The on-disk representation and the vfs both use 64-bit tv_sec values,
so let's change the last missing piece in the middle.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In error path of f2fs_move_rehashed_dirents, inode page could be writeback
state, so we should wait on inode page writeback before updating it.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

As Jens reported, we'd better assign REQ_RAHEAD to bio by the fact that
->readpages is called only from read-ahead.

In Documentation/filesystems/vfs.txt,

readpages: called by the VM to read pages associated with the address_space
  	object. This is essentially just a vector version of
  	readpage.  Instead of just one page, several pages are
  	requested.
	readpages is only used for read-ahead, so read errors are
  	ignored.  If anything goes wrong, feel free to give up.
Signed-off-by: NJens Axboe <axboe@kernel.dk>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch fix hungtask problem which can be reproduced as follow:

Thread 0~3:
while true
do
        touch /xxx/test/file_xxx
done

Thread 4 write a new checkpoint every three seconds.

In the meantime, fio start 16 threads for randwrite.

With my debug info, cycles num will exceed 1000 in function
f2fs_sync_dirty_inodes, and most of cycle will be dropped
into congestion_wait() and sleep more than 20ms. Cycles num
reduced to 3 with this patch.
Signed-off-by: NYunlei He <heyunlei@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

"ret" can be uninitialized on the success path when "in ==
F2FS_GOING_DOWN_FULLSYNC".

Fixes: 60b2b4ee ("f2fs: Fix deadlock in shutdown ioctl")
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Actually, we don't need to issue discard commands, if discard is on, as
mentioned in the comment.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

As Anatoly Trosinenko reported in bugzilla:

How to reproduce:
1. Compile the 73fcb1a3 version of the kernel using the config attached
2. Unpack and mount the attached filesystem image as F2FS
3. The kernel will BUG() on mount (BUGs are explicitly enabled in config)

[    2.233612] F2FS-fs (sda): Found nat_bits in checkpoint
[    2.248422] ------------[ cut here ]------------
[    2.248857] kernel BUG at fs/f2fs/node.c:1967!
[    2.249760] invalid opcode: 0000 [#1] SMP NOPTI
[    2.250219] Modules linked in:
[    2.251848] CPU: 0 PID: 944 Comm: mount Not tainted 4.17.0-rc5+ #1
[    2.252331] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1ubuntu1 04/01/2014
[    2.253305] RIP: 0010:build_free_nids+0x337/0x3f0
[    2.253672] RSP: 0018:ffffae7fc0857c50 EFLAGS: 00000246
[    2.254080] RAX: 00000000ffffffff RBX: 0000000000000123 RCX: 0000000000000001
[    2.254638] RDX: ffff9aa7063d5c00 RSI: 0000000000000122 RDI: ffff9aa705852e00
[    2.255190] RBP: ffff9aa705852e00 R08: 0000000000000001 R09: ffff9aa7059090c0
[    2.255719] R10: 0000000000000000 R11: 0000000000000000 R12: ffff9aa705852e00
[    2.256242] R13: ffff9aa7063ad000 R14: ffff9aa705919000 R15: 0000000000000123
[    2.256809] FS:  00000000023078c0(0000) GS:ffff9aa707800000(0000) knlGS:0000000000000000
[    2.258654] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    2.259153] CR2: 00000000005511ae CR3: 0000000005872000 CR4: 00000000000006f0
[    2.259801] Call Trace:
[    2.260583]  build_node_manager+0x5cd/0x600
[    2.260963]  f2fs_fill_super+0x66a/0x17c0
[    2.261300]  ? f2fs_commit_super+0xe0/0xe0
[    2.261622]  mount_bdev+0x16e/0x1a0
[    2.261899]  mount_fs+0x30/0x150
[    2.262398]  vfs_kern_mount.part.28+0x4f/0xf0
[    2.262743]  do_mount+0x5d0/0xc60
[    2.263010]  ? _copy_from_user+0x37/0x60
[    2.263313]  ? memdup_user+0x39/0x60
[    2.263692]  ksys_mount+0x7b/0xd0
[    2.263960]  __x64_sys_mount+0x1c/0x20
[    2.264268]  do_syscall_64+0x43/0xf0
[    2.264560]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
[    2.265095] RIP: 0033:0x48d31a
[    2.265502] RSP: 002b:00007ffc6fe60a08 EFLAGS: 00000246 ORIG_RAX: 00000000000000a5
[    2.266089] RAX: ffffffffffffffda RBX: 0000000000008000 RCX: 000000000048d31a
[    2.266607] RDX: 00007ffc6fe62fa5 RSI: 00007ffc6fe62f9d RDI: 00007ffc6fe62f94
[    2.267130] RBP: 00000000023078a0 R08: 0000000000000000 R09: 0000000000000000
[    2.267670] R10: 0000000000008000 R11: 0000000000000246 R12: 0000000000000000
[    2.268192] R13: 0000000000000000 R14: 00007ffc6fe60c78 R15: 0000000000000000
[    2.268767] Code: e8 5f c3 ff ff 83 c3 01 41 83 c7 01 81 fb c7 01 00 00 74 48 44 39 7d 04 76 42 48 63 c3 48 8d 04 c0 41 8b 44 06 05 83 f8 ff 75 c1 <0f> 0b 49 8b 45 50 48 8d b8 b0 00 00 00 e8 37 59 69 00 b9 01 00
[    2.270434] RIP: build_free_nids+0x337/0x3f0 RSP: ffffae7fc0857c50
[    2.271426] ---[ end trace ab20c06cd3c8fde4 ]---

