GitHub/moto-9609/android_kernel_motorola_exynos9610.git
7 years agobtrfs: scrub: fix target device intialization while setting up scrub context
David Sterba [Mon, 26 Jun 2017 13:19:00 +0000 (15:19 +0200)]
btrfs: scrub: fix target device intialization while setting up scrub context

The commit "btrfs: scrub: inline helper scrub_setup_wr_ctx" inlined a
helper but wrongly sets up the target device. Incidentally there's a
local variable with the same name as a parameter in the previous
function, so this got caught during runtime as crash in test btrfs/027.

Reported-by: Chris Mason <clm@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Fix qgroup reserved space underflow by only freeing reserved ranges
Qu Wenruo [Mon, 27 Feb 2017 07:10:39 +0000 (15:10 +0800)]
btrfs: qgroup: Fix qgroup reserved space underflow by only freeing reserved ranges

[BUG]
For the following case, btrfs can underflow qgroup reserved space
at an error path:
(Page size 4K, function name without "btrfs_" prefix)

         Task A                  |             Task B
----------------------------------------------------------------------
Buffered_write [0, 2K)           |
|- check_data_free_space()       |
|  |- qgroup_reserve_data()      |
|     Range aligned to page      |
|     range [0, 4K)          <<< |
|     4K bytes reserved      <<< |
|- copy pages to page cache      |
                                 | Buffered_write [2K, 4K)
                                 | |- check_data_free_space()
                                 | |  |- qgroup_reserved_data()
                                 | |     Range alinged to page
                                 | |     range [0, 4K)
                                 | |     Already reserved by A <<<
                                 | |     0 bytes reserved      <<<
                                 | |- delalloc_reserve_metadata()
                                 | |  And it *FAILED* (Maybe EQUOTA)
                                 | |- free_reserved_data_space()
                                      |- qgroup_free_data()
                                         Range aligned to page range
                                         [0, 4K)
                                         Freeing 4K
(Special thanks to Chandan for the detailed report and analyse)

[CAUSE]
Above Task B is freeing reserved data range [0, 4K) which is actually
reserved by Task A.

And at writeback time, page dirty by Task A will go through writeback
routine, which will free 4K reserved data space at file extent insert
time, causing the qgroup underflow.

[FIX]
For btrfs_qgroup_free_data(), add @reserved parameter to only free
data ranges reserved by previous btrfs_qgroup_reserve_data().
So in above case, Task B will try to free 0 byte, so no underflow.

Reported-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Tested-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Introduce extent changeset for qgroup reserve functions
Qu Wenruo [Mon, 27 Feb 2017 07:10:38 +0000 (15:10 +0800)]
btrfs: qgroup: Introduce extent changeset for qgroup reserve functions

Introduce a new parameter, struct extent_changeset for
btrfs_qgroup_reserved_data() and its callers.

Such extent_changeset was used in btrfs_qgroup_reserve_data() to record
which range it reserved in current reserve, so it can free it in error
paths.

The reason we need to export it to callers is, at buffered write error
path, without knowing what exactly which range we reserved in current
allocation, we can free space which is not reserved by us.

This will lead to qgroup reserved space underflow.

Reviewed-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Fix qgroup reserved space underflow caused by buffered write and quota...
Qu Wenruo [Mon, 27 Feb 2017 07:10:37 +0000 (15:10 +0800)]
btrfs: qgroup: Fix qgroup reserved space underflow caused by buffered write and quotas being enabled

[BUG]
Under the following case, we can underflow qgroup reserved space.

            Task A                |            Task B
---------------------------------------------------------------
 Quota disabled                   |
 Buffered write                   |
 |- btrfs_check_data_free_space() |
 |  *NO* qgroup space is reserved |
 |  since quota is *DISABLED*     |
 |- All pages are copied to page  |
    cache                         |
                                  | Enable quota
                                  | Quota scan finished
                                  |
                                  | Sync_fs
                                  | |- run_delalloc_range
                                  | |- Write pages
                                  | |- btrfs_finish_ordered_io
                                  |    |- insert_reserved_file_extent
                                  |       |- btrfs_qgroup_release_data()
                                  |          Since no qgroup space is
                                             reserved in Task A, we
                                             underflow qgroup reserved
                                             space
This can be detected by fstest btrfs/104.

[CAUSE]
In insert_reserved_file_extent() we tell qgroup to release the @ram_bytes
size of qgroup reserved_space in all cases.
And btrfs_qgroup_release_data() will check if quotas are enabled.

However in the above case, the buffered write happens before quota is
enabled, so we don't have the reserved space for that range.

[FIX]
In insert_reserved_file_extent(), we tell qgroup to release the acctual
byte number it released.
In the above case, since we don't have the reserved space, we tell
qgroups to release 0 byte, so the problem can be fixed.

And thanks to the @reserved parameter introduced by the qgroup rework,
and previous patch to return released bytes, the fix can be as small as
10 lines.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
[ changelog updates ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Return actually freed bytes for qgroup release or free data
Qu Wenruo [Mon, 27 Feb 2017 07:10:36 +0000 (15:10 +0800)]
btrfs: qgroup: Return actually freed bytes for qgroup release or free data

btrfs_qgroup_release/free_data() only returns 0 or a negative error
number (ENOMEM is the only possible error).

This is normally good enough, but sometimes we need the exact byte
count it freed/released.

Change it to return actually released/freed bytenr number instead of 0
for success.
And slightly modify related extent_changeset structure, since in btrfs
one no-hole data extent won't be larger than 128M, so "unsigned int"
is large enough for the use case.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Cleanup btrfs_qgroup_prepare_account_extents function
Qu Wenruo [Mon, 27 Feb 2017 07:10:35 +0000 (15:10 +0800)]
btrfs: qgroup: Cleanup btrfs_qgroup_prepare_account_extents function

Quite a lot of qgroup corruption happens due to wrong time of calling
btrfs_qgroup_prepare_account_extents().

Since the safest time is to call it just before
btrfs_qgroup_account_extents(), there is no need to separate these 2
functions.

Merging them will make code cleaner and less bug prone.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
[ changelog and comment adjustments ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: qgroup: Add quick exit for non-fs extents
Qu Wenruo [Mon, 27 Feb 2017 07:10:34 +0000 (15:10 +0800)]
btrfs: qgroup: Add quick exit for non-fs extents

Modify btrfs_qgroup_account_extent() to exit quicker for non-fs extents.

The quick exit condition is:
1) The extent belongs to a non-fs tree
   Only fs-tree extents can affect qgroup numbers and is the only case
   where extent can be shared between different trees.

   Although strictly speaking extent in data-reloc or tree-reloc tree
   can be shared, data/tree-reloc root won't appear in the result of
   btrfs_find_all_roots(), so we can ignore such case.

   So we can check the first root in old_roots/new_roots ulist.
   - if we find the 1st root is a not a fs/subvol root, then we can skip
     the extent
   - if we find the 1st root is a fs/subvol root, then we must continue
     calculation

OR

2) both 'nr_old_roots' and 'nr_new_roots' are 0
   This means either such extent got allocated then freed in current
   transaction or it's a new reloc tree extent, whose nr_new_roots is 0.
   Either way it won't affect qgroup accounting and can be skipped
   safely.

Such quick exit can make trace output more quite and less confusing:
(example with fs uuid and time stamp removed)

Before:
------
add_delayed_tree_ref: bytenr=29556736 num_bytes=16384 action=ADD_DELAYED_REF parent=0(-) ref_root=2(EXTENT_TREE) level=0 type=TREE_BLOCK_REF seq=0
btrfs_qgroup_account_extent: bytenr=29556736 num_bytes=16384 nr_old_roots=0 nr_new_roots=1
------
Extent tree block will trigger btrfs_qgroup_account_extent() trace point
while no qgroup number is changed, as extent tree won't affect qgroup
accounting.

After:
------
add_delayed_tree_ref: bytenr=29556736 num_bytes=16384 action=ADD_DELAYED_REF parent=0(-) ref_root=2(EXTENT_TREE) level=0 type=TREE_BLOCK_REF seq=0
------
Now such unrelated extent won't trigger btrfs_qgroup_account_extent()
trace point, making the trace less noisy.

Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
[ changelog and comment adjustments ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: rework delayed ref total_bytes_pinned accounting
Omar Sandoval [Tue, 6 Jun 2017 23:45:31 +0000 (16:45 -0700)]
Btrfs: rework delayed ref total_bytes_pinned accounting

The total_bytes_pinned counter is completely broken when accounting
delayed refs:

- If two drops for the same extent are merged, we will decrement
  total_bytes_pinned twice but only increment it once.
- If an add is merged into a drop or vice versa, we will decrement the
  total_bytes_pinned counter but never increment it.
- If multiple references to an extent are dropped, we will account it
  multiple times, potentially vastly over-estimating the number of bytes
  that will be freed by a commit and doing unnecessary work when we're
  close to ENOSPC.

The last issue is relatively minor, but the first two make the
total_bytes_pinned counter leak or underflow very often. These
accounting issues were introduced in b150a4f10d87 ("Btrfs: use a percpu
to keep track of possibly pinned bytes"), but they were papered over by
zeroing out the counter on every commit until d288db5dc011 ("Btrfs: fix
race of using total_bytes_pinned").

We need to make sure that an extent is accounted as pinned exactly once
if and only if we will drop references to it when when the transaction
is committed. Ideally we would only add to total_bytes_pinned when the
*last* reference is dropped, but this information isn't readily
available for data extents. Again, this over-estimation can lead to
extra commits when we're close to ENOSPC, but it's not as bad as before.

