_ref_mod;        /*   128     4 */
          int                        ref_mod;              /*   132     4 */
          unsigned int               must_insert_reserved:1; /*   136: 0  4 */
          unsigned int               is_data:1;            /*   136: 1  4 */
          unsigned int               is_system:1;          /*   136: 2  4 */
          unsigned int               processing:1;         /*   136: 3  4 */

          /* size: 144, cachelines: 3, members: 15 */
          /* sum members: 128, holes: 2, sum holes: 8 */
          /* sum bitfield members: 4 bits (0 bytes) */
          /* padding: 4 */
          /* bit_padding: 28 bits */
          /* forced alignments: 1 */
          /* last cacheline: 16 bytes */
  } __attribute__((__aligned__(8)));

This change reorders the 'href_node' and 'refs' members so that we have
the 'href_node' in the same cache line as the 'bytenr' field, while also
eliminating the two holes and reducing the structure size from 144 bytes
down to 136 bytes, so we can now have 30 ref heads per 4K page (on x86_64)
instead of 28. The new structure layout after this change is now:

  struct btrfs_delayed_ref_head {
          u64                        bytenr;               /*     0     8 */
          u64                        num_bytes;            /*     8     8 */
          struct rb_node             href_node __attribute__((__aligned__(8))); /*    16    24 */
          struct mutex               mutex;                /*    40    32 */
          /* --- cacheline 1 boundary (64 bytes) was 8 bytes ago --- */
          refcount_t                 refs;                 /*    72     4 */
          spinlock_t                 lock;                 /*    76     4 */
          struct rb_root_cached      ref_tree;             /*    80    16 */
          struct list_head           ref_add_list;         /*    96    16 */
          struct btrfs_delayed_extent_op * extent_op;      /*   112     8 */
          int                        total_ref_mod;        /*   120     4 */
          int                        ref_mod;              /*   124     4 */
          /* --- cacheline 2 boundary (128 bytes) --- */
          unsigned int               must_insert_reserved:1; /*   128: 0  4 */
          unsigned int               is_data:1;            /*   128: 1  4 */
          unsigned int               is_system:1;          /*   128: 2  4 */
          unsigned int               processing:1;         /*   128: 3  4 */

          /* size: 136, cachelines: 3, members: 15 */
          /* padding: 4 */
          /* bit_padding: 28 bits */
          /* forced alignments: 1 */
          /* last cacheline: 8 bytes */
  } __attribute__((__aligned__(8)));

Running the following fs_mark test shows some significant improvement.

  $ cat test.sh
  #!/bin/bash

  # 15G null block device
  DEV=/dev/nullb0
  MNT=/mnt/nullb0
  FILES=100000
  THREADS=$(nproc --all)
  FILE_SIZE=0

  echo "performance" | \
      tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

  mkfs.btrfs -f $DEV
  mount -o ssd $DEV $MNT

  OPTS="-S 0 -L 5 -n $FILES -s $FILE_SIZE -t $THREADS -k"
  for ((i = 1; i <= $THREADS; i++)); do
      OPTS="$OPTS -d $MNT/d$i"
  done

  fs_mark $OPTS

  umount $MNT

Before this change:

FSUse%        Count         Size    Files/sec     App Overhead
    10      1200000            0     112631.3         11928055
    16      2400000            0     189943.8         12140777
    23      3600000            0     150719.2         13178480
    50      4800000            0      99137.3         12504293
    53      6000000            0     111733.9         12670836

                    Total files/sec: 664165.5

After this change:

FSUse%        Count         Size    Files/sec     App Overhead
    10      1200000            0     148589.5         11565889
    16      2400000            0     227743.8         11561596
    23      3600000            0     191590.5         12550755
    30      4800000            0     179812.3         12629610
    53      6000000            0      92471.4         12352383

                    Total files/sec: 840207.5

Measuring the execution times of htree_insert(), in nanoseconds, during
those fs_mark runs:

Before this change:

  Range:  0.000 - 940647.000; Mean: 619.733; Median: 548.000; Stddev: 1834.231
  Percentiles:  90th: 980.000; 95th: 1208.000; 99th: 2090.000
     0.000 -    6.384:       257 |
     6.384 -   26.259:       977 |
    26.259 -   99.635:      4963 |
    99.635 -  370.526:    136800 #############
   370.526 - 1370.603:    566110 #####################################################
  1370.603 - 5062.704:     24945 ##
  5062.704 - 18693.248:      944 |
  18693.248 - 69014.670:     211 |
  69014.670 - 254791.959:     30 |
  254791.959 - 940647.000:     4 |

After this change:

  Range:  0.000 - 299200.000; Mean: 587.754; Median: 542.000; Stddev: 1030.422
  Percentiles:  90th: 918.000; 95th: 1113.000; 99th: 1987.000
     0.000 -    5.585:      163 |
     5.585 -   20.678:      452 |
    20.678 -   70.369:     1806 |
    70.369 -  233.965:    26268 ####
   233.965 -  772.564:   333519 #####################################################
   772.564 - 2545.771:    91820 ###############
  2545.771 - 8383.615:     2238 |
  8383.615 - 27603.280:     170 |
  27603.280 - 90879.297:     68 |
  90879.297 - 299200.000:    12 |

Mean, percentiles, maximum times are all better, as well as a lower
standard deviation.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use the same uptodate variable for end_bio_extent_readpage()
Qu Wenruo [Tue, 30 May 2023 01:45:28 +0000 (09:45 +0800)]
btrfs: use the same uptodate variable for end_bio_extent_readpage()

In function end_bio_extent_readpage() we call
endio_readpage_release_extent() to unlock the extent io tree.

However we pass PageUptodate(page) as @uptodate parameter for it, while
for previous end_page_read() call, we use a dedicated @uptodate local
variable.

This is not a big deal, as even for subpage cases, either the bio only
covers part of the page, then the @uptodate is always false, and the
subpage ranges can still be merged.

But for the sake of consistency, always use @uptodate variable when
possible.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: subpage: make alloc_extent_buffer() handle previously uptodate range efficiently
Qu Wenruo [Tue, 30 May 2023 01:45:27 +0000 (09:45 +0800)]
btrfs: subpage: make alloc_extent_buffer() handle previously uptodate range efficiently

Currently alloc_extent_buffer() would make the extent buffer uptodate if
the corresponding pages are also uptodate.

But this check is only checking PageUptodate, which is fine for regular
cases, but not for subpage cases, as we can have multiple extent buffers
in the same page.

So here we go btrfs_page_test_uptodate() instead.

The old code doesn't cause any problem, but is not efficient, as it
would cause extra metadata read even if the range is already uptodate.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: print assertion failure report and stack trace from the same line
David Sterba [Wed, 3 May 2023 19:08:16 +0000 (21:08 +0200)]
btrfs: print assertion failure report and stack trace from the same line

Assertions reports are split into two parts, the exact file and location
of the condition and then the stack trace printed from
btrfs_assertfail(). This means all the stack traces report the same line
and this is what's typically reported by various tools, making it harder
to distinguish the reports.

