OSDN Git Service

sagit-ice-cold/kernel_xiaomi_msm8998.git
6 years agof2fs: clear FI_HOT_DATA correctly
Chao Yu [Fri, 18 Aug 2017 15:37:36 +0000 (23:37 +0800)]
f2fs: clear FI_HOT_DATA correctly

This patch fixes to clear FI_HOT_DATA correctly in below path:
- error handling in f2fs_ioc_start_atomic_write
- after commit atomic write in f2fs_ioc_commit_atomic_write
- after drop atomic write in drop_inmem_pages

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: fix out-of-order execution in f2fs_issue_flush
Chao Yu [Mon, 21 Aug 2017 14:53:45 +0000 (22:53 +0800)]
f2fs: fix out-of-order execution in f2fs_issue_flush

In f2fs_issue_flush, due to out-of-order execution of CPU, wake_up can
be called before we insert issue_list, result in long latency of
wait_for_completion. Fix this by adding smp_mb() to force the order of
related codes.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: issue discard commands if gc_urgent is set
Jaegeuk Kim [Wed, 16 Aug 2017 04:27:19 +0000 (21:27 -0700)]
f2fs: issue discard commands if gc_urgent is set

It's time to issue all the discard commands, if user sets the idle time.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: introduce discard_granularity sysfs entry
Chao Yu [Mon, 7 Aug 2017 15:09:56 +0000 (23:09 +0800)]
f2fs: introduce discard_granularity sysfs entry

Commit d618ebaf0aa8 ("f2fs: enable small discard by default") enables
f2fs to issue 4K size discard in real-time discard mode. However, issuing
smaller discard may cost more lifetime but releasing less free space in
flash device. Since f2fs has ability of separating hot/cold data and
garbage collection, we can expect that small-sized invalid region would
expand soon with OPU, deletion or garbage collection on valid datas, so
it's better to delay or skip issuing smaller size discards, it could help
to reduce overmuch consumption of IO bandwidth and lifetime of flash
storage.

This patch makes f2fs selectng 64K size as its default minimal
granularity, and issue discard with the size which is not smaller than
minimal granularity. Also it exposes discard granularity as sysfs entry
for configuration in different scenario.

Jaegeuk Kim:
 We must issue all the accumulated discard commands when fstrim is called.
 So, I've added pend_list_tag[] to indicate whether we should issue the
 commands or not. If tag sets P_ACTIVE or P_TRIM, we have to issue them.
 P_TRIM is set once at a time, given fstrim trigger.
 In addition, issue_discard_thread is calling too much due to the number of
 discard commands remaining in the pending list. I added a timer to control
 it likewise gc_thread.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: remove unused function overprovision_sections
Yunlong Song [Mon, 14 Aug 2017 08:52:43 +0000 (16:52 +0800)]
f2fs: remove unused function overprovision_sections

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: check hot_data for roll-forward recovery
Jaegeuk Kim [Sun, 13 Aug 2017 04:33:23 +0000 (21:33 -0700)]
f2fs: check hot_data for roll-forward recovery

We need to check HOT_DATA to truncate any previous data block when doing
roll-forward recovery.

Cc: <stable@vger.kernel.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: add tracepoint for f2fs_gc
Chao Yu [Fri, 11 Aug 2017 10:00:15 +0000 (18:00 +0800)]
f2fs: add tracepoint for f2fs_gc

This patch adds tracepoint for f2fs_gc.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: retry to revoke atomic commit in -ENOMEM case
Chao Yu [Tue, 8 Aug 2017 11:09:08 +0000 (19:09 +0800)]
f2fs: retry to revoke atomic commit in -ENOMEM case

During atomic committing, if we encounter -ENOMEM in revoke path, it's
better to give a chance to retry revoking.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: let fill_super handle roll-forward errors
Jaegeuk Kim [Fri, 11 Aug 2017 00:35:04 +0000 (17:35 -0700)]
f2fs: let fill_super handle roll-forward errors

If we set CP_ERROR_FLAG in roll-forward error, f2fs is no longer to proceed
any IOs due to f2fs_cp_error(). But, for example, if some stale data is involved
on roll-forward process, we're able to get -ENOENT, getting fs stuck.
If we get any error, let fill_super set SBI_NEED_FSCK and try to recover back
to stable point.

Cc: <stable@vger.kernel.org>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: merge equivalent flags F2FS_GET_BLOCK_[READ|DIO]
Qiuyang Sun [Wed, 9 Aug 2017 09:27:30 +0000 (17:27 +0800)]
f2fs: merge equivalent flags F2FS_GET_BLOCK_[READ|DIO]

Currently, the two flags F2FS_GET_BLOCK_[READ|DIO] are totally equivalent
and can be used interchangably in all scenarios they are involved in.
Neither of the flags is referenced in f2fs_map_blocks(), making them both
the default case. To remove the ambiguity, this patch merges both flags
into F2FS_GET_BLOCK_DEFAULT, and introduces an enum for all distinct flags.

