When admin calls "reboot -f" - i.e., does a hard system reboot by
directly calling reboot(2) - ext4 filesystem mounted with errors=panic
can panic the system. This happens because the underlying device gets
disabled without unmounting the filesystem and thus some syscall running
in parallel to reboot(2) can result in the filesystem getting IO errors.
This is somewhat surprising to the users so try improve the behavior by
switching to errors=remount-ro behavior when the system is running
reboot(2).
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Ext4 needs to serialize unaligned direct AIO because the zeroing of
partial blocks of two competing unaligned AIOs can result in data
corruption.
However it decides not to serialize if the potentially unaligned aio is
past i_size with the rationale that no pending writes are possible past
i_size. Unfortunately if the i_size is not block aligned and the second
unaligned write lands past i_size, but still into the same block, it has
the potential of corrupting the previous unaligned write to the same
block.
This is (very simplified) reproducer from Frank
// 41472 = (10 * 4096) + 512
// 37376 = 41472 - 4096
ftruncate(fd, 41472);
io_prep_pwrite(iocbs[0], fd, buf[0], 4096, 37376);
io_prep_pwrite(iocbs[1], fd, buf[1], 4096, 41472);
io_submit(io_ctx, 1, &iocbs[1]);
io_submit(io_ctx, 1, &iocbs[2]);
io_getevents(io_ctx, 2, 2, events, NULL);
Without this patch the 512B range from 40960 up to the start of the
second unaligned write (41472) is going to be zeroed overwriting the data
written by the first write. This is a data corruption.
00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
*
00009200 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30
*
0000a000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
*
0000a200 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
With this patch the data corruption is avoided because we will recognize
the unaligned_aio and wait for the unwritten extent conversion.
00000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
*
00009200 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30
*
0000a200 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31 31
*
0000b200
Reported-by: Frank Sorenson <fsorenso@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Fixes: e9e3bcecf4 ("ext4: serialize unaligned asynchronous DIO")
Cc: stable@vger.kernel.org
We see the following NULL pointer dereference while running xfstests
generic/475:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
PGD 8000000c84bad067 P4D 8000000c84bad067 PUD c84e62067 PMD 0
Oops: 0000 [#1] SMP PTI
CPU: 7 PID: 9886 Comm: fsstress Kdump: loaded Not tainted 5.0.0-rc8 #10
RIP: 0010:ext4_do_update_inode+0x4ec/0x760
...
Call Trace:
? jbd2_journal_get_write_access+0x42/0x50
? __ext4_journal_get_write_access+0x2c/0x70
? ext4_truncate+0x186/0x3f0
ext4_mark_iloc_dirty+0x61/0x80
ext4_mark_inode_dirty+0x62/0x1b0
ext4_truncate+0x186/0x3f0
? unmap_mapping_pages+0x56/0x100
ext4_setattr+0x817/0x8b0
notify_change+0x1df/0x430
do_truncate+0x5e/0x90
? generic_permission+0x12b/0x1a0
This is triggered because the NULL pointer handle->h_transaction was
dereferenced in function ext4_update_inode_fsync_trans().
I found that the h_transaction was set to NULL in jbd2__journal_restart
but failed to attached to a new transaction while the journal is aborted.
Fix this by checking the handle before updating the inode.
Fixes: b436b9bef8 ("ext4: Wait for proper transaction commit on fsync")
Signed-off-by: Jiufei Xue <jiufei.xue@linux.alibaba.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: stable@kernel.org
We have a customer reporting crashes in lock_get_status() with many
"Leaked POSIX lock" messages preceeding the crash.
Leaked POSIX lock on dev=0x0:0x56 ...
Leaked POSIX lock on dev=0x0:0x56 ...
Leaked POSIX lock on dev=0x0:0x56 ...
Leaked POSIX lock on dev=0x0:0x53 ...
Leaked POSIX lock on dev=0x0:0x53 ...
Leaked POSIX lock on dev=0x0:0x53 ...
Leaked POSIX lock on dev=0x0:0x53 ...