  [403.2467] assertion failed: refcount_read(&block_group->refs) == 1, in fs/btrfs/block-group.c:4259
  [403.2479] ------------[ cut here ]------------
  [403.2484] kernel BUG at fs/btrfs/messages.c:259!
  [403.2488] invalid opcode: 0000 [#1] PREEMPT SMP KASAN
  [403.2493] CPU: 2 PID: 23202 Comm: umount Not tainted 6.2.0-rc4-default+ #67
  [403.2499] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552-rebuilt.opensuse.org 04/01/2014
  [403.2509] RIP: 0010:btrfs_assertfail+0x19/0x1b [btrfs]
  ...
  [403.2595] Call Trace:
  [403.2598]  <TASK>
  [403.2601]  btrfs_free_block_groups.cold+0x52/0xae [btrfs]
  [403.2608]  close_ctree+0x6c2/0x761 [btrfs]
  [403.2613]  ? __wait_for_common+0x2b8/0x360
  [403.2618]  ? btrfs_cleanup_one_transaction.cold+0x7a/0x7a [btrfs]
  [403.2626]  ? mark_held_locks+0x6b/0x90
  [403.2630]  ? lockdep_hardirqs_on_prepare+0x13d/0x200
  [403.2636]  ? __call_rcu_common.constprop.0+0x1ea/0x3d0
  [403.2642]  ? trace_hardirqs_on+0x2d/0x110
  [403.2646]  ? __call_rcu_common.constprop.0+0x1ea/0x3d0
  [403.2652]  generic_shutdown_super+0xb0/0x1c0
  [403.2657]  kill_anon_super+0x1e/0x40
  [403.2662]  btrfs_kill_super+0x25/0x30 [btrfs]
  [403.2668]  deactivate_locked_super+0x4c/0xc0

By making btrfs_assertfail a macro we'll get the same line number for
the BUG output:

  [63.5736] assertion failed: 0, in fs/btrfs/super.c:1572
  [63.5758] ------------[ cut here ]------------
  [63.5782] kernel BUG at fs/btrfs/super.c:1572!
  [63.5807] invalid opcode: 0000 [#2] PREEMPT SMP KASAN
  [63.5831] CPU: 0 PID: 859 Comm: mount Tainted: G      D            6.3.0-rc7-default+ #2062
  [63.5868] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a-rebuilt.opensuse.org 04/01/2014
  [63.5905] RIP: 0010:btrfs_mount+0x24/0x30 [btrfs]
  [63.5964] RSP: 0018:ffff88800e69fcd8 EFLAGS: 00010246
  [63.5982] RAX: 000000000000002d RBX: ffff888008fc1400 RCX: 0000000000000000
  [63.6004] RDX: 0000000000000000 RSI: ffffffffb90fd868 RDI: ffffffffbcc3ff20
  [63.6026] RBP: ffffffffc081b200 R08: 0000000000000001 R09: ffff88800e69fa27
  [63.6046] R10: ffffed1001cd3f44 R11: 0000000000000001 R12: ffff888005a3c370
  [63.6062] R13: ffffffffc058e830 R14: 0000000000000000 R15: 00000000ffffffff
  [63.6081] FS:  00007f7b3561f800(0000) GS:ffff88806c600000(0000) knlGS:0000000000000000
  [63.6105] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  [63.6120] CR2: 00007fff83726e10 CR3: 0000000002a9e000 CR4: 00000000000006b0
  [63.6137] Call Trace:
  [63.6143]  <TASK>
  [63.6148]  legacy_get_tree+0x80/0xd0
  [63.6158]  vfs_get_tree+0x43/0x120
  [63.6166]  do_new_mount+0x1f3/0x3d0
  [63.6176]  ? do_add_mount+0x140/0x140
  [63.6187]  ? cap_capable+0xa4/0xe0
  [63.6197]  path_mount+0x223/0xc10

This comes at a cost of bloating the final btrfs.ko module due all the
inlining, as long as assertions are compiled in. This is a must for
debugging builds but this is often enabled on release builds too.

Release build:

   text    data     bss     dec     hex filename
1251676   20317   16088 1288081  13a791 pre/btrfs.ko
1260612   29473   16088 1306173  13ee3d post/btrfs.ko

DELTA: +8936

CC: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: subpage: dump extra subpage bitmaps for debug
Qu Wenruo [Fri, 26 May 2023 12:30:53 +0000 (20:30 +0800)]
btrfs: subpage: dump extra subpage bitmaps for debug

There is a bug report that assert_eb_page_uptodate() gets triggered for
free space tree metadata.

Without proper dump for the subpage bitmaps it's much harder to debug.

Thus this patch would dump all the subpage bitmaps (split them into
their own bitmaps) for a easier debugging.

The output would look like this:
(Dumped after a tree block got read from disk)

  page:000000006e34bf49 refcount:4 mapcount:0 mapping:0000000067661ac4 index:0x1d1 pfn:0x110e9
  memcg:ffff0000d7d62000
  aops:btree_aops [btrfs] ino:1
  flags: 0x8000000000002002(referenced|private|zone=2)
  page_type: 0xffffffff()
  raw: 8000000000002002 0000000000000000 dead000000000122 ffff00000188bed0
  raw: 00000000000001d1 ffff0000c7992700 00000004ffffffff ffff0000d7d62000
  page dumped because: btrfs subpage dump
  BTRFS warning (device dm-1): start=30490624 len=16384 page=30474240 bitmaps: uptodate=4-7 error= dirty= writeback= ordered= checked=

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use alloc_ordered_workqueue() to create ordered workqueues
Tejun Heo [Thu, 25 May 2023 23:33:08 +0000 (13:33 -1000)]
btrfs: use alloc_ordered_workqueue() to create ordered workqueues

BACKGROUND
==========

When multiple work items are queued to a workqueue, their execution order
doesn't match the queueing order. They may get executed in any order and
simultaneously. When fully serialized execution - one by one in the queueing
order - is needed, an ordered workqueue should be used which can be created
with alloc_ordered_workqueue().

However, alloc_ordered_workqueue() was a later addition. Before it, an
ordered workqueue could be obtained by creating an UNBOUND workqueue with
@max_active==1. This originally was an implementation side-effect which was
broken by 4c16bd327c74 ("workqueue: restore WQ_UNBOUND/max_active==1 to be
ordered"). Because there were users that depended on the ordered execution,
5c0338c68706 ("workqueue: restore WQ_UNBOUND/max_active==1 to be ordered")
made workqueue allocation path to implicitly promote UNBOUND workqueues w/
@max_active==1 to ordered workqueues.

While this has worked okay, overloading the UNBOUND allocation interface
this way creates other issues. It's difficult to tell whether a given
workqueue actually needs to be ordered and users that legitimately want a
min concurrency level wq unexpectedly gets an ordered one instead. With
planned UNBOUND workqueue updates to improve execution locality and more
prevalence of chiplet designs which can benefit from such improvements, this
isn't a state we wanna be in forever.

This patch series audits all call sites that create an UNBOUND workqueue w/
@max_active==1 and converts them to alloc_ordered_workqueue() as necessary.

BTRFS
=====

* fs_info->scrub_workers initialized in scrub_workers_get() was setting
  @max_active to 1 when @is_dev_replace is set and it seems that the
  workqueue actually needs to be ordered if @is_dev_replace. Update the code
  so that alloc_ordered_workqueue() is used if @is_dev_replace.

* fs_info->discard_ctl.discard_workers initialized in
  btrfs_init_workqueues() was directly using alloc_workqueue() w/
  @max_active==1. Converted to alloc_ordered_workqueue().

* fs_info->fixup_workers and fs_info->qgroup_rescan_workers initialized in
  btrfs_queue_work() use the btrfs's workqueue wrapper, btrfs_workqueue,
  which are allocated with btrfs_alloc_workqueue().

  btrfs_workqueue implements automatic @max_active adjustment which is
  disabled when the specified max limit is below a certain threshold, so
  calling btrfs_alloc_workqueue() with @limit_active==1 yields an ordered
  workqueue whose @max_active won't be changed as the auto-tuning is
  disabled.

  This is rather brittle in that nothing clearly indicates that the two
  workqueues should be ordered or btrfs_alloc_workqueue() must disable
  auto-tuning when @limit_active==1.

  This patch factors out the common btrfs_workqueue init code into
  btrfs_init_workqueue() and add explicit btrfs_alloc_ordered_workqueue().
  The two workqueues are converted to use the new ordered allocation
  interface.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: drop gfp from parameter extent state helpers
David Sterba [Wed, 24 May 2023 23:04:39 +0000 (01:04 +0200)]
btrfs: drop gfp from parameter extent state helpers

Now that all extent state bit helpers effectively take the GFP_NOFS mask
(and GFP_NOWAIT is encoded in the bits) we can remove the parameter.
This reduces stack consumption in many functions and simplifies a lot of
code.