The fix implemented here is to increment total_bytes_pinned when the
total refmod count for an extent goes negative and decrement it if the
refmod count goes back to non-negative or after we've run all of the
delayed refs for that extent.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: return old and new total ref mods when adding delayed refs
Omar Sandoval [Tue, 6 Jun 2017 23:45:30 +0000 (16:45 -0700)]
Btrfs: return old and new total ref mods when adding delayed refs

We need this to decide when to account pinned bytes.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: always account pinned bytes when dropping a tree block ref
Omar Sandoval [Tue, 6 Jun 2017 23:45:29 +0000 (16:45 -0700)]
Btrfs: always account pinned bytes when dropping a tree block ref

Currently, we only increment total_bytes_pinned in
btrfs_free_tree_block() when dropping the last reference on the block.
However, when the delayed ref is run later, we will decrement
total_bytes_pinned regardless of whether it was the last reference or
not. This causes the counter to underflow when the reference we dropped
was not the last reference. Fix it by incrementing the counter
unconditionally, which is what btrfs_free_extent() does. This makes
total_bytes_pinned an overestimate when references to shared extents are
dropped, but in the worst case this will just make us try to commit the
transaction to try to free up space and find we didn't free enough.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: update total_bytes_pinned when pinning down extents
Omar Sandoval [Tue, 6 Jun 2017 23:45:28 +0000 (16:45 -0700)]
Btrfs: update total_bytes_pinned when pinning down extents

The extents marked in pin_down_extent() will be unpinned later in
unpin_extent_range(), which decrements total_bytes_pinned.
pin_down_extent() must increment the counter to avoid underflowing it.
Also adjust btrfs_free_tree_block() to avoid accounting for the same
extent twice.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: make BUG_ON() in add_pinned_bytes() an ASSERT()
Omar Sandoval [Tue, 6 Jun 2017 23:45:27 +0000 (16:45 -0700)]
Btrfs: make BUG_ON() in add_pinned_bytes() an ASSERT()

The value of flags is one of DATA/METADATA/SYSTEM, they must exist at
when add_pinned_bytes is called.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ added changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: make add_pinned_bytes() take an s64 num_bytes instead of u64
Omar Sandoval [Tue, 6 Jun 2017 23:45:26 +0000 (16:45 -0700)]
Btrfs: make add_pinned_bytes() take an s64 num_bytes instead of u64

There are a few places where we pass in a negative num_bytes, so make it
signed for clarity. Also move it up in the file since later patches will
need it there.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: fix validation of XATTR_ITEM dir items
David Sterba [Wed, 21 Jun 2017 15:43:24 +0000 (17:43 +0200)]
btrfs: fix validation of XATTR_ITEM dir items

The XATTR_ITEM is a type of a directory item so we use the common
validator helper. Unlike other dir items, it can have data. The way the
name len validation is currently implemented does not reflect that. We'd
have to adjust by the data_len when comparing the read and item limits.

However, this will not work for multi-item xattr dir items.

Example from tree dump of generic/337:

        item 7 key (257 XATTR_ITEM 751495445) itemoff 15667 itemsize 147
                location key (0 UNKNOWN.0 0) type XATTR
                transid 8 data_len 3 name_len 11
                name: user.foobar
                data 123
                location key (0 UNKNOWN.0 0) type XATTR
                transid 8 data_len 6 name_len 13
                name: user.WvG1c1Td
                data qwerty
                location key (0 UNKNOWN.0 0) type XATTR
                transid 8 data_len 5 name_len 19
                name: user.J3__T_Km3dVsW_
                data hello

At the point of btrfs_is_name_len_valid call we don't have access to the
data_len value of the 2nd and 3rd sub-item. So simple btrfs_dir_data_len(leaf,
di) would always return 3, although we'd need to get 6 and 5 respectively to
get the claculations right. (read_end + name_len + data_len vs item_end)

We'd have to also pass data_len externally, which is not point of the
name validation. The last check is supposed to test if there's at least
one dir item space after the one we're processing. I don't think this is
particularly useful, validation of the next item would catch that too.
So the check is removed and we don't weaken the validation. Now tests
btrfs/048, btrfs/053, generic/273 and generic/337 pass.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Verify dir_item in iterate_object_props
Su Yue [Tue, 6 Jun 2017 09:57:08 +0000 (17:57 +0800)]
btrfs: Verify dir_item in iterate_object_props

Call verify_dir_item before memcmp_extent_buffer reading name from
dir_item.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len before in btrfs_del_root_ref
Su Yue [Tue, 6 Jun 2017 09:57:07 +0000 (17:57 +0800)]
btrfs: Check name_len before in btrfs_del_root_ref

btrfs_del_root_ref calls btrfs_search_slot and reads name from root_ref.
Call btrfs_is_name_len_valid before memcmp.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len before reading btrfs_get_name
Su Yue [Tue, 6 Jun 2017 09:57:06 +0000 (17:57 +0800)]
btrfs: Check name_len before reading btrfs_get_name

In btrfs_get_name, there's btrfs_search_slot and reads name from
inode_ref/root_ref.

Call btrfs_is_name_len_valid in btrfs_get_name.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len before read in iterate_dir_item
Su Yue [Tue, 6 Jun 2017 09:57:05 +0000 (17:57 +0800)]
btrfs: Check name_len before read in iterate_dir_item

Since iterate_dir_item checks name_len in its own way,
so use btrfs_is_name_len_valid not 'verify_dir_item' to make more strict
name_len check.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ switched ENAMETOOLONG to EIO ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len in btrfs_check_ref_name_override
Su Yue [Tue, 6 Jun 2017 09:57:04 +0000 (17:57 +0800)]
btrfs: Check name_len in btrfs_check_ref_name_override

In btrfs_log_inode, btrfs_search_forward gets the buffer and then
btrfs_check_ref_name_override will read name from ref/extref for the
first time.

Call btrfs_is_name_len_valid before reading name.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Verify dir_item in replay_xattr_deletes
Su Yue [Tue, 6 Jun 2017 09:57:03 +0000 (17:57 +0800)]
btrfs: Verify dir_item in replay_xattr_deletes

replay_xattr_deletes calls btrfs_search_slot to get buffer and reads
name.

Call verify_dir_item to check name_len in replay_xattr_deletes to avoid
reading out of boundary.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len on add_inode_ref call path
Su Yue [Tue, 6 Jun 2017 09:57:02 +0000 (17:57 +0800)]
btrfs: Check name_len on add_inode_ref call path

replay_one_buffer first reads buffers and dispatches items accroding to
the item type.
In this patch, add_inode_ref handles inode_ref and inode_extref.
Then add_inode_ref calls ref_get_fields and extref_get_fields to read
ref/extref name for the first time.
So checking name_len before reading those two is fine.

add_inode_ref also calls inode_in_dir to match ref/extref in parent_dir.
The call graph includes btrfs_match_dir_item_name to read dir_item name
in the parent dir.
Checking first dir_item is not enough. Change it to verify every
dir_item while doing matches.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Check name_len with boundary in verify dir_item
Su Yue [Tue, 6 Jun 2017 09:57:01 +0000 (17:57 +0800)]
btrfs: Check name_len with boundary in verify dir_item

Originally, verify_dir_item verifies name_len of dir_item with fixed
values but not item boundary.
If corrupted name_len was not bigger than the fixed value, for example
255, the function will think the dir_item is fine. And then reading
beyond boundary will cause crash.

Example:
1. Corrupt one dir_item name_len to be 255.
        2. Run 'ls -lar /mnt/test/ > /dev/null'
dmesg:
[   48.451449] BTRFS info (device vdb1): disk space caching is enabled
[   48.451453] BTRFS info (device vdb1): has skinny extents
[   48.489420] general protection fault: 0000 [#1] SMP
[   48.489571] Modules linked in: ext4 jbd2 mbcache btrfs xor raid6_pq
[   48.489716] CPU: 1 PID: 2710 Comm: ls Not tainted 4.10.0-rc1 #5
[   48.489853] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-20170228_101828-anatol 04/01/2014
[   48.490008] task: ffff880035df1bc0 task.stack: ffffc90004800000
[   48.490008] RIP: 0010:read_extent_buffer+0xd2/0x190 [btrfs]
[   48.490008] RSP: 0018:ffffc90004803d98 EFLAGS: 00010202
[   48.490008] RAX: 000000000000001b RBX: 000000000000001b RCX: 0000000000000000
[   48.490008] RDX: ffff880079dbf36c RSI: 0005080000000000 RDI: ffff880079dbf368
[   48.490008] RBP: ffffc90004803dc8 R08: ffff880078e8cc48 R09: ffff880000000000
[   48.490008] R10: 0000160000000000 R11: 0000000000001000 R12: ffff880079dbf288
[   48.490008] R13: ffff880078e8ca88 R14: 0000000000000003 R15: ffffc90004803e20
[   48.490008] FS:  00007fef50c60800(0000) GS:ffff88007d400000(0000) knlGS:0000000000000000
[   48.490008] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   48.490008] CR2: 000055f335ac2ff8 CR3: 000000007356d000 CR4: 00000000001406e0
[   48.490008] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   48.490008] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   48.490008] Call Trace:
[   48.490008]  btrfs_real_readdir+0x3b7/0x4a0 [btrfs]
[   48.490008]  iterate_dir+0x181/0x1b0
[   48.490008]  SyS_getdents+0xa7/0x150
[   48.490008]  ? fillonedir+0x150/0x150
[   48.490008]  entry_SYSCALL_64_fastpath+0x18/0xad
[   48.490008] RIP: 0033:0x7fef5032546b
[   48.490008] RSP: 002b:00007ffeafcdb830 EFLAGS: 00000206 ORIG_RAX: 000000000000004e
[   48.490008] RAX: ffffffffffffffda RBX: 00007fef5061db38 RCX: 00007fef5032546b
[   48.490008] RDX: 0000000000008000 RSI: 000055f335abaff0 RDI: 0000000000000003
[   48.490008] RBP: 00007fef5061dae0 R08: 00007fef5061db48 R09: 0000000000000000
[   48.490008] R10: 000055f335abafc0 R11: 0000000000000206 R12: 00007fef5061db38
[   48.490008] R13: 0000000000008040 R14: 00007fef5061db38 R15: 000000000000270e
[   48.490008] RIP: read_extent_buffer+0xd2/0x190 [btrfs] RSP: ffffc90004803d98
[   48.499455] ---[ end trace 321920d8e8339505 ]---

Fix it by adding a parameter @slot and check name_len with item boundary
by calling btrfs_is_name_len_valid.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
rev
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Introduce btrfs_is_name_len_valid to avoid reading beyond boundary
Su Yue [Tue, 6 Jun 2017 09:57:00 +0000 (17:57 +0800)]
btrfs: Introduce btrfs_is_name_len_valid to avoid reading beyond boundary

Introduce function btrfs_is_name_len_valid.