POSIX: fl_owner=ffff8900e7b79380 fl_flags=0x1 fl_type=0x1 fl_pid=20709
Leaked POSIX lock on dev=0x0:0x4b ino...
Leaked locks on dev=0x0:0x4b ino=0xf911400000029:
POSIX: fl_owner=ffff89f41c870e00 fl_flags=0x1 fl_type=0x1 fl_pid=19592
stack segment: 0000 [#1] SMP
Modules linked in: binfmt_misc msr tcp_diag udp_diag inet_diag unix_diag af_packet_diag netlink_diag rpcsec_gss_krb5 arc4 ecb auth_rpcgss nfsv4 md4 nfs nls_utf8 lockd grace cifs sunrpc ccm dns_resolver fscache af_packet iscsi_ibft iscsi_boot_sysfs vmw_vsock_vmci_transport vsock xfs libcrc32c sb_edac edac_core crct10dif_pclmul crc32_pclmul ghash_clmulni_intel drbg ansi_cprng vmw_balloon aesni_intel aes_x86_64 lrw gf128mul glue_helper ablk_helper cryptd joydev pcspkr vmxnet3 i2c_piix4 vmw_vmci shpchp fjes processor button ac btrfs xor raid6_pq sr_mod cdrom ata_generic sd_mod ata_piix vmwgfx crc32c_intel drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops ttm serio_raw ahci libahci drm libata vmw_pvscsi sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod autofs4
Supported: Yes
CPU: 6 PID: 28250 Comm: lsof Not tainted 4.4.156-94.64-default #1
Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 04/05/2016
task: ffff88a345f28740 ti: ffff88c74005c000 task.ti: ffff88c74005c000
RIP: 0010:[<ffffffff8125dcab>] [<ffffffff8125dcab>] lock_get_status+0x9b/0x3b0
RSP: 0018:ffff88c74005fd90 EFLAGS: 00010202
RAX: ffff89bde83e20ae RBX: ffff89e870003d18 RCX: 0000000049534f50
RDX: ffffffff81a3541f RSI: ffffffff81a3544e RDI: ffff89bde83e20ae
RBP: 0026252423222120 R08: 0000000020584953 R09: 000000000000ffff
R10: 0000000000000000 R11: ffff88c74005fc70 R12: ffff89e5ca7b1340
R13: 00000000000050e5 R14: ffff89e870003d30 R15: ffff89e5ca7b1340
FS: 00007fafd64be800(0000) GS:ffff89f41fd00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000001c80018 CR3: 000000a522048000 CR4: 0000000000360670
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Stack:
0000000000000208 ffffffff81a3d6b6 ffff89e870003d30 ffff89e870003d18
ffff89e5ca7b1340 ffff89f41738d7c0 ffff89e870003d30 ffff89e5ca7b1340
ffffffff8125e08f 0000000000000000 ffff89bc22b67d00 ffff88c74005ff28
Call Trace:
[<ffffffff8125e08f>] locks_show+0x2f/0x70
[<ffffffff81230ad1>] seq_read+0x251/0x3a0
[<ffffffff81275bbc>] proc_reg_read+0x3c/0x70
[<ffffffff8120e456>] __vfs_read+0x26/0x140
[<ffffffff8120e9da>] vfs_read+0x7a/0x120
[<ffffffff8120faf2>] SyS_read+0x42/0xa0
[<ffffffff8161cbc3>] entry_SYSCALL_64_fastpath+0x1e/0xb7
When Linux closes a FD (close(), close-on-exec, dup2(), ...) it calls
filp_close() which also removes all posix locks.
The lock struct is initialized like so in filp_close() and passed
down to cifs
...
lock.fl_type = F_UNLCK;
lock.fl_flags = FL_POSIX | FL_CLOSE;
lock.fl_start = 0;
lock.fl_end = OFFSET_MAX;
...