Net effect on module on a release build:

   text    data     bss     dec     hex filename
1250432   20985   16088 1287505  13a551 pre/btrfs.ko
1247074   20985   16088 1284147  139833 post/btrfs.ko

DELTA: -3358

Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: pass NOWAIT for set/clear extent bits as another bit
David Sterba [Wed, 24 May 2023 23:04:37 +0000 (01:04 +0200)]
btrfs: pass NOWAIT for set/clear extent bits as another bit

The only flags we now pass to set_extent_bit/__clear_extent_bit are
GFP_NOFS and GFP_NOWAIT (a few functions handling mappings). This
requires an extra parameter to be passed everywhere but is almost always
the same.

Encode the GFP_NOWAIT as an artificial extent bit and extract the
real bits and gfp mask in the lowest level helpers. Now the passed
gfp mask is not actually used and can be removed.

Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: drop NOFAIL from set_extent_bit allocation masks
David Sterba [Wed, 24 May 2023 23:04:34 +0000 (01:04 +0200)]
btrfs: drop NOFAIL from set_extent_bit allocation masks

The __GFP_NOFAIL passed to set_extent_bit first appeared in 2010
(commit f0486c68e4bd9a ("Btrfs: Introduce contexts for metadata
reservation")), without any explanation why it would be needed.

Meanwhile we've updated the semantics of set_extent_bit to handle failed
allocations and do unlock, sleep and retry if needed.  The use of the
NOFAIL flag is also an outlier, we never want any of the set/clear
extent bit helpers to fail, they're used for many critical changes like
extent locking, besides the extent state bit changes.

Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_bits
David Sterba [Wed, 24 May 2023 23:04:32 +0000 (01:04 +0200)]
btrfs: open code set_extent_bits

This helper calls set_extent_bit with two more parameters set to default
values, but otherwise it's purpose is not clear.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_bits_nowait
David Sterba [Wed, 24 May 2023 23:04:30 +0000 (01:04 +0200)]
btrfs: open code set_extent_bits_nowait

The helper only passes GFP_NOWAIT as gfp flags and is used two times.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_dirty
David Sterba [Wed, 24 May 2023 23:04:28 +0000 (01:04 +0200)]
btrfs: open code set_extent_dirty

The helper is used a few times, that it's setting the DIRTY extent bit
is still clear.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_new
David Sterba [Wed, 24 May 2023 23:04:26 +0000 (01:04 +0200)]
btrfs: open code set_extent_new

The helper is used only once.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_delalloc
David Sterba [Wed, 24 May 2023 23:04:23 +0000 (01:04 +0200)]
btrfs: open code set_extent_delalloc

The helper is used once in fs code and a few times in the self test
code.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: open code set_extent_defrag
David Sterba [Wed, 24 May 2023 23:04:21 +0000 (01:04 +0200)]
btrfs: open code set_extent_defrag

The helper is used only once.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove a pointless NULL check in btrfs_lookup_fs_root
Christoph Hellwig [Tue, 23 May 2023 08:40:20 +0000 (10:40 +0200)]
btrfs: remove a pointless NULL check in btrfs_lookup_fs_root

btrfs_grab_root already checks for a NULL root itself.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: convert btrfs_get_global_root to use a switch statement
Christoph Hellwig [Tue, 23 May 2023 08:40:19 +0000 (10:40 +0200)]
btrfs: convert btrfs_get_global_root to use a switch statement

Use a switch statement instead of an endless chain of if statements
to make the code a little cleaner.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: fix the btrfs_get_global_root return value
Christoph Hellwig [Tue, 23 May 2023 08:40:18 +0000 (10:40 +0200)]
btrfs: fix the btrfs_get_global_root return value

btrfs_grab_root returns either the root or NULL, and the callers of
btrfs_get_global_root expect it to return the same.  But all the more
recently added roots instead return an ERR_PTR, so fix this.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add and fix comments in btrfs_fs_devices
Anand Jain [Wed, 24 May 2023 12:02:43 +0000 (20:02 +0800)]
btrfs: add and fix comments in btrfs_fs_devices

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: consolidate uuid comparisons in btrfs_validate_super
Anand Jain [Wed, 24 May 2023 12:02:42 +0000 (20:02 +0800)]
btrfs: consolidate uuid comparisons in btrfs_validate_super

There are three ways the fsid is validated in btrfs_validate_super():

- verify that super_copy::fsid is the same as fs_devices::fsid

- if the metadata_uuid flag is set, verify if super_copy::metadata_uuid
  and fs_devices::metadata_uuid are the same.

- a few lines below, often missed out, verify if dev_item::fsid is the
  same as fs_devices::metadata_uuid.

The function btrfs_validate_super() contains multiple if-statements with
memcmp() to check UUIDs. This patch consolidates them into a single
location.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: simplify how changed fsid and metadata_uuid is checked
Anand Jain [Wed, 24 May 2023 12:02:41 +0000 (20:02 +0800)]
btrfs: simplify how changed fsid and metadata_uuid is checked

We often check if the metadata_uuid is not the same as fsid, and then we
check if the given fsid matches the metadata_uuid. This patch refactors
this logic into function match_fsid_changed and utilize it.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: simplify fsid and metadata_uuid comparisons
Anand Jain [Wed, 24 May 2023 12:02:40 +0000 (20:02 +0800)]
btrfs: simplify fsid and metadata_uuid comparisons

Refactor the functions find_fsid() and find_fsid_with_metadata_uuid(),
as they currently share a common set of code to compare the fsid and
metadata_uuid. Create a common helper function, match_fsid_fs_devices().

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: return bool from check_tree_block_fsid instead of int
Anand Jain [Wed, 24 May 2023 12:02:39 +0000 (20:02 +0800)]
btrfs: return bool from check_tree_block_fsid instead of int

Simplify the return type of check_tree_block_fsid() from int (1 or 0) to
bool. Its only user is interested in knowing the success or failure.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add comment about metadata_uuid in btrfs_fs_devices
Anand Jain [Wed, 24 May 2023 12:02:38 +0000 (20:02 +0800)]
btrfs: add comment about metadata_uuid in btrfs_fs_devices

Add comment about metadata_uuid in btrfs_fs_devices.
No functional change.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: merge calls to alloc_fs_devices in device_list_add
Anand Jain [Wed, 24 May 2023 12:02:37 +0000 (20:02 +0800)]
btrfs: merge calls to alloc_fs_devices in device_list_add

Simplify has_metadata_uuid checks - by localizing the has_metadata_uuid
checked within alloc_fs_devices()'s second argument, it improves the
code readability.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: streamline fsid checks in alloc_fs_devices
Anand Jain [Wed, 24 May 2023 12:02:36 +0000 (20:02 +0800)]
btrfs: streamline fsid checks in alloc_fs_devices

We currently have redundant checks for the non-null value of fsid
simplify it.

And, no one is using alloc_fs_devices() with a NULL metadata_uuid
while fsid is not NULL, add an assert() to verify this condition.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: reduce struct btrfs_fs_devices size by moving fsid_change
Anand Jain [Wed, 24 May 2023 12:02:35 +0000 (20:02 +0800)]
btrfs: reduce struct btrfs_fs_devices size by moving fsid_change

Pack bool fsid_change and bool seeding with other bool declarations in the
struct btrfs_fs_devices, approximately 6 bytes is saved, depending on
the config.

   before: 512 bytes
   after: 496 bytes

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: merge write_one_subpage_eb into write_one_eb
Christoph Hellwig [Wed, 3 May 2023 15:24:41 +0000 (17:24 +0200)]
btrfs: merge write_one_subpage_eb into write_one_eb

Most of the code in write_one_subpage_eb and write_one_eb is shared,
so merge the two functions into one.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use per-buffer locking for extent_buffer reading
Christoph Hellwig [Wed, 3 May 2023 15:24:40 +0000 (17:24 +0200)]
btrfs: use per-buffer locking for extent_buffer reading

