mirror of
https://github.com/torvalds/linux
synced 2024-10-16 16:29:20 +00:00
801bec365e
When doing an incremental send we can end up not moving directories that
have the same name. This happens when the same parent directory has
different child directories with the same name in the parent and send
snapshots.
For example, consider the following scenario:
Parent snapshot:
. (ino 256)
|---- d/ (ino 257)
| |--- p1/ (ino 258)
|
|---- p1/ (ino 259)
Send snapshot:
. (ino 256)
|--- d/ (ino 257)
|--- p1/ (ino 259)
|--- p1/ (ino 258)
The directory named "d" (inode 257) has in both snapshots an entry with
the name "p1" but it refers to different inodes in both snapshots (inode
258 in the parent snapshot and inode 259 in the send snapshot). When
attempting to move inode 258, the operation is delayed because its new
parent, inode 259, was not yet moved/renamed (as the stream is currently
processing inode 258). Then when processing inode 259, we also end up
delaying its move/rename operation so that it happens after inode 258 is
moved/renamed. This decision to delay the move/rename rename operation
of inode 259 is due to the fact that the new parent inode (257) still
has inode 258 as its child, which has the same name has inode 259. So
we end up with inode 258 move/rename operation waiting for inode's 259
move/rename operation, which in turn it waiting for inode's 258
move/rename. This results in ending the send stream without issuing
move/rename operations for inodes 258 and 259 and generating the
following warnings in syslog/dmesg:
[148402.979747] ------------[ cut here ]------------
[148402.980588] WARNING: CPU: 14 PID: 4117 at fs/btrfs/send.c:6177 btrfs_ioctl_send+0xe03/0xe51 [btrfs]
[148402.981928] Modules linked in: btrfs crc32c_generic xor raid6_pq acpi_cpufreq tpm_tis ppdev tpm parport_pc psmouse parport sg pcspkr i2c_piix4 i2c_core evdev processor serio_raw button loop autofs4 ext4 crc16 jbd2 mbcache sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy [last unloaded: btrfs]
[148402.986999] CPU: 14 PID: 4117 Comm: btrfs Tainted: G W 4.6.0-rc7-btrfs-next-31+ #1
[148402.988136] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[148402.988136] 0000000000000000 ffff88022139fca8 ffffffff8126b42c 0000000000000000
[148402.988136] 0000000000000000 ffff88022139fce8 ffffffff81052b14 000018212139fac8
[148402.988136] ffff88022b0db400 0000000000000000 0000000000000001 0000000000000000
[148402.988136] Call Trace:
[148402.988136] [<ffffffff8126b42c>] dump_stack+0x67/0x90
[148402.988136] [<ffffffff81052b14>] __warn+0xc2/0xdd
[148402.988136] [<ffffffff81052beb>] warn_slowpath_null+0x1d/0x1f
[148402.988136] [<ffffffffa04bc831>] btrfs_ioctl_send+0xe03/0xe51 [btrfs]
[148402.988136] [<ffffffffa048b358>] btrfs_ioctl+0x14f/0x1f81 [btrfs]
[148402.988136] [<ffffffff8108e456>] ? arch_local_irq_save+0x9/0xc
[148402.988136] [<ffffffff8108eb51>] ? __lock_is_held+0x3c/0x57
[148402.988136] [<ffffffff8118da05>] vfs_ioctl+0x18/0x34
[148402.988136] [<ffffffff8118e00c>] do_vfs_ioctl+0x550/0x5be
[148402.988136] [<ffffffff81196f0c>] ? __fget+0x6b/0x77
[148402.988136] [<ffffffff81196fa1>] ? __fget_light+0x62/0x71
[148402.988136] [<ffffffff8118e0d1>] SyS_ioctl+0x57/0x79
[148402.988136] [<ffffffff8149e025>] entry_SYSCALL_64_fastpath+0x18/0xa8
[148402.988136] [<ffffffff8108e89d>] ? trace_hardirqs_off_caller+0x3f/0xaa
[148403.011373] ---[ end trace a4539270c8056f8b ]---
[148403.012296] ------------[ cut here ]------------
[148403.013071] WARNING: CPU: 14 PID: 4117 at fs/btrfs/send.c:6194 btrfs_ioctl_send+0xe19/0xe51 [btrfs]
[148403.014447] Modules linked in: btrfs crc32c_generic xor raid6_pq acpi_cpufreq tpm_tis ppdev tpm parport_pc psmouse parport sg pcspkr i2c_piix4 i2c_core evdev processor serio_raw button loop autofs4 ext4 crc16 jbd2 mbcache sr_mod cdrom sd_mod ata_generic virtio_scsi ata_piix libata virtio_pci virtio_ring virtio e1000 scsi_mod floppy [last unloaded: btrfs]
[148403.019708] CPU: 14 PID: 4117 Comm: btrfs Tainted: G W 4.6.0-rc7-btrfs-next-31+ #1
[148403.020104] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[148403.020104] 0000000000000000 ffff88022139fca8 ffffffff8126b42c 0000000000000000
[148403.020104] 0000000000000000 ffff88022139fce8 ffffffff81052b14 000018322139fac8
[148403.020104] ffff88022b0db400 0000000000000000 0000000000000001 0000000000000000
[148403.020104] Call Trace:
[148403.020104] [<ffffffff8126b42c>] dump_stack+0x67/0x90
[148403.020104] [<ffffffff81052b14>] __warn+0xc2/0xdd
[148403.020104] [<ffffffff81052beb>] warn_slowpath_null+0x1d/0x1f
[148403.020104] [<ffffffffa04bc847>] btrfs_ioctl_send+0xe19/0xe51 [btrfs]
[148403.020104] [<ffffffffa048b358>] btrfs_ioctl+0x14f/0x1f81 [btrfs]
[148403.020104] [<ffffffff8108e456>] ? arch_local_irq_save+0x9/0xc
[148403.020104] [<ffffffff8108eb51>] ? __lock_is_held+0x3c/0x57
[148403.020104] [<ffffffff8118da05>] vfs_ioctl+0x18/0x34
[148403.020104] [<ffffffff8118e00c>] do_vfs_ioctl+0x550/0x5be
[148403.020104] [<ffffffff81196f0c>] ? __fget+0x6b/0x77
[148403.020104] [<ffffffff81196fa1>] ? __fget_light+0x62/0x71
[148403.020104] [<ffffffff8118e0d1>] SyS_ioctl+0x57/0x79
[148403.020104] [<ffffffff8149e025>] entry_SYSCALL_64_fastpath+0x18/0xa8
[148403.020104] [<ffffffff8108e89d>] ? trace_hardirqs_off_caller+0x3f/0xaa
[148403.038981] ---[ end trace a4539270c8056f8c ]---
There's another issue caused by similar (but more complex) changes in the
directory hierarchy that makes move/rename operations fail, described with
the following example:
Parent snapshot:
.