Note the FL_CLOSE flag, which hints the VFS code that this unlocking
is done for closing the fd.
filp_close()
locks_remove_posix(filp, id);
vfs_lock_file(filp, F_SETLK, &lock, NULL);
return filp->f_op->lock(filp, cmd, fl) => cifs_lock()
rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock, xid);
rc = server->ops->mand_unlock_range(cfile, flock, xid);
if (flock->fl_flags & FL_POSIX && !rc)
rc = locks_lock_file_wait(file, flock)
Notice how we don't call locks_lock_file_wait() which does the
generic VFS lock/unlock/wait work on the inode if rc != 0.
If we are closing the handle, the SMB server is supposed to remove any
locks associated with it. Similarly, cifs.ko frees and wakes up any
lock and lock waiter when closing the file:
cifs_close()
cifsFileInfo_put(file->private_data)
/*
* Delete any outstanding lock records. We'll lose them when the file
* is closed anyway.
*/
down_write(&cifsi->lock_sem);
list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
list_del(&li->llist);
cifs_del_lock_waiters(li);
kfree(li);
}
list_del(&cifs_file->llist->llist);
kfree(cifs_file->llist);
up_write(&cifsi->lock_sem);
So we can safely ignore unlocking failures in cifs_lock() if they
happen with the FL_CLOSE flag hint set as both the server and the
client take care of it during the actual closing.
This is not a proper fix for the unlocking failure but it's safe and
it seems to prevent the lock leakages and crashes the customer
experiences.
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: NeilBrown <neil@brown.name>
Signed-off-by: Steve French <stfrench@microsoft.com>
Acked-by: Pavel Shilovsky <pshilov@microsoft.com>
For debugging purposes we often have to be able to query
additional information only available via SMB3 FSCTL
from the server from user space tools (e.g. like
cifs-utils's smbinfo). See MS-FSCC and MS-SMB2 protocol
specifications for more details.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
smb2_set_sparse does not return -errno, it returns a boolean where
true means success.
Change this to just ignore the return value just like the other callsites.
Additionally add code to handle the case where we must set the file sparse
and possibly also extending it.
Fixes xfstests: generic/236 generic/350 generic/420
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
As Sergey Senozhatsky pointed out __constant_cpu_to_le32()
is misspelled in a few definitions in the list of status
codes smb2status.h as __constanst_cpu_to_le32()
Signed-off-by: Steve French <stfrench@microsoft.com>
CC: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
This cleanup removes cifs specific code from SMB2/SMB3 code paths
which is cleaner and easier to maintain as the code to handle
special files is improved. Below is an example creating special files
using 'sfu' mount option over SMB3 to Windows (with this patch)
(Note that to Samba server, support for saving dos attributes
has to be enabled for the SFU mount option to work).
In the future this will also make implementation of creating
special files as reparse points easier (as Windows NFS server does
for example).
root@smf-Thinkpad-P51:~# stat -c "%F" /mnt2/char
character special file
root@smf-Thinkpad-P51:~# stat -c "%F" /mnt2/block
block special file
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Detected by CoverityScan CID#1438719 ("Unused Value")
buf is reset again before being used so these two lines of code
are useless.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
The passthrough queries from user space tools like smbinfo can be either
SMB3 QUERY_INFO or SMB3 FSCTL, but we are not checking for the latter.
Temporarily we return EOPNOTSUPP for SMB3 FSCTL passthrough requests
but once compounding fsctls is fixed can enable.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Can be helpful in debugging various xfstests that are currently
skipped or failing due to missing features in our current
implementation of fallocate.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
This allows fallocate -z to work against a Windows2016 share.
This is due to the SMB3 ZERO_RANGE command does not modify the filesize.
To address this we will now append a compounded SET-INFO to update the
end-of-file information.
This brings xfstests generic/469 closer to working against a windows share.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Define an _init() and a _free() function for SMB2_init so that we will
be able to use it with compounds.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Adds trace points for enter and exit (done vs. error) for:
compounded query and setinfo, hardlink, rename,
mkdir, rmdir, set_eof, delete (unlink)
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
When we open the shared root handle also ask for FILE_ALL_INFORMATION since
we can do this at zero cost as part of a compound.