Signed-off-by: Qiuyang Sun <sunqiuyang@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: support journalled quota
Chao Yu [Tue, 8 Aug 2017 02:54:31 +0000 (10:54 +0800)]
f2fs: support journalled quota

This patch supports to enable f2fs to accept quota information through
mount option:
- {usr,grp,prj}jquota=<quota file path>
- jqfmt=<quota type>

Then, in ->mount flow, we can recover quota file during log replaying,
by this, journelled quota can be supported.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: Fix wrong return values.]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: fix potential overflow when adjusting GC cycle
Chao Yu [Mon, 7 Aug 2017 15:12:46 +0000 (23:12 +0800)]
f2fs: fix potential overflow when adjusting GC cycle

While comparing signed and unsigned variables, compiler will converts the
signed value to unsigned one, due to this reason, {in,de}crease_sleep_time
may return overflowed result.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: avoid unneeded sync on quota file
Chao Yu [Mon, 7 Aug 2017 08:37:59 +0000 (16:37 +0800)]
f2fs: avoid unneeded sync on quota file

We only need to sync quota file with appointed quota type instead of all
types in f2fs_quota_{on,off}.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: introduce gc_urgent mode for background GC
Jaegeuk Kim [Mon, 7 Aug 2017 05:09:00 +0000 (22:09 -0700)]
f2fs: introduce gc_urgent mode for background GC

This patch adds a sysfs entry to control urgent mode for background GC.
If this is set, background GC thread conducts GC with gc_urgent_sleep_time
all the time.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: use IPU for cold files
Jaegeuk Kim [Sat, 5 Aug 2017 21:25:08 +0000 (14:25 -0700)]
f2fs: use IPU for cold files

We expect cold files write data sequentially, but sometimes some of small data
can be updated, which incurs fragmentation.
Let's avoid that.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: fix the size value in __check_sit_bitmap
Yunlong Song [Fri, 4 Aug 2017 09:07:15 +0000 (17:07 +0800)]
f2fs: fix the size value in __check_sit_bitmap

The current size value is not correct and will miss bitmap check.

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: add app/fs io stat
Chao Yu [Wed, 2 Aug 2017 15:21:48 +0000 (23:21 +0800)]
f2fs: add app/fs io stat

This patch enables inner app/fs io stats and introduces below virtual fs
nodes for exposing stats info:
/sys/fs/f2fs/<dev>/iostat_enable
/proc/fs/f2fs/<dev>/iostat_info

Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: fix wrong stat assignment]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: do not change the valid_block value if cur_valid_map was wrongly set or cleared
Yunlong Song [Wed, 2 Aug 2017 14:16:54 +0000 (22:16 +0800)]
f2fs: do not change the valid_block value if cur_valid_map was wrongly set or cleared

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: update cur_valid_map_mir together with cur_valid_map
Yunlong Song [Wed, 2 Aug 2017 13:20:13 +0000 (21:20 +0800)]
f2fs: update cur_valid_map_mir together with cur_valid_map

When cur_valid_map passes the f2fs_test_and_set(,clear)_bit test,
cur_valid_map_mir update is skipped unlikely, so fix it. The fix
now changes the mirror check together with cur_valid_map all the
time.

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: Fix unused variable and add unlikely for corner condition.]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: use printk_ratelimited for f2fs_msg
Jaegeuk Kim [Thu, 3 Aug 2017 03:58:29 +0000 (20:58 -0700)]
f2fs: use printk_ratelimited for f2fs_msg

This patch reduces contention of printks.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: expose features to sysfs entry
Jaegeuk Kim [Sat, 22 Jul 2017 00:14:09 +0000 (17:14 -0700)]
f2fs: expose features to sysfs entry

This patch exposes what features are supported by current f2fs build to sysfs
entry via:

/sys/fs/f2fs/features/
/sys/fs/f2fs/dev/features

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: support inode checksum
Chao Yu [Mon, 31 Jul 2017 12:19:09 +0000 (20:19 +0800)]
f2fs: support inode checksum

This patch adds to support inode checksum in f2fs.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: fix verification flow]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: return wrong error number on f2fs_quota_write
Jaegeuk Kim [Sun, 30 Jul 2017 16:45:14 +0000 (09:45 -0700)]
f2fs: return wrong error number on f2fs_quota_write

This must return size, not error number.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: provide f2fs_balance_fs to __write_node_page
Yunlong Song [Thu, 27 Jul 2017 12:11:00 +0000 (20:11 +0800)]
f2fs: provide f2fs_balance_fs to __write_node_page

Let node writeback also do f2fs_balance_fs to ensure there are always enough free
segments.