During loading NAT entries, we will do sanity check, once the entry info
is corrupted, it will cause BUG_ON directly to protect user data from
being overwrited.

In this case, it will be better to just return failure on mount() instead
of panic, so that user can get hint from kmsg and try fsck for recovery
immediately rather than after an abnormal reboot.

https://bugzilla.kernel.org/show_bug.cgi?id=199769Reported-by: NAnatoly Trosinenko <anatoly.trosinenko@gmail.com>
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Enable in-memory inode checksum to protect metadata blocks from
in-memory scribbles when checking consistency, which has no
performance requirements.
Signed-off-by: NWeichao Guo <guoweichao@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In fill_super, if root inode's attribute is incorrect, we need to
call f2fs_destroy_stats to release stats memory.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

readdir_ra is sysfs configuration instead of mount option, so it should
not be initialized in default_options(), otherwise after remount, it can
be reset to be enabled which may not as user wish, so let's move it to
f2fs_tuning_parameters().
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Let default_options() initialize s_res{u,g}id with default value like
other options.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

During orphan inode recovery, checkpoint should never succeed due to
SBI_POR_DOING flag, so we don't need acquire orphan ino which only be
used by checkpoint.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

[  155.018460] ======================================================
[  155.021431] WARNING: possible circular locking dependency detected
[  155.024339] 4.18.0-rc3+ #5 Tainted: G           OE
[  155.026879] ------------------------------------------------------
[  155.029783] umount/2901 is trying to acquire lock:
[  155.032187] 00000000c4282f1f (kn->count#130){++++}, at: kernfs_remove+0x1f/0x30
[  155.035439]
[  155.035439] but task is already holding lock:
[  155.038892] 0000000056e4307b (&type->s_umount_key#41){++++}, at: deactivate_super+0x33/0x50
[  155.042602]
[  155.042602] which lock already depends on the new lock.
[  155.042602]
[  155.047465]
[  155.047465] the existing dependency chain (in reverse order) is:
[  155.051354]
[  155.051354] -> #1 (&type->s_umount_key#41){++++}:
[  155.054768]        f2fs_sbi_store+0x61/0x460 [f2fs]
[  155.057083]        kernfs_fop_write+0x113/0x1a0
[  155.059277]        __vfs_write+0x36/0x180
[  155.061250]        vfs_write+0xbe/0x1b0
[  155.063179]        ksys_write+0x55/0xc0
[  155.065068]        do_syscall_64+0x60/0x1b0
[  155.067071]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  155.069529]
[  155.069529] -> #0 (kn->count#130){++++}:
[  155.072421]        __kernfs_remove+0x26f/0x2e0
[  155.074452]        kernfs_remove+0x1f/0x30
[  155.076342]        kobject_del.part.5+0xe/0x40
[  155.078354]        f2fs_put_super+0x12d/0x290 [f2fs]
[  155.080500]        generic_shutdown_super+0x6c/0x110
[  155.082655]        kill_block_super+0x21/0x50
[  155.084634]        kill_f2fs_super+0x9c/0xc0 [f2fs]
[  155.086726]        deactivate_locked_super+0x3f/0x70
[  155.088826]        cleanup_mnt+0x3b/0x70
[  155.090584]        task_work_run+0x93/0xc0
[  155.092367]        exit_to_usermode_loop+0xf0/0x100
[  155.094466]        do_syscall_64+0x162/0x1b0
[  155.096312]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  155.098603]
[  155.098603] other info that might help us debug this:
[  155.098603]
[  155.102418]  Possible unsafe locking scenario:
[  155.102418]
[  155.105134]        CPU0                    CPU1
[  155.107037]        ----                    ----
[  155.108910]   lock(&type->s_umount_key#41);
[  155.110674]                                lock(kn->count#130);
[  155.113010]                                lock(&type->s_umount_key#41);
[  155.115608]   lock(kn->count#130);
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Once we shutdown f2fs, we have to flush stale pages in order to unmount
the system. In order to make stable, we need to stop fault injection as well.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