The function compares parameter @name_len with item boundary then
returns true if name_len is valid.

Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ s/btrfs_leaf_data/BTRFS_LEAF_DATA_OFFSET/ ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: move dev stats accounting out of wait_dev_flush
David Sterba [Thu, 15 Jun 2017 14:20:43 +0000 (16:20 +0200)]
btrfs: move dev stats accounting out of wait_dev_flush

We should really just wait in wait_dev_flush and let the caller decide
what to do with the error value.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: account as waiting for IO, while waiting fot the flush bio completion
David Sterba [Thu, 15 Jun 2017 14:04:26 +0000 (16:04 +0200)]
btrfs: account as waiting for IO, while waiting fot the flush bio completion

Similar to what submit_bio_wait does, we should account for IO while
waiting for a bio completion. This has marginal visible effects, flush
bio is short-lived.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: preallocate device flush bio
David Sterba [Tue, 6 Jun 2017 15:06:06 +0000 (17:06 +0200)]
btrfs: preallocate device flush bio

For devices that support flushing, we allocate a bio, submit, wait for
it and then free it. The bio allocation does not fail so ENOMEM is not a
problem but we still may unnecessarily stress the allocation subsystem.

Instead, we can allocate the bio at the same time we allocate the device
and reuse it each time we need to flush the barriers. The bio is reset
before each use. Reference counting is simplified to just device
allocation (get) and freeing (put).

The bio used to be submitted through the integrity checker which will
find out that bio has no data attached and call submit_bio.

Status of the bio in flight needs to be tracked separately in case the
device caches get switched off between write and wait.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: incremental send, fix invalid path for unlink commands
Filipe Manana [Tue, 13 Jun 2017 13:13:11 +0000 (14:13 +0100)]
Btrfs: incremental send, fix invalid path for unlink commands

An incremental send can contain unlink operations with an invalid target
path when we rename some directory inode A, then rename some file inode B
to the old name of inode A and directory inode A is an ancestor of inode B
in the parent snapshot (but not anymore in the send snapshot).

Consider the following example scenario where this issue happens.

Parent snapshot:

 .                                                      (ino 256)
 |
 |--- dir1/                                             (ino 257)
       |--- dir2/                                       (ino 258)
       |     |--- file1                                 (ino 259)
       |     |--- file3                                 (ino 261)
       |
       |--- dir3/                                       (ino 262)
             |--- file22                                (ino 260)
             |--- dir4/                                 (ino 263)

Send snapshot:

 .                                                      (ino 256)
 |
 |--- dir1/                                             (ino 257)
       |--- dir2/                                       (ino 258)
       |--- dir3                                        (ino 260)
       |--- file3/                                      (ino 262)
             |--- dir4/                                 (ino 263)
                   |--- file11                          (ino 269)
                   |--- file33                          (ino 261)

When attempting to apply the corresponding incremental send stream, an
unlink operation contains an invalid path which makes the receiver fail.
The following is verbose output of the btrfs receive command:

 receiving snapshot snap2 uuid=7d5450da-a573-e043-a451-ec85f4879f0f (...)
 utimes
 utimes dir1
 utimes dir1/dir2
 link dir1/dir3/dir4/file11 -> dir1/dir2/file1
 unlink dir1/dir2/file1
 utimes dir1/dir2
 truncate dir1/dir3/dir4/file11 size=0
 utimes dir1/dir3/dir4/file11
 rename dir1/dir3 -> o262-7-0
 link dir1/dir3 -> o262-7-0/file22
 unlink dir1/dir3/file22
 ERROR: unlink dir1/dir3/file22 failed. Not a directory

The following steps happen during the computation of the incremental send
stream the lead to this issue:

1) Before we start processing the new and deleted references for inode
   260, we compute the full path of the deleted reference
   ("dir1/dir3/file22") and cache it in the list of deleted references
   for our inode.

2) We then start processing the new references for inode 260, for which
   there is only one new, located at "dir1/dir3". When processing this
   new reference, we check that inode 262, which was not yet processed,
   collides with the new reference and because of that we orphanize
   inode 262 so its new full path becomes "o262-7-0".

3) After the orphanization of inode 262, we create the new reference for
   inode 260 by issuing a link command with a target path of "dir1/dir3"
   and a source path of "o262-7-0/file22".

4) We then start processing the deleted references for inode 260, for
   which there is only one with the base name of "file22", and issue
   an unlink operation containing the target path computed at step 1,
   which is wrong because that path no longer exists and should be
   replaced with "o262-7-0/file22".

So fix this issue by recomputing the full path of deleted references if
when we processed the new references for an inode we ended up orphanizing
any other inode that is an ancestor of our inode in the parent snapshot.

A test case for fstests follows soon.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
[ adjusted after prev patch removed fs_path::dir_path and dir_path_len ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: send, fix invalid path after renaming and linking file
Filipe Manana [Wed, 7 Jun 2017 10:41:29 +0000 (11:41 +0100)]
Btrfs: send, fix invalid path after renaming and linking file

Currently an incremental snapshot can generate link operations which
contain an invalid target path. Such case happens when in the send
snapshot a file was renamed, a new hard link added for it and some
other inode (with a lower number) got renamed to the former name of
that file. Example:

Parent snapshot

 .                  (ino 256)
 |
 |--- f1            (ino 257)
 |--- f2            (ino 258)
 |--- f3            (ino 259)

Send snapshot

 .                  (ino 256)
 |
 |--- f2            (ino 257)
 |--- f3            (ino 258)
 |--- f4            (ino 259)
 |--- f5            (ino 258)

The following steps happen when computing the incremental send stream:

1) When processing inode 257, inode 258 is orphanized (renamed to
   "o258-7-0"), because its current reference has the same name as the
   new reference for inode 257;

2) When processing inode 258, we iterate over all its new references,
   which have the names "f3" and "f5". The first iteration sees name
   "f5" and renames the inode from its orphan name ("o258-7-0") to
   "f5", while the second iteration sees the name "f3" and, incorrectly,
   issues a link operation with a target name matching the orphan name,
   which no longer exists. The first iteration had reset the current
   valid path of the inode to "f5", but in the second iteration we lost
   it because we found another inode, with a higher number of 259, which
   has a reference named "f3" as well, so we orphanized inode 259 and
   recomputed the current valid path of inode 258 to its old orphan
   name because inode 259 could be an ancestor of inode 258 and therefore
   the current valid path could contain the pre-orphanization name of
   inode 259. However in this case inode 259 is not an ancestor of inode
   258 so the current valid path should not be recomputed.
   This makes the receiver fail with the following error:

   ERROR: link f3 -> o258-7-0 failed: No such file or directory

So fix this by not recomputing the current valid path for an inode
whenever we find a colliding reference from some not yet processed inode
(inode number higher then the one currently being processed), unless
that other inode is an ancestor of the one we are currently processing.

A test case for fstests will follow soon.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: fix invalid extent maps due to hole punching
Filipe Manana [Tue, 30 May 2017 04:29:09 +0000 (05:29 +0100)]
Btrfs: fix invalid extent maps due to hole punching

While punching a hole in a range that is not aligned with the sector size
(currently the same as the page size) we can end up leaving an extent map
in memory with a length that is smaller then the sector size or with a
start offset that is not aligned to the sector size. Both cases are not
expected and can lead to problems. This issue is easily detected
after the patch from commit a7e3b975a0f9 ("Btrfs: fix reported number of
inode blocks"), introduced in kernel 4.12-rc1, in a scenario like the
following for example:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ xfs_io -c "pwrite -S 0xaa -b 100K 0 100K" /mnt/foo
  $ xfs_io -c "fpunch 60K 90K" /mnt/foo
  $ xfs_io -c "pwrite -S 0xbb -b 100K 50K 100K" /mnt/foo
  $ xfs_io -c "pwrite -S 0xcc -b 50K 100K 50K" /mnt/foo
  $ umount /mnt

After the unmount operation we can see several warnings emmitted due to
underflows related to space reservation counters:

[ 2837.443299] ------------[ cut here ]------------
[ 2837.447395] WARNING: CPU: 8 PID: 2474 at fs/btrfs/inode.c:9444 btrfs_destroy_inode+0xe8/0x27e [btrfs]
[ 2837.452108] Modules linked in: dm_flakey dm_mod ppdev parport_pc psmouse parport sg pcspkr acpi_cpufreq tpm_tis tpm_tis_core i2c_piix4 i2c_core evdev tpm button se
rio_raw sunrpc loop autofs4 ext4 crc16 jbd2 mbcache btrfs raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c crc32c_gene
ric raid1 raid0 multipath linear md_mod sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy
[ 2837.458389] CPU: 8 PID: 2474 Comm: umount Tainted: G        W       4.10.0-rc8-btrfs-next-43+ #1
[ 2837.459754] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[ 2837.462379] Call Trace:
[ 2837.462379]  dump_stack+0x68/0x92
[ 2837.462379]  __warn+0xc2/0xdd
[ 2837.462379]  warn_slowpath_null+0x1d/0x1f
[ 2837.462379]  btrfs_destroy_inode+0xe8/0x27e [btrfs]
[ 2837.462379]  destroy_inode+0x3d/0x55
[ 2837.462379]  evict+0x177/0x17e
[ 2837.462379]  dispose_list+0x50/0x71
[ 2837.462379]  evict_inodes+0x132/0x141
[ 2837.462379]  generic_shutdown_super+0x3f/0xeb
[ 2837.462379]  kill_anon_super+0x12/0x1c
[ 2837.462379]  btrfs_kill_super+0x16/0x21 [btrfs]
[ 2837.462379]  deactivate_locked_super+0x30/0x68
[ 2837.462379]  deactivate_super+0x36/0x39
[ 2837.462379]  cleanup_mnt+0x58/0x76
[ 2837.462379]  __cleanup_mnt+0x12/0x14
[ 2837.462379]  task_work_run+0x77/0x9b
[ 2837.462379]  prepare_exit_to_usermode+0x9d/0xc5
[ 2837.462379]  syscall_return_slowpath+0x196/0x1b9
[ 2837.462379]  entry_SYSCALL_64_fastpath+0xab/0xad
[ 2837.462379] RIP: 0033:0x7f3ef3e6b9a7
[ 2837.462379] RSP: 002b:00007ffdd0d8de58 EFLAGS: 00000246 ORIG_RAX: 00000000000000a6
[ 2837.462379] RAX: 0000000000000000 RBX: 0000556f76a39060 RCX: 00007f3ef3e6b9a7
[ 2837.462379] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000556f76a3f910
[ 2837.462379] RBP: 0000556f76a3f910 R08: 0000556f76a3e670 R09: 0000000000000015
[ 2837.462379] R10: 00000000000006b4 R11: 0000000000000246 R12: 00007f3ef436ce64
[ 2837.462379] R13: 0000000000000000 R14: 0000556f76a39240 R15: 00007ffdd0d8e0e0
[ 2837.519355] ---[ end trace e79345fe24b30b8d ]---
[ 2837.596256] ------------[ cut here ]------------
[ 2837.597625] WARNING: CPU: 8 PID: 2474 at fs/btrfs/extent-tree.c:5699 btrfs_free_block_groups+0x246/0x3eb [btrfs]
[ 2837.603547] Modules linked in: dm_flakey dm_mod ppdev parport_pc psmouse parport sg pcspkr acpi_cpufreq tpm_tis tpm_tis_core i2c_piix4 i2c_core evdev tpm button serio_raw sunrpc loop autofs4 ext4 crc16 jbd2 mbcache btrfs raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c crc32c_generic raid1 raid0 multipath linear md_mod sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy
[ 2837.659372] CPU: 8 PID: 2474 Comm: umount Tainted: G        W       4.10.0-rc8-btrfs-next-43+ #1
[ 2837.663359] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[ 2837.663359] Call Trace:
[ 2837.663359]  dump_stack+0x68/0x92
[ 2837.663359]  __warn+0xc2/0xdd
[ 2837.663359]  warn_slowpath_null+0x1d/0x1f
[ 2837.663359]  btrfs_free_block_groups+0x246/0x3eb [btrfs]
[ 2837.663359]  close_ctree+0x1dd/0x2e1 [btrfs]
[ 2837.663359]  ? evict_inodes+0x132/0x141
[ 2837.663359]  btrfs_put_super+0x15/0x17 [btrfs]
[ 2837.663359]  generic_shutdown_super+0x6a/0xeb
[ 2837.663359]  kill_anon_super+0x12/0x1c
[ 2837.663359]  btrfs_kill_super+0x16/0x21 [btrfs]
[ 2837.663359]  deactivate_locked_super+0x30/0x68
[ 2837.663359]  deactivate_super+0x36/0x39
[ 2837.663359]  cleanup_mnt+0x58/0x76
[ 2837.663359]  __cleanup_mnt+0x12/0x14
[ 2837.663359]  task_work_run+0x77/0x9b
[ 2837.663359]  prepare_exit_to_usermode+0x9d/0xc5
[ 2837.663359]  syscall_return_slowpath+0x196/0x1b9
[ 2837.663359]  entry_SYSCALL_64_fastpath+0xab/0xad
[ 2837.663359] RIP: 0033:0x7f3ef3e6b9a7
[ 2837.663359] RSP: 002b:00007ffdd0d8de58 EFLAGS: 00000246 ORIG_RAX: 00000000000000a6
[ 2837.663359] RAX: 0000000000000000 RBX: 0000556f76a39060 RCX: 00007f3ef3e6b9a7
[ 2837.663359] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000556f76a3f910
[ 2837.663359] RBP: 0000556f76a3f910 R08: 0000556f76a3e670 R09: 0000000000000015
[ 2837.663359] R10: 00000000000006b4 R11: 0000000000000246 R12: 00007f3ef436ce64
[ 2837.663359] R13: 0000000000000000 R14: 0000556f76a39240 R15: 00007ffdd0d8e0e0
[ 2837.739445] ---[ end trace e79345fe24b30b8e ]---
[ 2837.745595] ------------[ cut here ]------------
[ 2837.746412] WARNING: CPU: 8 PID: 2474 at fs/btrfs/extent-tree.c:5700 btrfs_free_block_groups+0x261/0x3eb [btrfs]
[ 2837.747955] Modules linked in: dm_flakey dm_mod ppdev parport_pc psmouse parport sg pcspkr acpi_cpufreq tpm_tis tpm_tis_core i2c_piix4 i2c_core evdev tpm button serio_raw sunrpc loop autofs4 ext4 crc16 jbd2 mbcache btrfs raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c crc32c_generic raid1 raid0 multipath linear md_mod sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy
[ 2837.755395] CPU: 8 PID: 2474 Comm: umount Tainted: G        W       4.10.0-rc8-btrfs-next-43+ #1
[ 2837.756769] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[ 2837.758526] Call Trace:
[ 2837.758925]  dump_stack+0x68/0x92
[ 2837.759383]  __warn+0xc2/0xdd
[ 2837.759383]  warn_slowpath_null+0x1d/0x1f
[ 2837.759383]  btrfs_free_block_groups+0x261/0x3eb [btrfs]
[ 2837.759383]  close_ctree+0x1dd/0x2e1 [btrfs]
[ 2837.759383]  ? evict_inodes+0x132/0x141
[ 2837.759383]  btrfs_put_super+0x15/0x17 [btrfs]
[ 2837.759383]  generic_shutdown_super+0x6a/0xeb
[ 2837.759383]  kill_anon_super+0x12/0x1c
[ 2837.759383]  btrfs_kill_super+0x16/0x21 [btrfs]
[ 2837.759383]  deactivate_locked_super+0x30/0x68
[ 2837.759383]  deactivate_super+0x36/0x39
[ 2837.759383]  cleanup_mnt+0x58/0x76
[ 2837.759383]  __cleanup_mnt+0x12/0x14
[ 2837.759383]  task_work_run+0x77/0x9b
[ 2837.759383]  prepare_exit_to_usermode+0x9d/0xc5
[ 2837.759383]  syscall_return_slowpath+0x196/0x1b9
[ 2837.759383]  entry_SYSCALL_64_fastpath+0xab/0xad
[ 2837.759383] RIP: 0033:0x7f3ef3e6b9a7
[ 2837.759383] RSP: 002b:00007ffdd0d8de58 EFLAGS: 00000246 ORIG_RAX: 00000000000000a6
[ 2837.759383] RAX: 0000000000000000 RBX: 0000556f76a39060 RCX: 00007f3ef3e6b9a7
[ 2837.759383] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000556f76a3f910
[ 2837.759383] RBP: 0000556f76a3f910 R08: 0000556f76a3e670 R09: 0000000000000015
[ 2837.759383] R10: 00000000000006b4 R11: 0000000000000246 R12: 00007f3ef436ce64
[ 2837.759383] R13: 0000000000000000 R14: 0000556f76a39240 R15: 00007ffdd0d8e0e0
[ 2837.777063] ---[ end trace e79345fe24b30b8f ]---
[ 2837.778235] ------------[ cut here ]------------
[ 2837.778856] WARNING: CPU: 8 PID: 2474 at fs/btrfs/extent-tree.c:9825 btrfs_free_block_groups+0x348/0x3eb [btrfs]
[ 2837.791385] Modules linked in: dm_flakey dm_mod ppdev parport_pc psmouse parport sg pcspkr acpi_cpufreq tpm_tis tpm_tis_core i2c_piix4 i2c_core evdev tpm button serio_raw sunrpc loop autofs4 ext4 crc16 jbd2 mbcache btrfs raid10 raid456 async_raid6_recov async_memcpy async_pq async_xor async_tx xor raid6_pq libcrc32c crc32c_generic raid1 raid0 multipath linear md_mod sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy
[ 2837.797711] CPU: 8 PID: 2474 Comm: umount Tainted: G        W       4.10.0-rc8-btrfs-next-43+ #1
[ 2837.798594] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.9.1-0-gb3ef39f-prebuilt.qemu-project.org 04/01/2014
[ 2837.800118] Call Trace:
[ 2837.800515]  dump_stack+0x68/0x92
[ 2837.801015]  __warn+0xc2/0xdd
[ 2837.801471]  warn_slowpath_null+0x1d/0x1f
[ 2837.801698]  btrfs_free_block_groups+0x348/0x3eb [btrfs]
[ 2837.801698]  close_ctree+0x1dd/0x2e1 [btrfs]
[ 2837.801698]  ? evict_inodes+0x132/0x141
[ 2837.801698]  btrfs_put_super+0x15/0x17 [btrfs]
[ 2837.801698]  generic_shutdown_super+0x6a/0xeb
[ 2837.801698]  kill_anon_super+0x12/0x1c
[ 2837.801698]  btrfs_kill_super+0x16/0x21 [btrfs]
[ 2837.801698]  deactivate_locked_super+0x30/0x68
[ 2837.801698]  deactivate_super+0x36/0x39
[ 2837.801698]  cleanup_mnt+0x58/0x76
[ 2837.801698]  __cleanup_mnt+0x12/0x14
[ 2837.801698]  task_work_run+0x77/0x9b
[ 2837.801698]  prepare_exit_to_usermode+0x9d/0xc5
[ 2837.801698]  syscall_return_slowpath+0x196/0x1b9
[ 2837.801698]  entry_SYSCALL_64_fastpath+0xab/0xad
[ 2837.801698] RIP: 0033:0x7f3ef3e6b9a7
[ 2837.801698] RSP: 002b:00007ffdd0d8de58 EFLAGS: 00000246 ORIG_RAX: 00000000000000a6
[ 2837.801698] RAX: 0000000000000000 RBX: 0000556f76a39060 RCX: 00007f3ef3e6b9a7
[ 2837.801698] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 0000556f76a3f910
[ 2837.801698] RBP: 0000556f76a3f910 R08: 0000556f76a3e670 R09: 0000000000000015
[ 2837.801698] R10: 00000000000006b4 R11: 0000000000000246 R12: 00007f3ef436ce64
[ 2837.801698] R13: 0000000000000000 R14: 0000556f76a39240 R15: 00007ffdd0d8e0e0
[ 2837.818441] ---[ end trace e79345fe24b30b90 ]---
[ 2837.818991] BTRFS info (device sdc): space_info 1 has 7974912 free, is not full
[ 2837.819830] BTRFS info (device sdc): space_info total=8388608, used=417792, pinned=0, reserved=0, may_use=18446744073709547520, readonly=0