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: introduce f2fs_statfs_project
Chao Yu [Fri, 28 Jul 2017 16:32:53 +0000 (00:32 +0800)]
f2fs: introduce f2fs_statfs_project

This patch introduces f2fs_statfs_project, it enables to show usage
status of directory tree which is limited with project quota.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: don't need to wait for node writes for atomic write
Jaegeuk Kim [Fri, 28 Jul 2017 09:29:12 +0000 (02:29 -0700)]
f2fs: don't need to wait for node writes for atomic write

We have a node chain to serialize node block writes, so if any IOs for
node block writes are reordered, we'll get broken node chain. IOWs,
roll-forward recovery will see all or none node blocks given fsync
mark.

E.g.,
Node chain consists of:
 N1 -> N2 -> N3 -> NFSYNC -> N1' -> N2' -> N'FSYNC

Reordered to:
1) N1 -> N2 -> N3 -> N2' -> NFSYNC -> N'FSYNC -> power-cut
2) N1 -> N2 -> N3 -> N1' -> NFSYNC -> power-cut
3) N1 -> N2 -> NFSYNC -> N1' -> N'FSYNC -> N3 -> power-cut
4) N1 -> NFSYNC -> N1' -> N2' -> N'FSYNC -> N3 -> power-cut

Roll-forward recovery can proceed to:
1) N1 -> N2 -> N3 -> NFSYNC -> X
2) N1 -> N2 -> N3 -> NFSYNC -> N1' -> X
3) N1 -> N2 -> N3 -> FSYNC -> N1' -> X
4) N1 -> X

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: avoid naming confusion of sysfs init
Jaegeuk Kim [Wed, 26 Jul 2017 18:24:13 +0000 (11:24 -0700)]
f2fs: avoid naming confusion of sysfs init

This patch changes the function names of sysfs init to follow ext4.

f2fs_init_sysfs <-> f2fs_register_sysfs
f2fs_exit_sysfs <-> f2fs_unregister_sysfs

Suggested-by: Chao Yu <yuchao0@huawei.com>
Reivewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: support project quota
Chao Yu [Tue, 25 Jul 2017 16:01:41 +0000 (00:01 +0800)]
f2fs: support project quota

This patch adds to support plain project quota.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: record quota during dot{,dot} recovery
Chao Yu [Mon, 24 Jul 2017 09:12:06 +0000 (17:12 +0800)]
f2fs: record quota during dot{,dot} recovery

In ->lookup(), we will have a try to recover dot or dotdot for
corrupted directory, once disk quota is on, if it allocates new
block during dotdot recovery, we need to record disk quota info
for the allocation, so this patch fixes this issue by adding
missing dquot_initialize() in __recover_dot_dentries.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: enhance on-disk inode structure scalability
Chao Yu [Tue, 18 Jul 2017 16:19:06 +0000 (00:19 +0800)]
f2fs: enhance on-disk inode structure scalability

This patch add new flag F2FS_EXTRA_ATTR storing in inode.i_inline
to indicate that on-disk structure of current inode is extended.

In order to extend, we changed the inode structure a bit:

Original one:

struct f2fs_inode {
...
struct f2fs_extent i_ext;
__le32 i_addr[DEF_ADDRS_PER_INODE];
__le32 i_nid[DEF_NIDS_PER_INODE];
}

Extended one:

struct f2fs_inode {
        ...
        struct f2fs_extent i_ext;
union {
struct {
__le16 i_extra_isize;
__le16 i_padding;
__le32 i_extra_end[0];
};
__le32 i_addr[DEF_ADDRS_PER_INODE];
};
        __le32 i_nid[DEF_NIDS_PER_INODE];
}

Once F2FS_EXTRA_ATTR is set, we will steal four bytes in the head of
i_addr field for storing i_extra_isize and i_padding. with i_extra_isize,
we can calculate actual size of reserved space in i_addr, available
attribute fields included in total extra attribute fields for current
inode can be described as below:

  +--------------------+
  | .i_mode            |
  | ...                |
  | .i_ext             |
  +--------------------+
  | .i_extra_isize     |-----+
  | .i_padding         |     |
  | .i_prjid           |     |
  | .i_atime_extra     |     |
  | .i_ctime_extra     |     |
  | .i_mtime_extra     |<----+
  | .i_inode_cs        |<----- store blkaddr/inline from here
  | .i_xattr_cs        |
  | ...                |
  +--------------------+
  |                    |
  |    block address   |
  |                    |
  +--------------------+
  | .i_nid             |
  +--------------------+
  |   node_footer      |
  | (nid, ino, offset) |
  +--------------------+