It turns out losing meta pages in shutdown period makes f2fs very unstable
so that I could see many unexpected error conditions.

Let's keep meta pages for fault injection and sudden power-off tests.
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

When unmounting f2fs in force mode, we can get it stuck by io_schedule()
by some pending IOs in meta_inode.

io_schedule+0xd/0x30
wait_on_page_bit_common+0xc6/0x130
__filemap_fdatawait_range+0xbd/0x100
filemap_fdatawait_keep_errors+0x15/0x40
sync_inodes_sb+0x1cf/0x240
sync_filesystem+0x52/0x90
generic_shutdown_super+0x1d/0x110
kill_f2fs_super+0x28/0x80 [f2fs]
deactivate_locked_super+0x35/0x60
cleanup_mnt+0x36/0x70
task_work_run+0x79/0xa0
exit_to_usermode_loop+0x62/0x70
do_syscall_64+0xdb/0xf0
entry_SYSCALL_64_after_hwframe+0x44/0xa9
0xffffffffffffffff
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This fixes to support dio having unaligned buffers as buffered writes.

xfs_io -f -d -c "pwrite 0 512" $testfile
 -> okay

xfs_io -f -d -c "pwrite 1 512" $testfile
 -> EINVAL
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Let's flush journal nat entries for speed up in the next run.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

sgid directories have special semantics, making newly created files in
the directory belong to the group of the directory, and newly created
subdirectories will also become sgid.  This is historically used for
group-shared directories.

But group directories writable by non-group members should not imply
that such non-group members can magically join the group, so make sure
to clear the sgid bit on non-directories for non-members (but remember
that sgid without group execute means "mandatory locking", just to
confuse things even more).
Reported-by: NJann Horn <jannh@google.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

It turns out that systemd has a bug: it wants to load the autofs module
early because of some initialization ordering with udev, and it doesn't
do that correctly.  Everywhere else it does the proper "look up module
name" that does the proper alias resolution, but in that early code, it
just uses a hardcoded "autofs4" for the module name.

The result of that is that as of commit a2225d93 ("autofs: remove
left-over autofs4 stubs"), you get

    systemd[1]: Failed to insert module 'autofs4': No such file or directory

in the system logs, and a lack of module loading.  All this despite the
fact that we had very clearly marked 'autofs4' as an alias for this
module.

What's so ridiculous about this is that literally everything else does
the module alias handling correctly, including really old versions of
systemd (that just used 'modprobe' to do this), and even all the other
systemd module loading code.

Only that special systemd early module load code is broken, hardcoding
the module names for not just 'autofs4', but also "ipv6", "unix",
"ip_tables" and "virtio_rng".  Very annoying.

Instead of creating an _additional_ separate compatibility 'autofs4'
module, just rely on the fact that everybody else gets this right, and
just call the module 'autofs4' for compatibility reasons, with 'autofs'
as the alias name.

That will allow the systemd people to fix their bugs, adding the proper
alias handling, and maybe even fix the name of the module to be just
"autofs" (so that they can _test_ the alias handling).  And eventually,
we can revert this silly compatibility hack.