Cache this information as long as the lease is held and return and serve any
future requests from cache.
This allows us to serve "stat /<mountpoint>" directly from cache and avoid
a network roundtrip. Since clients often want to do this quite a lot
this improve performance slightly.
As an example: xfstest generic/533 performs 43 stat operations on the root
of the share while it is run. Which are eliminated with this patch.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
It can be helpful for debugging. According to MS-FSCC:
"A 32-bit unsigned integer that contains the serial number of the
volume. The serial number is an opaque value generated by the file
system at format time"
Signed-off-by: Steve French <stfrench@microsoft.com>
Acked-by: Pavel Shilovsky <pshilov@microsoft.com>
Since we can now wait for multiple requests atomically in
wait_for_free_request() we can now greatly simplify the handling
of the credits in this function.
This fixes a potential deadlock where many concurrent compound requests
could each have reserved 1 or 2 credits each but are all blocked
waiting for the final credits they need to be able to issue the requests
to the server.
Set a default timeout of 60 seconds for compounded requests.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
To help debug credit starvation problems where we timeout
waiting for server to grant the client credits.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
When the server required encryption (but we didn't connect to it with the
"seal" mount option) we weren't displaying in /proc/fs/cifs/DebugData that
the tcon for that share was encrypted. Similarly we were not displaying
that signing was required when ses->sign was enabled (we only
checked ses->server->sign). This makes it easier to debug when in
fact the connection is signed (or sealed), whether for performance
or security questions.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Reserve the last MAX_COMPOUND credits for any request asking for >1 credit.
This is to prevent future compound requests from becoming starved while waiting
for potentially many requests is there is a large number of concurrent
singe-credit requests.
However, we need to protect from servers that are very slow to hand out
new credits on new sessions so we only do this IFF there are 2*MAX_COMPOUND
(arbitrary) credits already in flight.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Change wait_for_free_credits() to allow waiting for >=1 credits instead of just
a single credit.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Android uses pin_file for uncrypt during OTA, and that should be managed by
CAP_SYS_ADMIN only.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Architectures like ppc64 use the deposited page table to store hardware
page table slot information. Make sure we deposit a page table when
using zero page at the pmd level for hash.
Without this we hit
Unable to handle kernel paging request for data at address 0x00000000
Faulting instruction address: 0xc000000000082a74
Oops: Kernel access of bad area, sig: 11 [#1]
....
NIP [c000000000082a74] __hash_page_thp+0x224/0x5b0
LR [c0000000000829a4] __hash_page_thp+0x154/0x5b0
Call Trace:
hash_page_mm+0x43c/0x740
do_hash_page+0x2c/0x3c
copy_from_iter_flushcache+0xa4/0x4a0
pmem_copy_from_iter+0x2c/0x50 [nd_pmem]
dax_copy_from_iter+0x40/0x70
dax_iomap_actor+0x134/0x360
iomap_apply+0xfc/0x1b0
dax_iomap_rw+0xac/0x130
ext4_file_write_iter+0x254/0x460 [ext4]
__vfs_write+0x120/0x1e0
vfs_write+0xd8/0x220
SyS_write+0x6c/0x110
system_call+0x3c/0x130
Fixes: b5beae5e22 ("powerpc/pseries: Add driver for PAPR SCM regions")
Cc: <stable@vger.kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Pull filesystem-dax updates from Dan Williams:
- Fix handling of PMD-sized entries in the Xarray that lead to a crash
scenario
- Miscellaneous cleanups and small fixes
* tag 'fsdax-for-5.1' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
dax: Flush partial PMDs correctly
fs/dax: NIT fix comment regarding start/end vs range
fs/dax: Convert to use vmf_error()
Pull UBI and UBIFS updates from Richard Weinberger:
- A new interface for UBI to deal better with read disturb
- Reject unsupported ioctl flags in UBIFS (xfstests found it)
* tag 'upstream-5.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/ubifs:
ubi: wl: Silence uninitialized variable warning
ubifs: Reject unsupported ioctl flags explicitly
ubi: Expose the bitrot interface
ubi: Introduce in_pq()
As readahead is an optimization, all errors are usually filtered out,
but still properly handled when the real read call is done. The commit
5e9d398240 ("btrfs: readpages() should submit IO as read-ahead") added
REQ_RAHEAD to readpages() because that's only used for readahead
(despite what one would expect from the callback name).