Instead of locking and unlocking every page or the extent, just add a
new EXTENT_BUFFER_READING bit that mirrors EXTENT_BUFFER_WRITEBACK
for synchronizing threads trying to read an extent_buffer and to wait
for I/O completion.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: stop using lock_extent in btrfs_buffer_uptodate
Christoph Hellwig [Wed, 3 May 2023 15:24:39 +0000 (17:24 +0200)]
btrfs: stop using lock_extent in btrfs_buffer_uptodate

The only other place that locks extents on the btree inode is
read_extent_buffer_subpage while reading in the partial page for a
buffer.  This means locking the extent in btrfs_buffer_uptodate does not
synchronize with anything on non-subpage file systems, and on subpage
file systems it only waits for a parallel read(-ahead) to finish,
which seems to be counter to what the callers actually expect.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: don't check for uptodate pages in read_extent_buffer_pages
Christoph Hellwig [Wed, 3 May 2023 15:24:38 +0000 (17:24 +0200)]
btrfs: don't check for uptodate pages in read_extent_buffer_pages

The only place that reads in pages and thus marks them uptodate for
the btree inode is read_extent_buffer_pages.  Which means that either
pages are already uptodate from an old buffer when creating a new
one in alloc_extent_buffer, or they will be updated by ca call
to read_extent_buffer_pages.  This means the checks for uptodate
pages in read_extent_buffer_pages and read_extent_buffer_subpage are
superfluous and can be removed.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: stop using PageError for extent_buffers
Christoph Hellwig [Wed, 3 May 2023 15:24:37 +0000 (17:24 +0200)]
btrfs: stop using PageError for extent_buffers

PageError is only used to limit the uptodate check in
assert_eb_page_uptodate.  But we have a much more useful flag indicating
the exact condition we are about with the EXTENT_BUFFER_WRITE_ERR flag,
so use that instead and help the kernel toward eventually removing
PageError.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove the io_pages field in struct extent_buffer
Christoph Hellwig [Wed, 3 May 2023 15:24:36 +0000 (17:24 +0200)]
btrfs: remove the io_pages field in struct extent_buffer

No need to track the number of pages under I/O now that each
extent_buffer is read and written using a single bio.  For the
read side we need to grab an extra reference for the duration of
the I/O to prevent eviction, though.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove the extent_buffer lookup in btree block checksumming
Christoph Hellwig [Wed, 3 May 2023 15:24:35 +0000 (17:24 +0200)]
btrfs: remove the extent_buffer lookup in btree block checksumming

The checksumming of btree blocks always operates on the entire
extent_buffer, and because btree blocks are always allocated contiguously
on disk they are never split by btrfs_submit_bio.

Simplify the checksumming code by finding the extent_buffer in the
btrfs_bio private data instead of trying to search through the bio_vec.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use a separate end_io handler for extent_buffer writing
Christoph Hellwig [Wed, 3 May 2023 15:24:34 +0000 (17:24 +0200)]
btrfs: use a separate end_io handler for extent_buffer writing

Now that we always use a single bio to write an extent_buffer, the buffer
can be passed to the end_io handler as private data.  This allows
to simplify the metadata write end I/O handler, and merge the subpage
end_io handler into the main one.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: don't use btrfs_bio_ctrl for extent buffer writing
Christoph Hellwig [Wed, 3 May 2023 15:24:33 +0000 (17:24 +0200)]
btrfs: don't use btrfs_bio_ctrl for extent buffer writing

The btrfs_bio_ctrl machinery is overkill for writing extent_buffers
as we always operate on PAGE_SIZE chunks (or one smaller one for the
subpage case) that are contiguous and are guaranteed to fit into a
single bio.  Replace it with open coded btrfs_bio_alloc, __bio_add_page
and btrfs_submit_bio calls.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: move page locking from lock_extent_buffer_for_io to write_one_eb
Christoph Hellwig [Wed, 3 May 2023 15:24:32 +0000 (17:24 +0200)]
btrfs: move page locking from lock_extent_buffer_for_io to write_one_eb

Locking the pages in lock_extent_buffer_for_io only for the non-subpage
case is very confusing.  Move it to write_one_eb to mirror the subpage
case and simplify the code. Now lock_extent_buffer_for_io does not leave
all the pages locked and each is individually locked/unlocked in
write_one_eb.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: submit a writeback bio per extent_buffer
Christoph Hellwig [Wed, 3 May 2023 15:24:31 +0000 (17:24 +0200)]
btrfs: submit a writeback bio per extent_buffer

Stop trying to cluster writes of multiple extent_buffers into a single
bio.  There is no need for that as the blk_plug mechanism used all the
way up in writeback_inodes_wb gives us the same I/O pattern even with
multiple bios.  Removing the clustering simplifies
lock_extent_buffer_for_io a lot and will also allow passing the eb
as private data to the end I/O handler.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: return bool from lock_extent_buffer_for_io
Christoph Hellwig [Wed, 3 May 2023 15:24:30 +0000 (17:24 +0200)]
btrfs: return bool from lock_extent_buffer_for_io

lock_extent_buffer_for_io never returns a negative error value, so switch
the return value to a simple bool.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ keep noinline_for_stack ]
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: do not try to unlock the extent for non-subpage metadata reads
Christoph Hellwig [Wed, 3 May 2023 15:24:29 +0000 (17:24 +0200)]
btrfs: do not try to unlock the extent for non-subpage metadata reads

Only subpage metadata reads lock the extent.  Don't try to unlock it and
waste cycles in the extent tree lookup for PAGE_SIZE or larger metadata.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use a separate end_io handler for read_extent_buffer
Christoph Hellwig [Wed, 3 May 2023 15:24:28 +0000 (17:24 +0200)]
btrfs: use a separate end_io handler for read_extent_buffer

Now that we always use a single bio to read an extent_buffer, the buffer
can be passed to the end_io handler as private data.  This allows
implementing a much simplified dedicated end I/O handler for metadata
reads.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove the mirror_num argument to btrfs_submit_compressed_read
Christoph Hellwig [Wed, 3 May 2023 15:24:27 +0000 (17:24 +0200)]
btrfs: remove the mirror_num argument to btrfs_submit_compressed_read

Given that read recovery for data I/O is handled in the storage layer,
the mirror_num argument to btrfs_submit_compressed_read is always 0,
so remove it.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: don't use btrfs_bio_ctrl for extent buffer reading
Christoph Hellwig [Wed, 3 May 2023 15:24:26 +0000 (17:24 +0200)]
btrfs: don't use btrfs_bio_ctrl for extent buffer reading

The btrfs_bio_ctrl machinery is overkill for reading extent_buffers
as we always operate on PAGE_SIZE chunks (or one smaller one for the
subpage case) that are contiguous and are guaranteed to fit into a
single bio.  Replace it with open coded btrfs_bio_alloc, __bio_add_page
and btrfs_submit_bio calls in a helper function shared between
the subpage and node size >= PAGE_SIZE cases.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: always read the entire extent_buffer
Christoph Hellwig [Wed, 3 May 2023 15:24:25 +0000 (17:24 +0200)]
btrfs: always read the entire extent_buffer

Currently read_extent_buffer_pages skips pages that are already uptodate
when reading in an extent_buffer.  While this reduces the amount of data
read, it increases the number of I/O operations as we now need to do
multiple I/Os when reading an extent buffer with one or more uptodate
pages in the middle of it.  On any modern storage device, be that hard
drives or SSDs this actually decreases I/O performance.  Fortunately
this case is pretty rare as the pages are always initially read together
and then aged the same way.  Besides simplifying the code a bit as-is
this will allow for major simplifications to the I/O completion handler
later on.