What happens in the above example is the following:

1) When punching the hole, at btrfs_punch_hole(), the variable tail_len
   is set to 2048 (as tail_start is 148Kb + 1 and offset + len is 150Kb).
   This results in the creation of an extent map with a length of 2Kb
   starting at file offset 148Kb, through find_first_non_hole() ->
   btrfs_get_extent().

2) The second write (first write after the hole punch operation), sets
   the range [50Kb, 152Kb[ to delalloc.

3) The third write, at btrfs_find_new_delalloc_bytes(), sees the extent
   map covering the range [148Kb, 150Kb[ and ends up calling
   set_extent_bit() for the same range, which results in splitting an
   existing extent state record, covering the range [148Kb, 152Kb[ into
   two 2Kb extent state records, covering the ranges [148Kb, 150Kb[ and
   [150Kb, 152Kb[.

4) Finally at lock_and_cleanup_extent_if_need(), immediately after calling
   btrfs_find_new_delalloc_bytes() we clear the delalloc bit from the
   range [100Kb, 152Kb[ which results in the btrfs_clear_bit_hook()
   callback being invoked against the two 2Kb extent state records that
   cover the ranges [148Kb, 150Kb[ and [150Kb, 152Kb[. When called against
   the first 2Kb extent state, it calls btrfs_delalloc_release_metadata()
   with a length argument of 2048 bytes. That function rounds up the length
   to a sector size aligned length, so it ends up considering a length of
   4096 bytes, and then calls calc_csum_metadata_size() which results in
   decrementing the inode's csum_bytes counter by 4096 bytes, so after
   it stays a value of 0 bytes. Then the same happens when
   btrfs_clear_bit_hook() is called against the second extent state that
   has a length of 2Kb, covering the range [150Kb, 152Kb[, the length is
   rounded up to 4096 and calc_csum_metadata_size() ends up being called
   to decrement 4096 bytes from the inode's csum_bytes counter, which
   at that time has a value of 0, leading to an underflow, which is
   exactly what triggers the first warning, at btrfs_destroy_inode().
   All the other warnings relate to several space accounting counters
   that underflow as well due to similar reasons.

A similar case but where the hole punching operation creates an extent map
with a start offset not aligned to the sector size is the following:

  $ mkfs.btrfs -f /dev/sdb
  $ mount /dev/sdb /mnt
  $ xfs_io -f -c "fpunch 695K 820K" $SCRATCH_MNT/bar
  $ xfs_io -c "pwrite -S 0xaa 1008K 307K" $SCRATCH_MNT/bar
  $ xfs_io -c "pwrite -S 0xbb -b 630K 1073K 630K" $SCRATCH_MNT/bar
  $ xfs_io -c "pwrite -S 0xcc -b 459K 1068K 459K" $SCRATCH_MNT/bar
  $ umount /mnt

During the unmount operation we get similar traces for the same reasons as
in the first example.

So fix the hole punching operation to make sure it never creates extent
maps with a length that is not aligned to the sector size nor with a start
offset that is not aligned to the sector size, as this breaks all
assumptions and it's a land mine.

Fixes: d77815461f04 ("btrfs: Avoid trucating page or punching hole in a already existed hole.")
Cc: <stable@vger.kernel.org>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: add cond_resched to btrfs_qgroup_trace_leaf_items
Jeff Mahoney [Tue, 20 Jun 2017 12:15:26 +0000 (08:15 -0400)]
btrfs: add cond_resched to btrfs_qgroup_trace_leaf_items

On an uncontended system, we can end up hitting soft lockups while
doing replace_path.  At the core, and frequently called is
btrfs_qgroup_trace_leaf_items, so it makes sense to add a cond_resched
there.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Round down values which are written for total_bytes_size
Nikolay Borisov [Fri, 16 Jun 2017 11:39:20 +0000 (14:39 +0300)]
btrfs: Round down values which are written for total_bytes_size

We got an internal report about a file system not wanting to mount
following 99e3ecfcb9f4 ("Btrfs: add more validation checks for
superblock").

BTRFS error (device sdb1): super_total_bytes 1000203816960 mismatch with
fs_devices total_rw_bytes 1000203820544

Subtracting the numbers we get a difference of less than a 4kb. Upon
closer inspection it became apparent that mkfs actually rounds down the
size of the device to a multiple of sector size. However, the same
cannot be said for various functions which modify the total size and are
called from btrfs_balance as well as when adding a new device. So this
patch ensures that values being saved into on-disk data structures are
always rounded down to a multiple of sectorsize.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Manually implement device_total_bytes getter/setter
Nikolay Borisov [Fri, 16 Jun 2017 11:39:19 +0000 (14:39 +0300)]
btrfs: Manually implement device_total_bytes getter/setter

The device->total_bytes member needs to always be rounded down to sectorsize
so that it corresponds to the value of super->total_bytes. However, there are
multiple places where the setter is fed a value which is not rounded which
can cause a fs to be unmountable due to the check introduced in
99e3ecfcb9f4 ("Btrfs: add more validation checks for superblock"). This patch
implements the getter/setter manually so that in a later patch I can add
necessary code to catch offenders.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: obsolete and remove mount option alloc_start
David Sterba [Wed, 14 Jun 2017 23:30:06 +0000 (01:30 +0200)]
btrfs: obsolete and remove mount option alloc_start

The mount option alloc_start was used in the past for debugging and
stressing the chunk allocator. Not meant to be used by users, so we're
not breaking anybody's setup.

There was some added complexity handling changes of the value and when
it was not same as default. Such code has likely been untested and I
think it's better to remove it.

This patch kills all use of alloc_start, and by doing that also fixes
a bug when alloc_size is set, potentially called from statfs:

in btrfs_calc_avail_data_space, traversing the list in RCU, the RCU
protection is temporarily dropped so btrfs_account_dev_extents_size can
be called and then RCU is locked again! Doing that inside
list_for_each_entry_rcu is just asking for trouble, but unlikely to be
observed in practice.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: move fs_info::fs_frozen to the flags
David Sterba [Thu, 15 Jun 2017 17:10:03 +0000 (19:10 +0200)]
btrfs: move fs_info::fs_frozen to the flags

We can keep the state among the other fs_info flags, there's no reason
why fs_frozen would need to be separate.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: cleanup duplicate return value in insert_inline_extent
David Sterba [Thu, 15 Jun 2017 17:09:51 +0000 (19:09 +0200)]
btrfs: cleanup duplicate return value in insert_inline_extent

The pattern when err is used for function exit and ret is used for
return values of callees is not used here.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: use GFP_KERNEL in btrfs_init_dev_replace_tgtdev
David Sterba [Thu, 15 Jun 2017 15:16:43 +0000 (17:16 +0200)]
btrfs: use GFP_KERNEL in btrfs_init_dev_replace_tgtdev

The function is called from ioctl context and we don't hold any locks
that take part in writeback. Right now it's only fs_info::volume_mutex.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: use GFP_KERNEL in btrfs_calc_avail_data_space
David Sterba [Thu, 15 Jun 2017 13:04:04 +0000 (15:04 +0200)]
btrfs: use GFP_KERNEL in btrfs_calc_avail_data_space

We don't hold any locks here. Inidirectly called from statfs.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Use btrfs_space_info_used instead of opencoding it
Nikolay Borisov [Wed, 14 Jun 2017 08:35:34 +0000 (11:35 +0300)]
btrfs: Use btrfs_space_info_used instead of opencoding it

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: wait part of the write_dev_flush() can be separated out
Anand Jain [Tue, 13 Jun 2017 09:05:41 +0000 (17:05 +0800)]
btrfs: wait part of the write_dev_flush() can be separated out