This causes a flood of messages and inflated read error stats, so skip
reporting in case it's readahead.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202403
Reported-by: LimeTech <tomm@lime-technology.com>
Fixes: 5e9d398240 ("btrfs: readpages() should submit IO as read-ahead")
CC: stable@vger.kernel.org # 4.19+
Signed-off-by: David Sterba <dsterba@suse.com>
When we are mixing buffered writes with direct IO writes against the same
file and snapshotting is happening concurrently, we can end up with a
corrupt file content in the snapshot. Example:
1) Inode/file is empty.
2) Snapshotting starts.
2) Buffered write at offset 0 length 256Kb. This updates the i_size of the
inode to 256Kb, disk_i_size remains zero. This happens after the task
doing the snapshot flushes all existing delalloc.
3) DIO write at offset 256Kb length 768Kb. Once the ordered extent
completes it sets the inode's disk_i_size to 1Mb (256Kb + 768Kb) and
updates the inode item in the fs tree with a size of 1Mb (which is
the value of disk_i_size).
4) The dealloc for the range [0, 256Kb[ did not start yet.
5) The transaction used in the DIO ordered extent completion, which updated
the inode item, is committed by the snapshotting task.
6) Snapshot creation completes.
7) Dealloc for the range [0, 256Kb[ is flushed.
After that when reading the file from the snapshot we always get zeroes for
the range [0, 256Kb[, the file has a size of 1Mb and the data written by
the direct IO write is found. From an application's point of view this is
a corruption, since in the source subvolume it could never read a version
of the file that included the data from the direct IO write without the
data from the buffered write included as well. In the snapshot's tree,
file extent items are missing for the range [0, 256Kb[.
The issue, obviously, does not happen when using the -o flushoncommit
mount option.
Fix this by flushing delalloc for all the roots that are about to be
snapshotted when committing a transaction. This guarantees total ordering
when updating the disk_i_size of an inode since the flush for dealloc is
done when a transaction is in the TRANS_STATE_COMMIT_START state and wait
is done once no more external writers exist. This is similar to what we
do when using the flushoncommit mount option, but we do it only if the
transaction has snapshots to create and only for the roots of the
subvolumes to be snapshotted. The bulk of the dealloc is flushed in the
snapshot creation ioctl, so the flush work we do inside the transaction
is minimized.
This issue, involving buffered and direct IO writes with snapshotting, is
often triggered by fstest btrfs/078, and got reported by fsck when not
using the NO_HOLES features, for example:
$ cat results/btrfs/078.full
(...)
_check_btrfs_filesystem: filesystem on /dev/sdc is inconsistent
*** fsck.btrfs output ***
[1/7] checking root items
[2/7] checking extents
[3/7] checking free space cache
[4/7] checking fs roots
root 258 inode 264 errors 100, file extent discount
Found file extent holes:
start: 524288, len: 65536
ERROR: errors found in fs roots
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When Filipe added the recursive directory logging stuff in
2f2ff0ee5e ("Btrfs: fix metadata inconsistencies after directory
fsync") he specifically didn't take the directory i_mutex for the
children directories that we need to log because of lockdep. This is
generally fine, but can lead to this WARN_ON() tripping if we happen to
run delayed deletion's in between our first search and our second search
of dir_item/dir_indexes for this directory. We expect this to happen,
so the WARN_ON() isn't necessary. Drop the WARN_ON() and add a comment
so we know why this case can happen.
CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we do a shrinking truncate against an inode which is already present
in the respective log tree and then rename it, as part of logging the new
name we end up logging an inode item that reflects the old size of the
file (the one which we previously logged) and not the new smaller size.
The decision to preserve the size previously logged was added by commit
1a4bcf470c ("Btrfs: fix fsync data loss after adding hard link to
inode") in order to avoid data loss after replaying the log. However that
decision is only needed for the case the logged inode size is smaller then
the current size of the inode, as explained in that commit's change log.
If the current size of the inode is smaller then the previously logged
size, we know a shrinking truncate happened and therefore need to use
that smaller size.
Example to trigger the problem:
$ mkfs.btrfs -f /dev/sdb
$ mount /dev/sdb /mnt
$ xfs_io -f -c "pwrite -S 0xab 0 8000" /mnt/foo
$ xfs_io -c "fsync" /mnt/foo
$ xfs_io -c "truncate 3000" /mnt/foo
$ mv /mnt/foo /mnt/bar
$ xfs_io -c "fsync" /mnt/bar
<power failure>
$ mount /dev/sdb /mnt
$ od -t x1 -A d /mnt/bar
0000000 ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab ab
*
0008000
Once we rename the file, we log its name (and inode item), and because
the inode was already logged before in the current transaction, we log it
with a size of 8000 bytes because that is the size we previously logged
(with the first fsync). As part of the rename, besides logging the inode,
we do also sync the log, which is done since commit d4682ba03e
("Btrfs: sync log after logging new name"), so the next fsync against our
inode is effectively a no-op, since no new changes happened since the
rename operation. Even if did not sync the log during the rename
operation, the same problem (fize size of 8000 bytes instead of 3000
bytes) would be visible after replaying the log if the log ended up
getting synced to disk through some other means, such as for example by
fsyncing some other modified file. In the example above the fsync after
the rename operation is there just because not every filesystem may
guarantee logging/journalling the inode (and syncing the log/journal)
during the rename operation, for example it is needed for f2fs, but not
for ext4 and xfs.
Fix this scenario by, when logging a new name (which is triggered by
rename and link operations), using the current size of the inode instead
of the previously logged inode size.
A test case for fstests follows soon.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202695
CC: stable@vger.kernel.org # 4.4+
Reported-by: Seulbae Kim <seulbae@gatech.edu>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
With below testcase, we will fail to find existed xattr entry:
1. mkfs.f2fs -O extra_attr -O flexible_inline_xattr /dev/zram0
2. mount -t f2fs -o inline_xattr_size=1 /dev/zram0 /mnt/f2fs/
3. touch /mnt/f2fs/file
4. setfattr -n "user.name" -v 0 /mnt/f2fs/file
5. getfattr -n "user.name" /mnt/f2fs/file
/mnt/f2fs/file: user.name: No such attribute
The reason is for inode which has very small inline xattr size,
__find_inline_xattr() will fail to traverse any entry due to first
entry may not be loaded from xattr node yet, later, we may skip to
check entire xattr datas in __find_xattr(), result in such wrong
condition.
This patch adds condition to check such case to avoid this issue.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Paul Bandha reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=202709
When I run the poc on the mounted f2fs img I get a buffer overflow in
read_inline_xattr due to there being no sanity check on the value of
i_inline_xattr_size.