|---- a/ (ino 262)
| |---- c/ (ino 268)
|
|---- d/ (ino 263)
|---- ance/ (ino 267)
|---- e/ (ino 264)
|---- f/ (ino 265)
|---- ance/ (ino 266)
Send snapshot:
.
|---- a/ (ino 262)
|---- c/ (ino 268)
| |---- ance/ (ino 267)
|
|---- d/ (ino 263)
| |---- ance/ (ino 266)
|
|---- f/ (ino 265)
|---- e/ (ino 264)
When the inode 265 is processed, the path for inode 267 is computed, which
at that time corresponds to "d/ance", and it's stored in the names cache.
Later on when processing inode 266, we end up orphanizing (renaming to a
name matching the pattern o<ino>-<gen>-<seq>) inode 267 because it has
the same name as inode 266 and it's currently a child of the new parent
directory (inode 263) for inode 266. After the orphanization and while we
are still processing inode 266, a rename operation for inode 266 is
generated. However the source path for that rename operation is incorrect
because it ends up using the old, pre-orphanization, name of inode 267.
The no longer valid name for inode 267 was previously cached when
processing inode 265 and it remains usable and considered valid until
the inode currently being processed has a number greater than 267.
This resulted in the receiving side failing with the following error:
ERROR: rename d/ance/ance -> d/ance failed: No such file or directory
So fix these issues by detecting such circular dependencies for rename
operations and by clearing the cached name of an inode once the inode
is orphanized.
A test case for fstests will follow soon.
Signed-off-by: Robbie Ko <robbieko@synology.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
[Rewrote change log to be more detailed and organized, and improved
comments]
Signed-off-by: Filipe Manana <fdmanana@suse.com>
|
||
|---|---|---|
| .. | ||
| tests | ||
| acl.c | ||
| async-thread.c | ||
| async-thread.h | ||
| backref.c | ||
| backref.h | ||
| btrfs_inode.h | ||
| check-integrity.c | ||
| check-integrity.h | ||
| compression.c | ||
| compression.h | ||
| ctree.c | ||
| ctree.h | ||
| delayed-inode.c | ||
| delayed-inode.h | ||
| delayed-ref.c | ||
| delayed-ref.h | ||
| dev-replace.c | ||
| dev-replace.h | ||
| dir-item.c | ||
| disk-io.c | ||
| disk-io.h | ||
| export.c | ||
| export.h | ||
| extent-tree.c | ||
| extent_io.c | ||
| extent_io.h | ||
| extent_map.c | ||
| extent_map.h | ||
| file-item.c | ||
| file.c | ||
| free-space-cache.c | ||
| free-space-cache.h | ||
| free-space-tree.c | ||
| free-space-tree.h | ||
| hash.c | ||
| hash.h | ||
| inode-item.c | ||
| inode-map.c | ||
| inode-map.h | ||
| inode.c | ||
| ioctl.c | ||
| Kconfig | ||
| locking.c | ||
| locking.h | ||
| lzo.c | ||
| Makefile | ||
| math.h | ||
| ordered-data.c | ||
| ordered-data.h | ||
| orphan.c | ||
| print-tree.c | ||
| print-tree.h | ||
| props.c | ||
| props.h | ||
| qgroup.c | ||
| qgroup.h | ||
| raid56.c | ||
| raid56.h | ||
| rcu-string.h | ||
| reada.c | ||
| relocation.c | ||
| root-tree.c | ||
| scrub.c | ||
| send.c | ||
| send.h | ||
| struct-funcs.c | ||
| super.c | ||
| sysfs.c | ||
| sysfs.h | ||
| transaction.c | ||
| transaction.h | ||
| tree-defrag.c | ||
| tree-log.c | ||
| tree-log.h | ||
| ulist.c | ||
| ulist.h | ||
| uuid-tree.c | ||
| volumes.c | ||
| volumes.h | ||
| xattr.c | ||
| xattr.h | ||
| zlib.c | ||