Note that the case where all pages are uptodate is still handled by an
optimized fast path that does not read any data from disk.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: merge verify_parent_transid and btrfs_buffer_uptodate
Christoph Hellwig [Wed, 3 May 2023 15:24:24 +0000 (17:24 +0200)]
btrfs: merge verify_parent_transid and btrfs_buffer_uptodate

verify_parent_transid is only called by btrfs_buffer_uptodate, which
confusingly inverts the return value.  Merge the two functions and
reflow the parent_transid so that error handling is in a branch.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: move setting the buffer uptodate out of validate_extent_buffer
Christoph Hellwig [Wed, 3 May 2023 15:24:23 +0000 (17:24 +0200)]
btrfs: move setting the buffer uptodate out of validate_extent_buffer

Setting the buffer uptodate in a function that is named as a validation
helper is a it confusing.  Move the call from validate_extent_buffer to
the one of its two callers that didn't already have a duplicate call
to set_extent_buffer_uptodate.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: subpage: fix error handling in end_bio_subpage_eb_writepage
Christoph Hellwig [Wed, 3 May 2023 15:24:22 +0000 (17:24 +0200)]
btrfs: subpage: fix error handling in end_bio_subpage_eb_writepage

Call btrfs_page_clear_uptodate instead of ClearPageUptodate to properly
manage the uptodate bit for the subpage case.

Reported-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: mark extent_buffer_under_io static
Christoph Hellwig [Wed, 3 May 2023 15:24:21 +0000 (17:24 +0200)]
btrfs: mark extent_buffer_under_io static

extent_buffer_under_io is only used in extent_io.c, so mark it static.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: trigger orphan inode cleanup during START_SYNC ioctl
Qu Wenruo [Thu, 11 May 2023 23:42:05 +0000 (07:42 +0800)]
btrfs: trigger orphan inode cleanup during START_SYNC ioctl

There is an internal error report that scrub found an error in an orphan
inode's data.

However there are very limited ways to cleanup such orphan inodes:

- btrfs_start_pre_rw_mount()
  This happens at either mount, or RO->RW switch.
  This is not a viable solution for root fs which may not be unmounted
  or RO mounted.

  Furthermore this doesn't cover every subvolume, it only covers the
  currently cached subvolumes.

- btrfs_lookup_dentry()
  This happens when we first lookup the subvolume dentry.
  But dentry can be cached thus it's not ensured to be triggered every
  time.

- create_snapshot()
  This only happens for the created snapshot, not the source one.

This means if we didn't trigger orphan items cleanup, there is really no
other way to manually trigger it. Add this step to the START_SYNC ioctl.
This is a slight change in the semantics of the ioctl but as sync can be
potentially slow and is usually paired with WAIT_SYNC ioctl.

The errors are not handled because the main point of the ioctl is the
async commit, orphan cleanup is a side effect.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: fix comment referring to no longer existing btrfs_clean_tree_block()
Filipe Manana [Wed, 17 May 2023 11:03:44 +0000 (12:03 +0100)]
btrfs: fix comment referring to no longer existing btrfs_clean_tree_block()

There's a comment at btrfs_init_new_buffer() that refers to a function
named btrfs_clean_tree_block(), however the function was renamed to
btrfs_clear_buffer_dirty() in commit 190a83391bc4 ("btrfs: rename
btrfs_clean_tree_block to btrfs_clear_buffer_dirty"). So update the
comment to refer to the current name.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: change for_rename argument of btrfs_record_unlink_dir() to bool
Filipe Manana [Wed, 17 May 2023 11:02:16 +0000 (12:02 +0100)]
btrfs: change for_rename argument of btrfs_record_unlink_dir() to bool

The for_rename argument of btrfs_record_unlink_dir() is defined as an
integer, but the argument is in fact used as a boolean. So change it to
a boolean to make its use more clear.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove pointless label and goto at btrfs_record_unlink_dir()
Filipe Manana [Wed, 17 May 2023 11:02:15 +0000 (12:02 +0100)]
btrfs: remove pointless label and goto at btrfs_record_unlink_dir()

There's no point of having a label and goto at btrfs_record_unlink_dir()
because the function is trivial and can just return early if we are not
in a rename context. So remove the label and goto and instead return
early if we are not in a rename.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: update comments at btrfs_record_unlink_dir() to be more clear
Filipe Manana [Wed, 17 May 2023 11:02:14 +0000 (12:02 +0100)]
btrfs: update comments at btrfs_record_unlink_dir() to be more clear

Update the comments at btrfs_record_unlink_dir() so that they mention
where new names are logged and where old names are removed. Also, while
at it make the width of the comments closer to 80 columns and capitalize
the sentences and finish them with punctuation.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use inode_logged() at btrfs_record_unlink_dir()
Filipe Manana [Wed, 17 May 2023 11:02:13 +0000 (12:02 +0100)]
btrfs: use inode_logged() at btrfs_record_unlink_dir()

At btrfs_record_unlink_dir() we directly check the logged_trans field of
the given inodes to check if they were previously logged in the current
transaction, and if any of them were, then we can avoid setting the field
last_unlink_trans of the directory to the id of the current transaction if
we are in a rename path. Avoiding that can later prevent falling back to
a transaction commit if anyone attempts to log the directory.

However the logged_trans field, store in struct btrfs_inode, is transient,
not persisted in the inode item on its subvolume b+tree, so that means
that if an inode is evicted and then loaded again, its original value is
lost and it's reset to 0. So directly checking the logged_trans field can
lead to some false negative, and that only results in a performance impact
as mentioned before.

Instead of directly checking the logged_trans field of the inodes, use the
inode_logged() helper, which will check in the log tree if an inode was
logged before in case its logged_trans field has a value of 0. This way
we can avoid setting the directory inode's last_unlink_trans and cause
future logging attempts of it to fallback to transaction commits. The
following test script shows one example where this happens without this
patch:

  $ cat test.sh
  #!/bin/bash

  DEV=/dev/nullb0
  MNT=/mnt/nullb0

  num_init_files=10000
  num_new_files=10000

  mkfs.btrfs -f $DEV
  mount -o ssd $DEV $MNT

  mkdir $MNT/testdir
  for ((i = 1; i <= $num_init_files; i++)); do
      echo -n > $MNT/testdir/file_$i
   done

  echo -n > $MNT/testdir/foo

  sync

  # Add some files so that there's more work in the transaction other
  # than just renaming file foo.
  for ((i = 1; i <= $num_new_files; i++)); do
      echo -n > $MNT/testdir/new_file_$i
  done

  # Change the file, fsync it.
  setfattr -n user.x1 -v 123 $MNT/testdir/foo
  xfs_io -c "fsync" $MNT/testdir/foo

  # Now triggger eviction of file foo but no eviction for our test
  # directory, since it is being used by the process below. This will
  # set logged_trans of the file's inode to 0 once it is loaded again.
  (
      cd $MNT/testdir
      while true; do
          :
      done
  ) &
  pid=$!

  echo 2 > /proc/sys/vm/drop_caches

  kill $pid
  wait $pid

  # Move foo out of our testdir. This will set last_unlink_trans
  # of the directory inode to the current transaction, because
  # logged_trans of both the directory and the file are set to 0.
  mv $MNT/testdir/foo $MNT/foo

  # Change file foo again and fsync it.
  # This fsync will result in a transaction commit because the rename
  # above has set last_unlink_trans of the parent directory to the id
  # of the current transaction and because our inode for file foo has
  # last_unlink_trans set to the current transaction, since it was
  # evicted and reloaded and it was previously modified in the current
  # transaction (the xattr addition).
  xfs_io -c "pwrite 0 64K" $MNT/foo
  start=$(date +%s%N)
  xfs_io -c "fsync" $MNT/foo
  end=$(date +%s%N)
  dur=$(( (end - start) / 1000000 ))

  echo "file fsync took: $dur milliseconds"

  umount $MNT

Before this patch:   fsync took 19 milliseconds
After this patch:    fsync took  5 milliseconds

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use inode_logged() at need_log_inode()
Filipe Manana [Wed, 17 May 2023 11:02:12 +0000 (12:02 +0100)]
btrfs: use inode_logged() at need_log_inode()

At need_log_inode() we directly check the ->logged_trans field of the
given inode to check if it was previously logged in the transaction, with
the goal of skipping logging the inode again when it's not necessary.
The ->logged_trans field in not persisted in the inode item or elsewhere,
it's only stored in memory (struct btrfs_inode), so it's transient and
lost once the inode is evicted and then loaded again. Once an inode is
loaded, we are conservative and set ->logged_trans to 0, which may mean
that either the inode was never logged in the current transaction or it
was logged but evicted before being loaded again.