Submit and wait parts of write_dev_flush() can be split into two
separate functions for better readability.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove redundant null bdev counting during flush submission
Anand Jain [Tue, 13 Jun 2017 09:05:40 +0000 (17:05 +0800)]
btrfs: remove redundant null bdev counting during flush submission

There is no extra benefit to count null bdev during the submit loop,
as these null devices will be anyway checked during command
completion device loop just after the submit loop. We are holding the
device_list_mutex, the device->bdev status won't change in between.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: write_dev_flush does not return ENOMEM anymore
Anand Jain [Tue, 13 Jun 2017 09:32:29 +0000 (17:32 +0800)]
btrfs: write_dev_flush does not return ENOMEM anymore

Since commit "btrfs: btrfs_io_bio_alloc never fails, skip error handling"
write_dev_flush will not return ENOMEM in the sending part. We do not
need to check for it in the callers.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ updated changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: compression must free at least one sector size
Timofey Titovets [Tue, 6 Jun 2017 11:41:15 +0000 (14:41 +0300)]
Btrfs: compression must free at least one sector size

We already skip storing data where compression does not make the result
at least one byte less.  Let's make the logic better and check
that compression frees at least one sector size of bytes, otherwise it's
not that useful.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ changelog updated ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: sink gfp parameter to btrfs_io_bio_alloc
David Sterba [Mon, 12 Jun 2017 15:29:41 +0000 (17:29 +0200)]
btrfs: sink gfp parameter to btrfs_io_bio_alloc

We can hardcode GFP_NOFS to btrfs_io_bio_alloc, although it means we
change it back from GFP_KERNEL in scrub. I'd rather save a few stack
bytes from not passing the gfp flags in the remaining, more imporatant,
contexts and the bio allocating API now looks more consistent.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: add helper to initialize the non-bio part of btrfs_io_bio
David Sterba [Mon, 12 Jun 2017 15:29:39 +0000 (17:29 +0200)]
btrfs: add helper to initialize the non-bio part of btrfs_io_bio

We use btrfs_bioset for bios and ask to allocate the entire size of
btrfs_io_bio from btrfs bio_alloc_bioset. The member 'bio' is
initialized but the bytes from 0 to offset of 'bio' are left
uninitialized. Although we initialize some of the members in our
helpers, we should initialize the whole structures.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: document mandatory order of bio in btrfs_io_bio
David Sterba [Mon, 12 Jun 2017 15:29:36 +0000 (17:29 +0200)]
btrfs: document mandatory order of bio in btrfs_io_bio

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: btrfs_ioctl_search_key documentation
Hans van Kranenburg [Mon, 5 Jun 2017 22:20:32 +0000 (00:20 +0200)]
Btrfs: btrfs_ioctl_search_key documentation

A programmer who is trying to implement calling the btrfs SEARCH
or SEARCH_V2 ioctl will probably soon end up reading this struct
definition.

Properly document the input fields to prevent common misconceptions:
 1. The search space is linear, not 3 dimensional. The invidual min/max
 values for objectid, type and offset cannot be used to filter the
 result, they only define the endpoints of an interval.
 2. The transaction id (a.k.a. generation) filter applies only on
 transaction id of the last COW operation on a whole metadata page, not
 on individual items.

Ad 1. The first misunderstanding was helped by the previous misleading
comments on min/max type and offset:
  "keys returned will be >= min and <= max".

Ad 2. For example, running btrfs balance will happily cause rewriting of
metadata pages that contain a filesystem tree of a read only subvolume,
causing transids to be increased.

Also, improve descriptions of tree_id and nr_items and add in/out
annotations.

Signed-off-by: Hans van Kranenburg <hans.van.kranenburg@mendix.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: skip checksum verification if IO error occurs
Liu Bo [Fri, 14 Apr 2017 01:11:48 +0000 (18:11 -0700)]
Btrfs: skip checksum verification if IO error occurs

Currently dio read also goes to verify checksum if -EIO has been returned,
although it usually fails on checksum, it's not necessary at all, we could
directly check if there is another copy to read.

And with this, the behavior of dio read is now consistent with that of
buffered read.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ use bool for uptodate ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: tolerate errors if we have retried successfully
Liu Bo [Wed, 17 May 2017 21:42:00 +0000 (15:42 -0600)]
Btrfs: tolerate errors if we have retried successfully

With raid1 profile, dio read isn't tolerating IO errors if read length is
less than the stripe length (64K).

Our bio didn't get split in btrfs_submit_direct_hook() if (dip->flags &
BTRFS_DIO_ORIG_BIO_SUBMITTED) is true and that happens when the read
length is less than 64k.  In this case, if the underlying device returns
error somehow, bio->bi_error has recorded that error.

If we could recover the correct data from another copy in profile raid1/10/5/6,
with btrfs_subio_endio_read() returning 0, bio would have the correct data in
its vector, but bio->bi_error is not updated accordingly so that the following
dio_end_io(dio_bio, bio->bi_error) makes directIO think this read has failed.

This fixes the problem by setting bio's error to 0 if a good copy has been
found.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: pass bytes to btrfs_bio_alloc
David Sterba [Fri, 2 Jun 2017 16:35:36 +0000 (18:35 +0200)]
btrfs: pass bytes to btrfs_bio_alloc

Most callers of btrfs_bio_alloc convert from bytes to sectors. Hide that
in the helper and simplify the logic in the callsers.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: opencode trivial compressed_bio_alloc, simplify error handling
David Sterba [Fri, 2 Jun 2017 16:01:51 +0000 (18:01 +0200)]
btrfs: opencode trivial compressed_bio_alloc, simplify error handling

compressed_bio_alloc is now a trivial wrapper around btrfs_bio_alloc, no
point keeping it. The error handling can be simplified, as we know
btrfs_bio_alloc will never fail.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove redundant parameters from btrfs_bio_alloc
David Sterba [Fri, 2 Jun 2017 15:55:44 +0000 (17:55 +0200)]
btrfs: remove redundant parameters from btrfs_bio_alloc

All callers pass gfp_flags=GFP_NOFS and nr_vecs=BIO_MAX_PAGES.

submit_extent_page adds __GFP_HIGH that does not make a difference in
our case as it allows access to memory reserves but otherwise does not
change the constraints.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: sink gfp parameter to btrfs_bio_clone
David Sterba [Fri, 2 Jun 2017 15:48:13 +0000 (17:48 +0200)]
btrfs: sink gfp parameter to btrfs_bio_clone

All callers pass GFP_NOFS.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: btrfs_io_bio_alloc never fails, skip error handling
David Sterba [Fri, 2 Jun 2017 15:38:30 +0000 (17:38 +0200)]
btrfs: btrfs_io_bio_alloc never fails, skip error handling

Update direct callers of btrfs_io_bio_alloc that do error handling, that
we can now remove.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: btrfs_bio_clone never fails, skip error handling
David Sterba [Fri, 2 Jun 2017 15:38:30 +0000 (17:38 +0200)]
btrfs: btrfs_bio_clone never fails, skip error handling

Update direct callers of btrfs_bio_clone that do error handling, that we
can now remove.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: btrfs_bio_alloc never fails, skip error handling
David Sterba [Fri, 2 Jun 2017 15:38:30 +0000 (17:38 +0200)]
btrfs: btrfs_bio_alloc never fails, skip error handling

Update direct callers of btrfs_bio_alloc that do error handling, that we
can now remove.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: bioset allocations will never fail, adapt our helpers
David Sterba [Fri, 2 Jun 2017 15:26:26 +0000 (17:26 +0200)]
btrfs: bioset allocations will never fail, adapt our helpers

Christoph pointed out that bio allocations backed by a bioset will never
fail.  As we always use a bioset for all bio allocations, we can skip
the error handling.  This patch adjusts our low-level helpers, the
cascaded changes to all callers will come next.

CC: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: switch to kvmalloc and GFP_KERNEL in lzo/zlib alloc_workspace
David Sterba [Wed, 31 May 2017 15:21:15 +0000 (17:21 +0200)]
btrfs: switch to kvmalloc and GFP_KERNEL in lzo/zlib alloc_workspace

The compression workspace buffers are larger than a page so we use
vmalloc, unconditionally. This is not always necessary as there might be
contiguous memory available.

Let's use the kvmalloc helpers that will try kmalloc first and fallback
to vmalloc. For that they require GFP_KERNEL flags. As we now have the
alloc_workspace calls protected by memalloc_nofs in the critical
contexts, we can safely use GFP_KERNEL.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: switch kmallocs to GFP_KERNEL in lzo/zlib alloc_workspace
David Sterba [Wed, 31 May 2017 15:21:15 +0000 (17:21 +0200)]
btrfs: switch kmallocs to GFP_KERNEL in lzo/zlib alloc_workspace

As alloc_workspace is now protected by memalloc_nofs where needed,
we can switch the kmalloc to use GFP_KERNEL.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: add memalloc_nofs protections around alloc_workspace callback
David Sterba [Wed, 31 May 2017 15:14:56 +0000 (17:14 +0200)]
btrfs: add memalloc_nofs protections around alloc_workspace callback

The workspaces are preallocated at the beginning where we can safely use
GFP_KERNEL, but in some cases the find_workspace might reach the
allocation again, now in a more restricted context when the bios or
pages are being compressed.