I created the img by just modifying the value of i_inline_xattr_size
in the inode:
i_name [test1.txt]
i_ext: fofs:0 blkaddr:0 len:0
i_extra_isize [0x 18 : 24]
i_inline_xattr_size [0x ffff : 65535]
i_addr[ofs] [0x 0 : 0]
mkdir /mnt/f2fs
mount ./f2fs1.img /mnt/f2fs
gcc poc.c -o poc
./poc
int main() {
int y = syscall(SYS_listxattr, "/mnt/f2fs/test1.txt", NULL, 0);
printf("ret %d", y);
printf("errno: %d\n", errno);
}
BUG: KASAN: slab-out-of-bounds in read_inline_xattr+0x18f/0x260
Read of size 262140 at addr ffff88011035efd8 by task f2fs1poc/3263
CPU: 0 PID: 3263 Comm: f2fs1poc Not tainted 4.18.0-custom #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.11.1-0-g0551a4be2c-prebuilt.qemu-project.org 04/01/2014
Call Trace:
dump_stack+0x71/0xab
print_address_description+0x83/0x250
kasan_report+0x213/0x350
memcpy+0x1f/0x50
read_inline_xattr+0x18f/0x260
read_all_xattrs+0xba/0x190
f2fs_listxattr+0x9d/0x3f0
listxattr+0xb2/0xd0
path_listxattr+0x93/0xe0
do_syscall_64+0x9d/0x220
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Let's add sanity check for inode.i_inline_xattr_size during f2fs_iget()
to avoid this issue.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds some kernel messages when user sets wrong inline_xattr_size.
Fixes: 500e0b28ec ("f2fs: fix to check inline_xattr_size boundary correctly")
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_mpage_readpages(), if page is beyond EOF, we should just
zero out it, but previously, before checking previous mapping
info, we missed to check filesize boundary, fix it.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Gao Xiang reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=202749
f2fs may skip pageout() due to incorrect page reference count.
The problem here is that MM defined the rule [1] very clearly that
once page was set with PG_private flag, we should increment the
refcount in that page, also main flows like pageout(), migrate_page()
will assume there is one additional page reference count if
page_has_private() returns true.
But currently, f2fs won't add/del refcount when changing PG_private
flag. Anyway, f2fs should follow MM's rule to make MM's related flows
running as expected.
[1] https://lore.kernel.org/lkml/2b19b3c4-2bc4-15fa-15cc-27a13e5c7af1@aol.com/
Reported-by: Gao Xiang <gaoxiang25@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Since 8c242db9b8 ("f2fs: fix stale ATOMIC_WRITTEN_PAGE private pointer"),
we've started to not skip clear private flag for atomic_write page
truncation, so removing old wrong comment in f2fs_invalidate_page().
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Jiqun Li reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=202747
System can panic due to using wrong allocate/free function pair
in xattr interface:
- use kvmalloc to allocate memory
- use kzfree to free memory
Let's fix to use kvfree instead of kzfree, BTW, we are safe to
get rid of kzfree, since there is no such confidential data stored
as xattr, we don't need to zero it before free memory.
Fixes: 5222595d09 ("f2fs: use kvmalloc, if kmalloc is failed")
Reported-by: Jiqun Li <jiqun.li@unisoc.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As Seulbae Kim reported in bugzilla:
https://bugzilla.kernel.org/show_bug.cgi?id=202637
We didn't recover permission field correctly after sudden power-cut,
the reason is in setattr we didn't add inode into global dirty list
once i_mode is changed, so latter checkpoint triggered by fsync will
not flush last i_mode into disk, result in this problem, fix it.
Reported-by: Seulbae Kim <seulbae@gatech.edu>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This follows to give random number to i_generation along with commit
2325306802 ("ext4: improve smp scalability for inode generation")
This can be used for DUN for UFS HW encryption.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
VFS will take inode_lock for readdir, therefore no need to
take page lock in readdir at all just as the majority of
other generic filesystems.
This patch improves concurrency since .iterate_shared
was introduced to VFS years ago.
Signed-off-by: Gao Xiang <gaoxiang25@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In error path of IPU, we didn't account iostat correctly, fix it.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For IPU path of f2fs_do_write_data_page(), in its error path, we
need to release encrypted page and fscrypt context, otherwise it
will cause memory leak.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch changes to allow failure of f2fs_bio_alloc() in
__submit_flush_wait(), which can simulate flush error in checkpoint()
for covering more error paths.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
With current retry mechanism in f2fs_fill_super, first fill_super
fails due to no memory, then second fill_super runs w/o recovery,
if we succeed, we may lose fsynced data, it doesn't make sense.
Let's retry fill_super only if it occurs non-ENOMEM error during
recovery.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