Instead of checking the inode's ->logged_trans field directly, we can
use instead the helper inode_logged(), which will really check if the
inode was logged before in the current transaction in case we have a
->logged_trans field with a value of 0. This will prevent unnecessarily
logging an inode when it's not needed, and in some cases preventing a
transaction commit, in case the logging requires a fallback to a
transaction commit. The following test script shows a scenario where
due to eviction we fallback a transaction commit when trying to fsync
a file that was renamed:

  $ cat test.sh
  #!/bin/bash

  DEV=/dev/nullb0
  MNT=/mnt/nullb0

  num_init_files=10000
  num_new_files=10000

  mkfs.btrfs -f $DEV
  mount -o ssd $DEV $MNT

  mkdir $MNT/testdir
  for ((i = 1; i <= $num_init_files; i++)); do
      echo -n > $MNT/testdir/file_$i
  done

  echo -n > $MNT/testdir/foo

  sync

  # Add some files so that there's more work in the transaction other
  # than just renaming file foo.
  for ((i = 1; i <= $num_new_files; i++)); do
      echo -n > $MNT/testdir/new_file_$i
  done

  # Fsync the directory first.
  xfs_io -c "fsync" $MNT/testdir

  # Rename file foo.
  mv $MNT/testdir/foo $MNT/testdir/bar

  # Now trigger eviction of the test directory's inode.
  # Once loaded again, it will have logged_trans set to 0 and
  # last_unlink_trans set to the current transaction.
  echo 2 > /proc/sys/vm/drop_caches

  # Fsync file bar (ex-foo).
  # Before the patch the fsync would result in a transaction commit
  # because the inode for file bar has last_unlink_trans set to the
  # current transaction, so it will attempt to log the parent directory
  # as well, which will fallback to a full transaction commit because
  # it also has its last_unlink_trans set to the current transaction,
  # due to the inode eviction.
  start=$(date +%s%N)
  xfs_io -c "fsync" $MNT/testdir/bar
  end=$(date +%s%N)
  dur=$(( (end - start) / 1000000 ))

  echo "file fsync took: $dur milliseconds"

  umount $MNT

Before this patch:  fsync took 22 milliseconds
After this patch:   fsync took  8 milliseconds

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: scrub: remove more unused functions
Jiapeng Chong [Fri, 12 May 2023 05:44:57 +0000 (13:44 +0800)]
btrfs: scrub: remove more unused functions

These functions are defined in the scrub.c file, but last callers were
removed in e9255d6c4054 ("btrfs: scrub: remove the old scrub recheck
code").

fs/btrfs/scrub.c:553:20: warning: unused function 'scrub_stripe_index_and_offset'.
fs/btrfs/scrub.c:543:19: warning: unused function 'scrub_nr_raid_mirrors'.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=4937
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: handle tree backref walk error properly
Qu Wenruo [Thu, 11 May 2023 08:01:44 +0000 (16:01 +0800)]
btrfs: handle tree backref walk error properly

[BUG]
Smatch reports the following errors related to commit ("btrfs: output
affected files when relocation fails"):

fs/btrfs/inode.c:283 print_data_reloc_error()
error: uninitialized symbol 'ref_level'.

[CAUSE]
That part of code is mostly copied from scrub, but unfortunately scrub
code from the beginning is not doing the error handling properly.

The offending code looks like this:

do {
ret = tree_backref_for_extent();
btrfs_warn_rl();
} while (ret != 1);

There are several problems involved:

- No error handling
  If that tree_backref_for_extent() failed, we would output the same
  error again and again, never really exit as it requires ret == 1 to
  exit.

- Always do one extra output
  As tree_backref_for_extent() only return > 0 if there is no more
  backref item.
  This means after the last item we hit, we would output an invalid
  error message for ret > 0 case.

[FIX]
Fix the old code by:

- Move @ref_root and @ref_level into the if branch
  And do not initialize them, so we can catch such uninitialized values
  just like what we do in the inode.c

- Explicitly check the return value of tree_backref_for_extent()
  And handle ret < 0 and ret > 0 cases properly.

- No more do {} while () loop
  Instead go while (true) {} loop since we will handle @ret manually.

Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: don't hold an extra reference for redirtied buffers
Christoph Hellwig [Mon, 8 May 2023 14:58:39 +0000 (07:58 -0700)]
btrfs: don't hold an extra reference for redirtied buffers

When btrfs_redirty_list_add redirties a buffer, it also acquires
an extra reference that is released on transaction commit.  But
this is not required as buffers that are dirty or under writeback
are never freed (look for calls to extent_buffer_under_io())).

Remove the extra reference and the infrastructure used to drop it
again.

History behind redirty logic:

In the first place, it used releasing_list to hold all the
to-be-released extent buffers, and decided which buffers to re-dirty at
the commit time. Then, in a later version, the behaviour got changed to
re-dirty a necessary buffer and add re-dirtied one to the list in
btrfs_free_tree_block(). In short, the list was there mostly for the
patch series' historical reason.

Reviewed-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
[ add Naohiro's comment regarding history ]
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: fix dirty_metadata_bytes for redirtied buffers
Christoph Hellwig [Mon, 8 May 2023 14:58:38 +0000 (07:58 -0700)]
btrfs: fix dirty_metadata_bytes for redirtied buffers

dirty_metadata_bytes is decremented in both places that clear the dirty
bit in a buffer, but only incremented in btrfs_mark_buffer_dirty, which
means that a buffer that is redirtied using btrfs_redirty_list_add won't
be added to dirty_metadata_bytes, but it will be subtracted when written
out, leading an inconsistency in the counter.

Move the dirty_metadata_bytes from btrfs_mark_buffer_dirty into
set_extent_buffer_dirty to also account for the redirty case, and remove
the now unused set_extent_buffer_dirty return value.

Fixes: d3575156f662 ("btrfs: zoned: redirty released extent buffers")
CC: stable@vger.kernel.org # 5.15+
Reviewed-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: unexport btrfs_run_discard_work and make it static
Johannes Thumshirn [Tue, 9 May 2023 17:12:01 +0000 (10:12 -0700)]
btrfs: unexport btrfs_run_discard_work and make it static

Mark btrfs_run_discard_work static and move it above its callers.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: rename del_ptr to btrfs_del_ptr and export it
Josef Bacik [Sat, 29 Apr 2023 20:07:21 +0000 (16:07 -0400)]
btrfs: rename del_ptr to btrfs_del_ptr and export it

This exists internal to ctree.c, however btrfs check needs to use it for
some of its operations.  I'd rather not duplicate that code inside of
btrfs check as this is low level and I want to keep this code in one
place, so rename the function to btrfs_del_ptr and export it so that it
can be used inside of btrfs-progs safely.  Add a comment to make sure
this doesn't get removed by a future cleanup.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add a btrfs_csum_type_size helper
Josef Bacik [Sat, 29 Apr 2023 20:07:20 +0000 (16:07 -0400)]
btrfs: add a btrfs_csum_type_size helper

This is needed in btrfs-progs for the tools that convert the checksum
types for file systems and a few other things.  We don't have it in the
kernel as we just want to get the size for the super blocks type.
However I don't want to have to manually add this every time we sync
ctree.c into btrfs-progs, so add the helper in the kernel with a note so
it doesn't get removed by a later cleanup.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add __KERNEL__ check for btrfs_no_printk
Josef Bacik [Sat, 29 Apr 2023 20:07:19 +0000 (16:07 -0400)]
btrfs: add __KERNEL__ check for btrfs_no_printk

We want to override this in btrfs-progs, so wrap this in the __KERNEL__
check so we can easily sync this to btrfs-progs and have our local
version of btrfs_no_printk do the work.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: move split_flags/combine_flags helpers to inode-item.h
Josef Bacik [Sat, 29 Apr 2023 20:07:18 +0000 (16:07 -0400)]
btrfs: move split_flags/combine_flags helpers to inode-item.h