To avoid potential lockup when alloc_workspace -> vmalloc would silently
use the GFP_KERNEL, add the memalloc_nofs helpers around the critical
call site.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: adjust includes after vmalloc removal
David Sterba [Wed, 31 May 2017 17:44:31 +0000 (19:44 +0200)]
btrfs: adjust includes after vmalloc removal

As we don't use vmalloc/vzalloc/vfree directly in ctree.c, we can now
use the proper header that defines kvmalloc.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: use GFP_KERNEL in init_ipath
David Sterba [Wed, 31 May 2017 17:32:09 +0000 (19:32 +0200)]
btrfs: use GFP_KERNEL in init_ipath

Now that init_ipath is called either from a safe context or with
memalloc_nofs protection, we can switch to GFP_KERNEL allocations in
init_path and init_data_container.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: add memalloc_nofs protection around init_ipath
David Sterba [Wed, 31 May 2017 17:21:38 +0000 (19:21 +0200)]
btrfs: scrub: add memalloc_nofs protection around init_ipath

init_ipath is called from a safe ioctl context and from scrub when
printing an error.  The protection is added for three reasons:

* init_data_container calls vmalloc and this does not work as expected
  in the GFP_NOFS context, so this silently does GFP_KERNEL and might
  deadlock in some cases
* keep the context constraint of GFP_NOFS, used by scrub
* we want to use GFP_KERNEL unconditionally inside init_ipath or its
  callees

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: send: use kvmalloc in iterate_dir_item
David Sterba [Wed, 31 May 2017 16:40:02 +0000 (18:40 +0200)]
btrfs: send: use kvmalloc in iterate_dir_item

We use a growing buffer for xattrs larger than a page size, at some
point vmalloc is unconditionally used for larger buffers. We can still
try to avoid it using the kvmalloc helper.

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: replace opencoded kvzalloc with the helper
David Sterba [Wed, 31 May 2017 16:40:02 +0000 (18:40 +0200)]
btrfs: replace opencoded kvzalloc with the helper

The logic of kmalloc and vmalloc fallback is opencoded in
several places, we can now use the existing helper.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: lzo: compressed data size must be less then input size
Timofey Titovets [Mon, 29 May 2017 23:18:04 +0000 (02:18 +0300)]
Btrfs: lzo: compressed data size must be less then input size

Logic already skips if compression makes data bigger, let's sync lzo
with zlib and also return error if compressed size is equal to
input size.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: simplify code with bio_io_error
Guoqing Jiang [Fri, 2 Jun 2017 08:08:50 +0000 (16:08 +0800)]
btrfs: simplify code with bio_io_error

bio_io_error was introduced in the commit 4246a0b63bd8f56a1469b
("block: add a bi_error field to struct bio"), so use it to simplify
code.

Signed-off-by: Guoqing Jiang <gqjiang@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: use memalloc_nofs and kvzalloc() for free space tree bitmaps
Omar Sandoval [Mon, 5 Jun 2017 07:12:31 +0000 (00:12 -0700)]
Btrfs: use memalloc_nofs and kvzalloc() for free space tree bitmaps

First, instead of open-coding the vmalloc() fallback, use the new
kvzalloc() helper. Second, use memalloc_nofs_{save,restore}() instead of
GFP_NOFS, as vmalloc() uses some GFP_KERNEL allocations internally which
could lead to deadlocks.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: use generic slab for for btrfs_transaction
David Sterba [Tue, 28 Mar 2017 10:06:05 +0000 (12:06 +0200)]
btrfs: use generic slab for for btrfs_transaction

Observing the number of slab objects of btrfs_transaction, there's just
one active on an almost quiescent filesystem, and the number of objects
goes to about ten when sync is in progress. Then the nubmer goes down to
1.  This matches the expectations of the transaction lifetime.

For such use the separate slab cache is not justified, as we do not
reuse objects frequently. For the shortlived transaction, the generic
slab (size 512) should be ok. We can optimistically expect that the 512
slabs are not all used (fragmentation) and there are free slots to take
when we do the allocation, compared to potentially allocating a whole new
page for the separate slab.

We'll lose the stats about the object use, which could be added later if
we really need them.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: embed scrub_wr_ctx into scrub context
David Sterba [Tue, 16 May 2017 17:10:32 +0000 (19:10 +0200)]
btrfs: scrub: embed scrub_wr_ctx into scrub context

The structure scrub_wr_ctx is not used anywhere just the scrub context,
we can move the members there. The tgtdev is renamed so it's more clear
that it belongs to the "wr" part.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: use fs_info::sectorsize and drop it from scrub context
David Sterba [Tue, 16 May 2017 17:10:41 +0000 (19:10 +0200)]
btrfs: scrub: use fs_info::sectorsize and drop it from scrub context

As we now have the node/block sizes in fs_info, we can use them and can
drop the local copies.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: add statx support
Yonghong Song [Fri, 12 May 2017 22:07:43 +0000 (15:07 -0700)]
Btrfs: add statx support

Return enhanced file attributes from the btrfs, including:
  (1). inode creation time as stx_btime, and
  (2). Certain BTRFS_INODE_xxx flags are mapped to stx_attributes flags.

Example output:
[root@localhost ~]# cat t.sh
touch t
chattr +aic t
~/linux/samples/statx/test-statx t
chattr -aic t
touch t
echo "========================================"
~/linux/samples/statx/test-statx t
/bin/rm t
[root@localhost ~]# ./t.sh
statx(t) = 0
results=fff
     Size: 0               Blocks: 0          IO Block: 4096    regular file
Device: 00:1c           Inode: 63962       Links: 1
Access: (0644/-rw-r--r--)  Uid:     0   Gid:     0
Access: 2017-05-11 16:03:13.999856591-0700
Modify: 2017-05-11 16:03:13.999856591-0700
Change: 2017-05-11 16:03:14.000856663-0700
   Birth: 2017-05-11 16:03:13.999856591-0700
Attributes: 0000000000000034 (........ ........ ........ ........ ........ ........ ........ .-ai.c..)
========================================
statx(t) = 0
results=fff
  Size: 0               Blocks: 0          IO Block: 4096    regular file
Device: 00:1c           Inode: 63962       Links: 1
Access: (0644/-rw-r--r--)  Uid:     0   Gid:     0
Access: 2017-05-11 16:03:14.006857097-0700
Modify: 2017-05-11 16:03:14.006857097-0700
Change: 2017-05-11 16:03:14.006857097-0700
  Birth: 2017-05-11 16:03:13.999856591-0700
Attributes: 0000000000000000 (........ ........ ........ ........ ........ ........ ........ .---.-..)
[root@localhost ~]#

Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: lzo: fix typo in error message after failed deflate
Timofey Titovets [Thu, 25 May 2017 18:12:19 +0000 (21:12 +0300)]
Btrfs: lzo: fix typo in error message after failed deflate

Fix copy paste typo in debug message for lzo.c, lzo is not deflate.

Signed-off-by: Timofey Titovets <nefelim4ag@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: btrfs_wait_tree_block_writeback can be void return
Jeff Layton [Thu, 25 May 2017 10:39:52 +0000 (06:39 -0400)]
btrfs: btrfs_wait_tree_block_writeback can be void return

Nothing checks its return value.

Is it safe to skip checking return value of btrfs_wait_tree_block_writeback?

Liu Bo: I think yes, it's used in walk_log_tree which is called in two
places, free_log_tree and log replay.  For free_log_tree, it waits for
any running writeback of the extent buffer under freeing to finish in
case we need to access the eb pointer from page->private, and it's OK to
not check the return value, while for log replay, it's doesn't wait
because wc->wait is not set. So neither cares about the writeback error.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
[ added more explanation to changelog, from Liu Bo ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove __BTRFS_LEAF_DATA_SIZE
Nikolay Borisov [Mon, 22 May 2017 10:16:11 +0000 (13:16 +0300)]
btrfs: remove __BTRFS_LEAF_DATA_SIZE

__BTRFS_LAF_DATA_SIZE is used only by BTRFS_LEAF_DATA_SIZE. Make the
latter subsume the former.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: rename btrfs_leaf_data to BTRFS_LEAF_DATA_OFFSET
Nikolay Borisov [Mon, 29 May 2017 06:43:43 +0000 (09:43 +0300)]
btrfs: rename btrfs_leaf_data to BTRFS_LEAF_DATA_OFFSET

Commit 5f39d397dfbe ("Btrfs: Create extent_buffer interface
for large blocksizes") refactored btrfs_leaf_data function to take
extent_buffer rather than struct btrfs_leaf. However, as it turns out the
parameter being passed is never used. Furthermore this function no longer
returns the leaf data but rather the offset to it. So rename the function
to BTRFS_LEAF_DATA_OFFSET to make it consistent with other BTRFS_LEAF_*
helpers and turn it into a macro.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
[ removed () from the macro ]
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: reduce arguments for decompress_bio ops
Anand Jain [Fri, 26 May 2017 07:44:59 +0000 (15:44 +0800)]
btrfs: reduce arguments for decompress_bio ops

struct compressed_bio pointer can be used instead.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: btrfs_decompress_bio() could accept compressed_bio instead
Anand Jain [Fri, 26 May 2017 07:44:58 +0000 (15:44 +0800)]
btrfs: btrfs_decompress_bio() could accept compressed_bio instead

Instead of sending each argument of struct compressed_bio, send
the compressed_bio itself.

Also by having struct compressed_bio in btrfs_decompress_bio()
it would help tracing.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Refactor update_space_info
Nikolay Borisov [Mon, 22 May 2017 06:35:50 +0000 (09:35 +0300)]
btrfs: Refactor update_space_info

Following the factoring out of the creation code udpate_space_info can
only be called for already-existing space_info structs. As such it
cannot fail.  Remove superfluous error handling and make the function
return void.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Separate space_info create/update
Nikolay Borisov [Mon, 22 May 2017 06:35:49 +0000 (09:35 +0300)]
btrfs: Separate space_info create/update

Currently the struct space_info creation code is intermixed in the
udpate_space_info function. There are well-defined points at which the
we actually want to create brand-new space_info structs (e.g. during
mount of the filesystem as well as sometimes when adding/initialising
new chunks). In such cases update_space_info is called with 0 as the
bytes parameter. All of this makes for spaghetti code.