See also

    https://github.com/systemd/systemd/issues/9501
    https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=902946

for the systemd bug reports upstream and in the Debian bug tracker
respectively.

Fixes: a2225d93 ("autofs: remove left-over autofs4 stubs")
Reported-by: NBen Hutchings <ben@decadent.org.uk>
Reported-by: NMichael Biebl <biebl@debian.org>
Cc: Ian Kent <raven@themaw.net>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Use huge_ptep_get() to translate huge ptes to normal ptes so we can
check them with the huge_pte_* functions.  Otherwise some architectures
will check the wrong values and will not wait for userspace to bring in
the memory.

Link: http://lkml.kernel.org/r/20180626132421.78084-1-frankja@linux.ibm.com
Fixes: 369cd212 ("userfaultfd: hugetlbfs: userfaultfd_huge_must_wait for hugepmd ranges")
Signed-off-by: NJanosch Frank <frankja@linux.ibm.com>
Reviewed-by: NDavid Hildenbrand <david@redhat.com>
Reviewed-by: NMike Kravetz <mike.kravetz@oracle.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The poll() changes were not well thought out, and completely
unexplained.  They also caused a huge performance regression, because
"->poll()" was no longer a trivial file operation that just called down
to the underlying file operations, but instead did at least two indirect
calls.

Indirect calls are sadly slow now with the Spectre mitigation, but the
performance problem could at least be largely mitigated by changing the
"->get_poll_head()" operation to just have a per-file-descriptor pointer
to the poll head instead.  That gets rid of one of the new indirections.

But that doesn't fix the new complexity that is completely unwarranted
for the regular case.  The (undocumented) reason for the poll() changes
was some alleged AIO poll race fixing, but we don't make the common case
slower and more complex for some uncommon special case, so this all
really needs way more explanations and most likely a fundamental
redesign.

[ This revert is a revert of about 30 different commits, not reverted
  individually because that would just be unnecessarily messy  - Linus ]

Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

If a power failure happens while the qgroup rescan kthread is running,
the next mount operation will always fail. This is because of a recent
regression that makes qgroup_rescan_init() incorrectly return -EINVAL
when we are mounting the filesystem (through btrfs_read_qgroup_config()).
This causes the -EINVAL error to be returned regardless of any qgroup
flags being set instead of returning the error only when neither of
the flags BTRFS_QGROUP_STATUS_FLAG_RESCAN nor BTRFS_QGROUP_STATUS_FLAG_ON
are set.

A test case for fstests follows up soon.

Fixes: 9593bf49 ("btrfs: qgroup: show more meaningful qgroup_rescan_init error message")
Signed-off-by: NFilipe Manana <fdmanana@suse.com>
Reviewed-by: NQu Wenruo <wqu@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The vm_fault_t conversion commit introduced a ret2 variable for tracking
the integer return values from internal btrfs functions.  It was
sometimes returning VM_FAULT_LOCKED for pages that were actually invalid
and had been removed from the radix.  Something like this:

    ret2 = btrfs_delalloc_reserve_space() // returns zero on success

    lock_page(page)
    if (page->mapping != inode->i_mapping)
	goto out_unlock;

...

out_unlock:
    if (!ret2) {
	    ...
	    return VM_FAULT_LOCKED;
    }

This ends up triggering this WARNING in btrfs_destroy_inode()
    WARN_ON(BTRFS_I(inode)->block_rsv.size);

xfstests generic/095 was able to reliably reproduce the errors.

Since out_unlock: is only used for errors, this fix moves it below the
if (!ret2) check we use to return VM_FAULT_LOCKED for success.

Fixes: a528a241 (btrfs: change return type of btrfs_page_mkwrite to vm_fault_t)
Signed-off-by: NChris Mason <clm@fb.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Commit ff3d27a0 ("btrfs: qgroup: Finish rescan when hit the last leaf
of extent tree") added a new exit for rescan finish.

However after finishing quota rescan, we set
fs_info->qgroup_rescan_progress to (u64)-1 before we exit through the
original exit path.
While we missed that assignment of (u64)-1 in the new exit path.