These are more related to the inode item flags on disk than the
in-memory btrfs_inode, move the helpers to inode-item.h.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: move btrfs_verify_level_key into tree-checker.c
Josef Bacik [Sat, 29 Apr 2023 20:07:17 +0000 (16:07 -0400)]
btrfs: move btrfs_verify_level_key into tree-checker.c

This is more a buffer validation helper, move it into the tree-checker
files where it makes more sense.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add __btrfs_check_node helper
Josef Bacik [Sat, 29 Apr 2023 20:07:16 +0000 (16:07 -0400)]
btrfs: add __btrfs_check_node helper

This helper returns a btrfs_tree_block_status for the various errors,
and then btrfs_check_node() will return -EUCLEAN if it gets anything
other than BTRFS_TREE_BLOCK_CLEAN which will be used by the kernel.  In
the future btrfs-progs will use this helper instead.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: extend btrfs_leaf_check to return btrfs_tree_block_status
Josef Bacik [Sat, 29 Apr 2023 20:07:15 +0000 (16:07 -0400)]
btrfs: extend btrfs_leaf_check to return btrfs_tree_block_status

Instead of blanket returning -EUCLEAN for all the failures in
btrfs_check_leaf, use btrfs_tree_block_status and return the appropriate
status for each failure.  Rename the helper to __btrfs_check_leaf and
then make a wrapper of btrfs_check_leaf that will return -EUCLEAN to
non-clean error codes.  This will allow us to have the
__btrfs_check_leaf variant in btrfs-progs while keeping the behavior in
the kernel consistent.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use btrfs_tree_block_status for leaf item errors
Josef Bacik [Sat, 29 Apr 2023 20:07:14 +0000 (16:07 -0400)]
btrfs: use btrfs_tree_block_status for leaf item errors

We have a variety of item specific errors that can occur.  For now
simply put these under the umbrella of BTRFS_TREE_BLOCK_INVALID_ITEM,
this can be fleshed out as we need in the future.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: add btrfs_tree_block_status definitions to tree-checker.h
Josef Bacik [Sat, 29 Apr 2023 20:07:13 +0000 (16:07 -0400)]
btrfs: add btrfs_tree_block_status definitions to tree-checker.h

We use this in btrfs-progs to determine if we can fix different types of
corruptions.  We don't care about this in the kernel, however it would
be good to share this code between the kernel and btrfs-progs, so add
the status definitions so we can start converting the tree-checker code
over to using these status flags instead of blanket returning -EUCLEAN.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: simplify btrfs_check_leaf_* helpers into a single helper
Josef Bacik [Sat, 29 Apr 2023 20:07:12 +0000 (16:07 -0400)]
btrfs: simplify btrfs_check_leaf_* helpers into a single helper

We have two helpers for checking leaves, because we have an extra check
for debugging in btrfs_mark_buffer_dirty(), and at that stage we may
have item data that isn't consistent yet.  However we can handle this
case internally in the helper, if BTRFS_HEADER_FLAG_WRITTEN is set we
know the buffer should be internally consistent, otherwise we need to
skip checking the item data.

Simplify this helper down a single helper and handle the item data
checking logic internally to the helper.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove level argument from btrfs_set_block_flags
Josef Bacik [Sat, 29 Apr 2023 20:07:11 +0000 (16:07 -0400)]
btrfs: remove level argument from btrfs_set_block_flags

We just pass in btrfs_header_level(eb) for the level, and we're passing
in the eb already, so simply get the level from the eb inside of
btrfs_set_block_flags.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: move btrfs_check_trunc_cache_free_space into block-rsv.c
Josef Bacik [Sat, 29 Apr 2023 20:07:10 +0000 (16:07 -0400)]
btrfs: move btrfs_check_trunc_cache_free_space into block-rsv.c

This is completely related to block rsv's, move it out of the free space
cache code and into block-rsv.c.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: scrub: use recovered data stripes as cache to avoid unnecessary read
Qu Wenruo [Thu, 13 Apr 2023 05:57:18 +0000 (13:57 +0800)]
btrfs: scrub: use recovered data stripes as cache to avoid unnecessary read

For P/Q stripe scrub, we have quite some duplicated read IO:

- Data stripes read for verification
  This is triggered by the scrub_submit_initial_read() inside
  scrub_raid56_parity_stripe().

- Data stripes read (again) for P/Q stripe verification
  This is triggered by scrub_assemble_read_bios() from scrub_rbio().

  Although we can have hit rbio cache and avoid unnecessary read, the
  chance is very low, as scrub would easily flush the whole rbio cache.

This means, even we're just scrubbing a single P/Q stripe, we would read
the data stripes twice for the best case scenario.  If we need to
recover some data stripes, it would cause more reads on the same data
stripes, again and again.

However before we call raid56_parity_submit_scrub_rbio() we already
have all data stripes repaired and their contents ready to use.
But RAID56 cache is unaware about the scrub cache, thus RAID56 layer
itself still needs to re-read the data stripes.

To avoid such cache miss, this patch would:

- Introduce a new helper, raid56_parity_cache_data_pages()
  This function would grab the pages from an array, and copy the content
  to the rbio, marking all the involved sectors uptodate.

  The page copy is unavoidable because of the cache pages of rbio are all
  self managed, thus can not utilize outside pages without screwing up
  the lifespan.

- Use the repaired data stripes as cache inside
  scrub_raid56_parity_stripe()

By this, we ensure all the data sectors of the scrub rbio are already
uptodate, and no need to read them again from disk.

Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: assert tree lock is held when removing free space entries
Filipe Manana [Thu, 4 May 2023 11:04:26 +0000 (12:04 +0100)]
btrfs: assert tree lock is held when removing free space entries

Removing a free space entry from an in memory space cache requires having
the corresponding btrfs_free_space_ctl's 'tree_lock' held. We have several
code paths that remove an entry, so add assertions where appropriate to
verify we are holding the lock, as the lock is acquired by some other
function up in the call chain, which makes it easy to miss in the future.

Note: for this to work we need to lock the local btrfs_free_space_ctl at
load_free_space_cache(), which was not being done because it's local,
declared on the stack, so no other task has access to it.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: assert tree lock is held when linking free space
Filipe Manana [Thu, 4 May 2023 11:04:25 +0000 (12:04 +0100)]
btrfs: assert tree lock is held when linking free space

When linking a free space entry, at link_free_space(), the caller should
be holding the spinlock 'tree_lock' of the given btrfs_free_space_ctl
argument, which is necessary for manipulating the red black tree of free
space entries (done by tree_insert_offset(), which already asserts the
lock is held) and for manipulating the 'free_space', 'free_extents',
'discardable_extents' and 'discardable_bytes' counters of the given
struct btrfs_free_space_ctl.

So assert that the spinlock 'tree_lock' of the given btrfs_free_space_ctl
is held by the current task. We have multiple code paths that end up
calling link_free_space(), and all currently take the lock before calling
it.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: assert tree lock is held when searching for free space entries
Filipe Manana [Thu, 4 May 2023 11:04:24 +0000 (12:04 +0100)]
btrfs: assert tree lock is held when searching for free space entries

When searching for a free space entry by offset, at tree_search_offset(),
we are supposed to have the btrfs_free_space_ctl's 'tree_lock' held, so
assert that. We have multiple callers of tree_search_offset(), and all
currently hold the necessary lock before calling it.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: assert proper locks are held at tree_insert_offset()
Filipe Manana [Thu, 4 May 2023 11:04:23 +0000 (12:04 +0100)]
btrfs: assert proper locks are held at tree_insert_offset()

There are multiple code paths leading to tree_insert_offset(), and each
path takes the necessary locks before tree_insert_offset() is called,
since they do other things that require those locks to be held. This makes
it easy to miss the locking somewhere, so make tree_insert_offset() assert
that the required locks are being held by the calling task.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: simplify arguments to tree_insert_offset()
Filipe Manana [Thu, 4 May 2023 11:04:22 +0000 (12:04 +0100)]
btrfs: simplify arguments to tree_insert_offset()

For the in-memory component of space caching (free space cache and free
space tree), three of the arguments passed to tree_insert_offset() can
always be taken from the new free space entry that we are about to add.