Fix it by factoring out the creation code in a separate
create_space_info structure. This also allows to simplify the internals.
Also remove BUG_ON from do_alloc_chunk since the callers handle errors.
Furthermore it will make the update_space_info function not fail,
allowing us to remove error handling in callers. This will come in a
follow up patch.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: let btrfs_print_leaf print more about block group
Liu Bo [Fri, 26 May 2017 00:08:12 +0000 (18:08 -0600)]
Btrfs: let btrfs_print_leaf print more about block group

This adds chunk_objectid and flags, with flags we can recognize whether
the block group is about data or metadata.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: skip commit transaction if we don't have enough pinned bytes
Liu Bo [Fri, 19 May 2017 17:39:15 +0000 (11:39 -0600)]
Btrfs: skip commit transaction if we don't have enough pinned bytes

We commit transaction in order to reclaim space from pinned bytes because
it could process delayed refs, and in may_commit_transaction(), we check
first if pinned bytes are enough for the required space, we then check if
that plus bytes reserved for delayed insert are enough for the required
space.

This changes the code to the above logic.

Fixes: b150a4f10d87 ("Btrfs: use a percpu to keep track of possibly pinned bytes")
Tested-by: Nikolay Borisov <nborisov@suse.com>
Reported-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: simplify cleanup of wr_ctx in scrub_free_ctx
David Sterba [Tue, 16 May 2017 17:10:29 +0000 (19:10 +0200)]
btrfs: scrub: simplify cleanup of wr_ctx in scrub_free_ctx

We don't need to take the mutex and zero out wr_cur_bio, as this is
called after the scrub finished.

Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: inline helper scrub_free_wr_ctx
David Sterba [Tue, 16 May 2017 17:10:26 +0000 (19:10 +0200)]
btrfs: scrub: inline helper scrub_free_wr_ctx

The helper scrub_free_wr_ctx is used only once and fits into
scrub_free_ctx as it continues sctx shutdown, no need to keep it
separate.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: scrub: inline helper scrub_setup_wr_ctx
David Sterba [Tue, 16 May 2017 17:10:23 +0000 (19:10 +0200)]
btrfs: scrub: inline helper scrub_setup_wr_ctx

The helper scrub_setup_wr_ctx is used only once and fits into
scrub_setup_ctx as it continues intialization, no need to keep it
separate.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove root usage from can_overcommit
Jeff Mahoney [Wed, 17 May 2017 15:38:36 +0000 (11:38 -0400)]
btrfs: remove root usage from can_overcommit

can_overcommit using the root to determine the allocation profile
is the only use of a root in the call graph below reserve_metadata_bytes.

It turns out that we only need to know whether the allocation is for
the chunk root or not -- and we can pass that around as a bool instead.

This allows us to pull root usage out of the reservation path all the
way up to reserve_metadata_bytes itself, which uses it only to compare
against fs_info->chunk_root to set the bool.  In turn, this eliminates
a bunch of races where we use a particular root too early in the mount
process.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: cleanup root usage by btrfs_get_alloc_profile
Jeff Mahoney [Wed, 17 May 2017 15:38:35 +0000 (11:38 -0400)]
btrfs: cleanup root usage by btrfs_get_alloc_profile

There are two places where we don't already know what kind of alloc
profile we need before calling btrfs_get_alloc_profile, but we need
access to a root everywhere we call it.

This patch adds helpers for btrfs_{data,metadata,system}_alloc_profile()
and relegates btrfs_system_alloc_profile to a static for use in those
two cases.  The next patch will eliminate one of those.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: fix bool type in btrfs_page_exists_in_range
David Sterba [Thu, 11 May 2017 23:02:22 +0000 (01:02 +0200)]
btrfs: fix bool type in btrfs_page_exists_in_range

We use only a simple bool indicator, int is not a problem here.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove unused member list from btrfs_end_io_wq
David Sterba [Thu, 13 Apr 2017 17:11:04 +0000 (19:11 +0200)]
btrfs: remove unused member list from btrfs_end_io_wq

The end io work queue items have been tracked by the work queues since
"Btrfs: Add async worker threads for pre and post IO checksumming"
(8b7128429235d9bd72cfd5e) (2008).

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove unused members dir_path from recorded_ref
David Sterba [Thu, 13 Apr 2017 17:11:04 +0000 (19:11 +0200)]
btrfs: remove unused members dir_path from recorded_ref

The two members do not seem to be used since the initial commit.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove unused member list from async_submit_bio
David Sterba [Thu, 13 Apr 2017 17:11:04 +0000 (19:11 +0200)]
btrfs: remove unused member list from async_submit_bio

The list used to track checksums in the early version (2.6.29), but I
was able not pinpoint the commit that stopped using it. Everything
apparently works without it for a long time.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: remove unused member err from reada_extent
David Sterba [Thu, 13 Apr 2017 17:11:04 +0000 (19:11 +0200)]
btrfs: remove unused member err from reada_extent

Seems to be unused since the initial commit, we ignore readahead errors
anyway, the full read will handle that if necessary.

Signed-off-by: David Sterba <dsterba@suse.com>
7 years agobtrfs: Remove unnecessary branching in free-space-tree.c
Sahil Kang [Wed, 17 May 2017 10:33:45 +0000 (03:33 -0700)]
btrfs: Remove unnecessary branching in free-space-tree.c

Both btrfs_create_free_space_tree and btrfs_clear_free_space_tree
contain:

  if (ret)
          return ret;

  return 0;

The if statement is only false when ret equals zero, and since we return
zero in such cases, we can safely remove the branching.

Signed-off-by: Sahil Kang <sahil.kang@asilaycomputing.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: hardcode GFP_NOFS for btrfs_bio_clone_partial
Liu Bo [Tue, 16 May 2017 17:57:14 +0000 (10:57 -0700)]
Btrfs: hardcode GFP_NOFS for btrfs_bio_clone_partial

We only pass GFP_NOFS to btrfs_bio_clone_partial, so lets hardcode it.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: work around maybe-uninitialized warning
Arnd Bergmann [Thu, 18 May 2017 13:33:29 +0000 (15:33 +0200)]
Btrfs: work around maybe-uninitialized warning

A rewrite of btrfs_submit_direct_hook appears to have introduced a warning:

fs/btrfs/inode.c: In function 'btrfs_submit_direct_hook':
fs/btrfs/inode.c:8467:14: error: 'bio' may be used uninitialized in this function [-Werror=maybe-uninitialized]

Where the 'bio' variable was previously initialized unconditionally, it
is now set in the "while (submit_len > 0)" loop that would never execute
if submit_len is zero.

Assuming this cannot happen in practice, we can avoid the warning
by simply replacing the while{} loop with a do{}while() loop so
the compiler knows that it will always be entered at least once.

Fixes changes introduced in "Btrfs: use bio_clone_bioset_partial to
simplify DIO submit".

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: unify naming of btrfs_io_bio
Liu Bo [Mon, 17 Apr 2017 22:00:28 +0000 (15:00 -0700)]
Btrfs: unify naming of btrfs_io_bio

All dio endio functions are using io_bio for struct btrfs_io_bio, this
makes btrfs_submit_direct to follow this convention.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: check-integrity use bvec_iter
Liu Bo [Fri, 14 Apr 2017 23:11:52 +0000 (16:11 -0700)]
Btrfs: check-integrity use bvec_iter

Some check-integrity code depends on bio->bi_vcnt, this changes it to use
bio segments because some bios passing here may not have a reliable
bi_vcnt.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: record error if one block has failed to retry
Liu Bo [Tue, 16 May 2017 00:20:07 +0000 (17:20 -0700)]
Btrfs: record error if one block has failed to retry

In the nocsum case of dio read endio, it returns immediately if an error
gets returned when repairing, which leaves the rest blocks unrepaired.  The
behavior is different from how buffered read endio works in the same case.
This changes it to record error only and go on repairing the rest blocks.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: change how we iterate bios in endio
Liu Bo [Mon, 15 May 2017 22:33:27 +0000 (15:33 -0700)]
Btrfs: change how we iterate bios in endio

Since dio submit has used bio_clone_fast, the submitted bio may not have a
reliable bi_vcnt, for the bio vector iterations in checksum related
functions, bio->bi_iter is not modified yet and it's safe to use
bio_for_each_segment, while for those bio vector iterations in dio read's
endio, we now save a copy of bvec_iter in struct btrfs_io_bio when cloning
bios and use the helper __bio_for_each_segment with the saved bvec_iter to
access each bvec.

Also for dio reads which don't get split, we also need to save a copy of
bio iterator in btrfs_bio_clone to let __bio_for_each_segments to access
each bvec in dio read's endio.  Note that it doesn't affect other calls of
btrfs_bio_clone() because they don't need to use this iterator.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: use bio_clone_bioset_partial to simplify DIO submit
Liu Bo [Tue, 16 May 2017 16:51:39 +0000 (09:51 -0700)]
Btrfs: use bio_clone_bioset_partial to simplify DIO submit

Currently when mapping bio to limit bio to a single stripe length, we
split bio by adding page to bio one by one, but later we don't modify
the vector of bio at all, thus we can use bio_clone_fast to use the
original bio vector directly.

Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
7 years agoBtrfs: new helper btrfs_bio_clone_partial
Liu Bo [Tue, 16 May 2017 00:43:31 +0000 (17:43 -0700)]
Btrfs: new helper btrfs_bio_clone_partial

This adds a new helper btrfs_bio_clone_partial, it'll allocate a cloned
bio that only owns a part of the original bio's data.

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