The end result is, the quota status item doesn't have the same value.
(-1 vs the last bytenr + 1)
Although it doesn't affect quota accounting, it's still better to keep
the original behavior.
Reported-by: NMisono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Fixes: ff3d27a0 ("btrfs: qgroup: Finish rescan when hit the last leaf of extent tree")
Signed-off-by: NQu Wenruo <wqu@suse.com>
Reviewed-by: NMisono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

kmemleak reported some memory leak on reading proc files. After adding
some debug lines, find that proc_seq_fops is using seq_release as
release handler, which won't handle the free of 'private' field of
seq_file, while in fact the open handler proc_seq_open could create
the private data with __seq_open_private when state_size is greater
than zero. So after reading files created with proc_create_seq_private,
such as /proc/timer_list and /proc/vmallocinfo, the private mem of a
seq_file is not freed. Fix it by adding the paired proc_seq_release
as the default release handler of proc_seq_ops instead of seq_release.

Fixes: 44414d82 ("proc: introduce proc_create_seq_private")
Reviewed-by: NChristoph Hellwig <hch@lst.de>
CC: Christoph Hellwig <hch@lst.de>
Signed-off-by: NChunyu Hu <chuhu@redhat.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

In any case, d_splice_alias() does not drop reference of original
dentry.
Signed-off-by: N"Yan, Zheng" <zyan@redhat.com>
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NIlya Dryomov <idryomov@gmail.com>

In __xfs_ag_resv_init we incorrectly calculate the amount by which to
decrease fdblocks when reserving blocks for the rmapbt.  Because rmapbt
allocations do not decrease fdblocks, we must decrease fdblocks by the
entire size of the requested reservation in order to achieve our goal of
always having enough free blocks to satisfy an rmapbt expansion.

This is in contrast to the refcountbt/finobt, which /do/ subtract from
fdblocks whenever they allocate a block.  For this allocation type we
preserve the existing behavior where we decrease fdblocks only by the
requested reservation minus the size of the existing tree.

This fixes the problem where the available block counts reported by
statfs change across a remount if there had been an rmapbt size change
since mount time.
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>

If a user asks us to zero_range part of a file, the end of the range is
EOF, and not aligned to a page boundary, invoke writeback of the EOF
page to ensure that the post-EOF part of the page is zeroed.  This
ensures that we don't expose stale memory contents via mmap, if in a
clumsy manner.

Found by running generic/127 when it runs zero_range and mapread at EOF
one after the other.
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>

In commit 8ad560d2 ("xfs: strengthen rtalloc query range checks")
we strengthened the input parameter checks in the rtbitmap range query
function, but introduced an off-by-one error in the process.  The call
to xfs_rtfind_forw deals with the high key being rextents, but we clamp
the high key to rextents - 1.  This causes the returned results to stop
one block short of the end of the rtdev, which is incorrect.
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Initialize the extent count field of the high key so that when we use
the high key to synthesize an 'unknown owner' record (i.e. used space
record) at the end of the queried range we have a field with which to
compute rm_blockcount.  This is not strictly necessary because the
synthesizer never uses the rm_blockcount field, but we can shut up the
static code analysis anyway.

Coverity-id: 1437358
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

The reflink iflag could have changed since the earlier unlocked check,
so if we got ILOCK_SHARED for a write and but we're now a reflink inode
we have to switch to ILOCK_EXCL and relock.

This helps us avoid blowing lock assertions in things like generic/166:

XFS: Assertion failed: xfs_isilocked(ip, XFS_ILOCK_EXCL), file: fs/xfs/xfs_reflink.c, line: 383
WARNING: CPU: 1 PID: 24707 at fs/xfs/xfs_message.c:104 assfail+0x25/0x30 [xfs]
Modules linked in: deadline_iosched dm_snapshot dm_bufio ext4 mbcache jbd2 dm_flakey xfs libcrc32c dax_pmem device_dax nd_pmem sch_fq_codel af_packet [last unloaded: scsi_debug]
CPU: 1 PID: 24707 Comm: xfs_io Not tainted 4.18.0-rc1-djw #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.10.2-1ubuntu1 04/01/2014
RIP: 0010:assfail+0x25/0x30 [xfs]
Code: ff 0f 0b c3 90 66 66 66 66 90 48 89 f1 41 89 d0 48 c7 c6 e8 ef 1b a0 48 89 fa 31 ff e8 54 f9 ff ff 80 3d fd ba 0f 00 00 75 03 <0f> 0b c3 0f 0b 66 0f 1f 44 00 00 66 66 66 66 90 48 63 f6 49 89 f9
RSP: 0018:ffffc90006423ad8 EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff880030b65e80 RCX: 0000000000000000
RDX: 00000000ffffffc0 RSI: 000000000000000a RDI: ffffffffa01b0447
RBP: ffffc90006423c10 R08: 0000000000000000 R09: 0000000000000000
R10: ffff88003d43fc30 R11: f000000000000000 R12: ffff880077cda000
R13: 0000000000000000 R14: ffffc90006423c30 R15: ffffc90006423bf9
FS:  00007feba8986800(0000) GS:ffff88003ec00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000000138ab58 CR3: 000000003d40a000 CR4: 00000000000006a0
Call Trace:
 xfs_reflink_allocate_cow+0x24c/0x3d0 [xfs]
 xfs_file_iomap_begin+0x6d2/0xeb0 [xfs]
 ? iomap_to_fiemap+0x80/0x80
 iomap_apply+0x5e/0x130
 iomap_dio_rw+0x2e0/0x400
 ? iomap_to_fiemap+0x80/0x80
 ? xfs_file_dio_aio_write+0x133/0x4a0 [xfs]
 xfs_file_dio_aio_write+0x133/0x4a0 [xfs]
 xfs_file_write_iter+0x7b/0xb0 [xfs]
 __vfs_write+0x16f/0x1f0
 vfs_write+0xc8/0x1c0
 ksys_pwrite64+0x74/0x90
 do_syscall_64+0x56/0x180
 entry_SYSCALL_64_after_hwframe+0x49/0xbe
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Zorro Lang reports that generic/485 blows an assert on a filesystem with
512 byte blocks.  The test tries to fallocate a post-eof extent at the
maximum file size and calls insert range to shift the extents right by
two blocks.  On a 512b block filesystem this causes startoff to overflow
the 54-bit startoff field, leading to the assert.

Therefore, always check the rightmost extent to see if it would overflow
prior to invoking the insert range machinery.

Reported-by: zlang@redhat.com
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=200137Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

If we somehow end up with a filesystem that has fewer free blocks than
the blocks set aside to avoid ENOSPC deadlocks, it's possible that the
free space calculation in xfs_reserve_blocks will spit out a negative
number (because percpu_counter_sum returns s64).  We fail to notice
this negative number and set fdblks_delta to it.  Now we increment
fdblocks(!) and the unsigned type of m_resblks means that we end up
setting a ridiculously huge m_resblks reservation.

Avoid this comedy of errors by detecting the negative free space and
returning -ENOSPC.
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

In commit e89c0413 ("xfs: implement the GETFSMAP ioctl") we
created the ability to obtain empty transactions.  These transactions
have no log or block reservations and therefore can't modify anything.
Since they're also NO_WRITECOUNT they can run while the fs is frozen,
so we don't need to WARN_ON about that usage.
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: NAllison Henderson <allison.henderson@oracle.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

If we failed during a rename exchange operation after starting/joining a
transaction, we would end up replacing the return value, stored in the
local 'ret' variable, with the return value from btrfs_end_transaction().
So this could end up returning 0 (success) to user space despite the
operation having failed and aborted the transaction, because if there are
multiple tasks having a reference on the transaction at the time
btrfs_end_transaction() is called by the rename exchange, that function
returns 0 (otherwise it returns -EIO and not the original error value).
So fix this by not overwriting the return value on error after getting
a transaction handle.