So simplify tree_insert_offset() to take the new entry instead of the
'offset', 'node' and 'bitmap' arguments. This will also allow to make
further changes simpler.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use precomputed end offsets at do_trimming()
Filipe Manana [Thu, 4 May 2023 11:04:21 +0000 (12:04 +0100)]
btrfs: use precomputed end offsets at do_trimming()

The are two computations of end offsets at do_trimming() that are not
necessary, as they were previously computed and stored in local const
variables. So just use the variables instead, to make the source code
shorter and easier to read.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: avoid searching twice for previous node when merging free space entries
Filipe Manana [Thu, 4 May 2023 11:04:20 +0000 (12:04 +0100)]
btrfs: avoid searching twice for previous node when merging free space entries

At try_merge_free_space(), avoid calling twice rb_prev() to find the
previous node, as that requires looping through the red black tree, so
store the result of the rb_prev() call and then use it.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: avoid extra memory allocation when copying free space cache
Filipe Manana [Thu, 4 May 2023 11:04:19 +0000 (12:04 +0100)]
btrfs: avoid extra memory allocation when copying free space cache

At copy_free_space_cache(), we add a new entry to the block group's ctl
before we free the entry from the temporary ctl. Adding a new entry
requires the allocation of a new struct btrfs_free_space, so we can
avoid a temporary extra allocation by freeing the entry from the
temporary ctl before we add a new entry to the main ctl, which possibly
also reduces the chances for a memory allocation failure in case of very
high memory pressure. So just do that.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: simplify transid initialization in btrfs_ioctl_wait_sync
Tom Rix [Tue, 2 May 2023 14:51:29 +0000 (10:51 -0400)]
btrfs: simplify transid initialization in btrfs_ioctl_wait_sync

A small code simplification, move the default value of transid to its
initialization and remove the else-statement.

Signed-off-by: Tom Rix <trix@redhat.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: output affected files when relocation fails
Qu Wenruo [Wed, 3 May 2023 04:40:01 +0000 (12:40 +0800)]
btrfs: output affected files when relocation fails

[PROBLEM]
When relocation fails (mostly due to checksum mismatch), we only got
very cryptic error messages like:

  BTRFS info (device dm-4): relocating block group 13631488 flags data
  BTRFS warning (device dm-4): csum failed root -9 ino 257 off 0 csum 0x373e1ae3 expected csum 0x98757625 mirror 1
  BTRFS error (device dm-4): bdev /dev/mapper/test-scratch1 errs: wr 0, rd 0, flush 0, corrupt 1, gen 0
  BTRFS info (device dm-4): balance: ended with status: -5

The end user has to decipher the above messages and use various tools to
locate the affected files and find a way to fix the problem (mostly
deleting the file).  This is not an easy work even for experienced
developer, not to mention the end users.

[SCRUB IS DOING BETTER]
By contrast, scrub is providing much better error messages:

  BTRFS error (device dm-4): unable to fixup (regular) error at logical 13631488 on dev /dev/mapper/test-scratch1 physical 13631488
  BTRFS warning (device dm-4): checksum error at logical 13631488 on dev /dev/mapper/test-scratch1, physical 13631488, root 5, inode 257, offset 0, length 4096, links 1 (path: file)
  BTRFS info (device dm-4): scrub: finished on devid 1 with status: 0

Which provides the affected files directly to the end user.

[IMPROVEMENT]
Instead of the generic data checksum error messages, which is not doing
a good job for data reloc inodes, this patch introduce a scrub like
backref walking based solution.

When a sector fails its checksum for data reloc inode, we go the
following workflow:

- Get the real logical bytenr
  For data reloc inode, the file offset is the offset inside the block
  group.
  Thus the real logical bytenr is @file_off + @block_group->start.

- Do an extent type check
  If it's tree blocks it's much easier to handle, just go through
  all the tree block backref.

- Do a backref walk and inode path resolution for data extents
  This is mostly the same as scrub.
  But unfortunately we can not reuse the same function as the output
  format is different.

Now the new output would be more user friendly:

  BTRFS info (device dm-4): relocating block group 13631488 flags data
  BTRFS warning (device dm-4): csum failed root -9 ino 257 off 0 logical 13631488 csum 0x373e1ae3 expected csum 0x98757625 mirror 1
  BTRFS warning (device dm-4): checksum error at logical 13631488 mirror 1 root 5 inode 257 offset 0 length 4096 links 1 (path: file)
  BTRFS error (device dm-4): bdev /dev/mapper/test-scratch1 errs: wr 0, rd 0, flush 0, corrupt 2, gen 0
  BTRFS info (device dm-4): balance: ended with status: -5

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: remove hipri_workers workqueue
Christoph Hellwig [Wed, 3 May 2023 07:06:15 +0000 (09:06 +0200)]
btrfs: remove hipri_workers workqueue

Now that btrfs_wq_submit_bio is never called for synchronous I/O,
the hipri_workers workqueue is not used anymore and can be removed.

Reviewed-by: Chris Mason <clm@fb.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: determine synchronous writers from bio or writeback control
Christoph Hellwig [Wed, 3 May 2023 07:06:14 +0000 (09:06 +0200)]
btrfs: determine synchronous writers from bio or writeback control

The writeback_control structure already passes down the information about
a writeback being synchronous from the core VM code, and thus information
is propagated into the bio REQ_SYNC flag through the wbc_to_write_flags
helper.

Use that information to decide if checksums calculation is offloaded to
a workqueue instead of btrfs_inode::sync_writers field that not only
bloats the inode but also has too wide scope, being inode wide instead
of limited to the actual writeback request.

The sync writes were set in:

- btrfs_do_write_iter - regular IO, sync status is set
- start_ordered_ops - ordered write start, writeback with WB_SYNC_ALL
  mode
- btrfs_write_marked_extents - write marked extents, writeback with
  WB_SYNC_ALL mode

Reviewed-by: Chris Mason <clm@fb.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: submit IO synchronously for fast checksum implementations
Christoph Hellwig [Wed, 3 May 2023 07:06:13 +0000 (09:06 +0200)]
btrfs: submit IO synchronously for fast checksum implementations

Most modern hardware supports very fast accelerated crc32c calculation.
If that is supported the CPU overhead of the checksum calculation is
very limited, and offloading the calculation to special worker threads
has a lot of overhead for no gain.

E.g. on an Intel Optane device is actually very much slows down even
1M buffered writes with fio:

Unpatched:

write: IOPS=3316, BW=3316MiB/s (3477MB/s)(200GiB/61757msec); 0 zone resets

With synchronous CRCs:

write: IOPS=4882, BW=4882MiB/s (5119MB/s)(200GiB/41948msec); 0 zone resets

With a lot of variation during the unpatched run going down as low as
1100MB/s, while the synchronous CRC version has about the same peak write
speed but much lower dips, and fewer kworkers churning around.
Both tests had fio saturated at 100% CPU.

(thanks to Jens Axboe via Chris Mason for the benchmarking)

Reviewed-by: Chris Mason <clm@fb.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use SECTOR_SHIFT to convert LBA to physical offset
Anand Jain [Sat, 15 Apr 2023 11:32:38 +0000 (19:32 +0800)]
btrfs: use SECTOR_SHIFT to convert LBA to physical offset

Using SECTOR_SHIFT to convert LBA to physical address makes it more
readable.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
17 months agobtrfs: use SECTOR_SHIFT to convert physical offset to LBA
Anand Jain [Sat, 15 Apr 2023 09:51:23 +0000 (17:51 +0800)]
btrfs: use SECTOR_SHIFT to convert physical offset to LBA

Use SECTOR_SHIFT while converting a physical address to an LBA, makes
it more readable.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>