Fixes: cdd1fedf ("btrfs: add support for RENAME_EXCHANGE and RENAME_WHITEOUT")
CC: stable@vger.kernel.org # 4.9+
Signed-off-by: NFilipe Manana <fdmanana@suse.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

When a corrupt inode is detected during xfs_iflush_cluster, we can
get a shutdown ASSERT failure like this:

XFS (pmem1): Metadata corruption detected at xfs_symlink_shortform_verify+0x5c/0xa0, inode 0x86627 data fork
XFS (pmem1): Unmount and run xfs_repair
XFS (pmem1): xfs_do_force_shutdown(0x8) called from line 3372 of file fs/xfs/xfs_inode.c.  Return address = ffffffff814f4116
XFS (pmem1): Corruption of in-memory data detected.  Shutting down filesystem
XFS (pmem1): xfs_do_force_shutdown(0x1) called from line 222 of file fs/xfs/libxfs/xfs_defer.c.  Return address = ffffffff814a8a88
XFS (pmem1): xfs_do_force_shutdown(0x1) called from line 222 of file fs/xfs/libxfs/xfs_defer.c.  Return address = ffffffff814a8ef9
XFS (pmem1): Please umount the filesystem and rectify the problem(s)
XFS: Assertion failed: xfs_isiflocked(ip), file: fs/xfs/xfs_inode.h, line: 258
.....
Call Trace:
 xfs_iflush_abort+0x10a/0x110
 xfs_iflush+0xf3/0x390
 xfs_inode_item_push+0x126/0x1e0
 xfsaild+0x2c5/0x890
 kthread+0x11c/0x140
 ret_from_fork+0x24/0x30

Essentially, xfs_iflush_abort() has been called twice on the
original inode that that was flushed. This happens because the
inode has been flushed to teh buffer successfully via
xfs_iflush_int(), and so when another inode is detected as corrupt
in xfs_iflush_cluster, the buffer is marked stale and EIO, and
iodone callbacks are run on it.

Running the iodone callbacks walks across the original inode and
calls xfs_iflush_abort() on it. When xfs_iflush_cluster() returns
to xfs_iflush(), it runs the error path for that function, and that
calls xfs_iflush_abort() on the inode a second time, leading to the
above assert failure as the inode is not flush locked anymore.

This bug has been there a long time.

The simple fix would be to just avoid calling xfs_iflush_abort() in
xfs_iflush() if we've got a failure from xfs_iflush_cluster().
However, xfs_iflush_cluster() has magic delwri buffer handling that
means it may or may not have run IO completion on the buffer, and
hence sometimes we have to call xfs_iflush_abort() from
xfs_iflush(), and sometimes we shouldn't.

After reading through all the error paths and the delwri buffer
code, it's clear that the error handling in xfs_iflush_cluster() is
unnecessary. If the buffer is delwri, it leaves it on the delwri
list so that when the delwri list is submitted it sees a shutdown
fliesystem in xfs_buf_submit() and that marks the buffer stale, EIO
and runs IO completion. i.e. exactly what xfs+iflush_cluster() does
when it's not a delwri buffer. Further, marking a buffer stale
clears the _XBF_DELWRI_Q flag on the buffer, which means when
submission of the buffer occurs, it just skips over it and releases
it.

IOWs, the error handling in xfs_iflush_cluster doesn't need to care
if the buffer is already on a the delwri queue or not - it just
needs to mark the buffer stale, EIO and run completions. That means
we can just use the easy fix for xfs_iflush() to avoid the double
abort.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDarrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

When the inode is in extent format, it can't have more extents that
fit in the inode fork. We don't currenty check this, and so this
corruption goes unnoticed by the inode verifiers. This can lead to
crashes operating on invalid in-memory structures.

Attempts to access such a inode will now error out in the verifier
rather than allowing modification operations to proceed.
Reported-by: NWen Xu <wen.xu@gatech.edu>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NDarrick J. Wong <darrick.wong@oracle.com>
[darrick: fix a typedef, add some braces and breaks to shut up compiler warnings]
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

Instead of using xfs_bmapi_read to find delalloc extents and then punch
them out using xfs_bunmapi, opencode the loop to iterate over the extents
and call xfs_bmap_del_extent_delay directly.  This both simplifies the
code and reduces the number of extent tree lookups required.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDarrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>

If this condition ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) !=
		   (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM))
is hit, we will go to free the uninitialized cmp.src_pages and
cmp.dst_pages.

Fixes: 67b07bd4 ("Btrfs: reuse cmp workspace in EXTENT_SAME ioctl")
Signed-off-by: NLu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: NDavid Sterba <dsterba@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>