Hence, with this patch, we would enhance scalability of f2fs inode for
storing more newly added attribute.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: make max inline size changeable
Chao Yu [Tue, 18 Jul 2017 16:19:05 +0000 (00:19 +0800)]
f2fs: make max inline size changeable

This patch tries to make below macros calculating max inline size,
inline dentry field size considerring reserving size-changeable
space:
- MAX_INLINE_DATA
- NR_INLINE_DENTRY
- INLINE_DENTRY_BITMAP_SIZE
- INLINE_RESERVED_SIZE

Then, when inline_{data,dentry} options is enabled, it allows us to
reserve inline space with different size flexibly for adding newly
introduced inode attribute.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: add ioctl to expose current features
Jaegeuk Kim [Fri, 21 Jul 2017 19:58:59 +0000 (12:58 -0700)]
f2fs: add ioctl to expose current features

This patch adds an ioctl to provide feature information to user.
For exapmle, SQLite can use this ioctl to detect whether f2fs support atomic
write or not.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: make background threads of f2fs being aware of freezing
Chao Yu [Sat, 22 Jul 2017 00:52:23 +0000 (08:52 +0800)]
f2fs: make background threads of f2fs being aware of freezing

When ->freeze_fs is called from lvm for doing snapshot, it needs to
make sure there will be no more changes in filesystem's data, however,
previously, background threads like GC thread wasn't aware of freezing,
so in environment with active background threads, data of snapshot
becomes unstable.

This patch fixes this issue by adding sb_{start,end}_intwrite in
below background threads:
- GC thread
- flush thread
- discard thread

Note that, don't use sb_start_intwrite() in gc_thread_func() due to:

generic/241 reports below bug:

 ======================================================
 WARNING: possible circular locking dependency detected
 4.13.0-rc1+ #32 Tainted: G           O
 ------------------------------------------------------
 f2fs_gc-250:0/22186 is trying to acquire lock:
  (&sbi->gc_mutex){+.+...}, at: [<f8fa7f0b>] f2fs_sync_fs+0x7b/0x1b0 [f2fs]

 but task is already holding lock:
  (sb_internal#2){++++.-}, at: [<f8fb5609>] gc_thread_func+0x159/0x4a0 [f2fs]

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #2 (sb_internal#2){++++.-}:
        __lock_acquire+0x405/0x7b0
        lock_acquire+0xae/0x220
        __sb_start_write+0x11d/0x1f0
        f2fs_evict_inode+0x2d6/0x4e0 [f2fs]
        evict+0xa8/0x170
        iput+0x1fb/0x2c0
        f2fs_sync_inode_meta+0x3f/0xf0 [f2fs]
        write_checkpoint+0x1b1/0x750 [f2fs]
        f2fs_sync_fs+0x85/0x1b0 [f2fs]
        f2fs_do_sync_file.isra.24+0x137/0xa30 [f2fs]
        f2fs_sync_file+0x34/0x40 [f2fs]
        vfs_fsync_range+0x4a/0xa0
        do_fsync+0x3c/0x60
        SyS_fdatasync+0x15/0x20
        do_fast_syscall_32+0xa1/0x1b0
        entry_SYSENTER_32+0x4c/0x7b

 -> #1 (&sbi->cp_mutex){+.+...}:
        __lock_acquire+0x405/0x7b0
        lock_acquire+0xae/0x220
        __mutex_lock+0x4f/0x830
        mutex_lock_nested+0x25/0x30
        write_checkpoint+0x2f/0x750 [f2fs]
        f2fs_sync_fs+0x85/0x1b0 [f2fs]
        sync_filesystem+0x67/0x80
        generic_shutdown_super+0x27/0x100
        kill_block_super+0x22/0x50
        kill_f2fs_super+0x3a/0x40 [f2fs]
        deactivate_locked_super+0x3d/0x70
        deactivate_super+0x40/0x60
        cleanup_mnt+0x39/0x70
        __cleanup_mnt+0x10/0x20
        task_work_run+0x69/0x80
        exit_to_usermode_loop+0x57/0x92
        do_fast_syscall_32+0x18c/0x1b0
        entry_SYSENTER_32+0x4c/0x7b

 -> #0 (&sbi->gc_mutex){+.+...}:
        validate_chain.isra.36+0xc50/0xdb0
        __lock_acquire+0x405/0x7b0
        lock_acquire+0xae/0x220
        __mutex_lock+0x4f/0x830
        mutex_lock_nested+0x25/0x30
        f2fs_sync_fs+0x7b/0x1b0 [f2fs]
        f2fs_balance_fs_bg+0xb9/0x200 [f2fs]
        gc_thread_func+0x302/0x4a0 [f2fs]
        kthread+0xe9/0x120
        ret_from_fork+0x19/0x24

 other info that might help us debug this:

 Chain exists of:
   &sbi->gc_mutex --> &sbi->cp_mutex --> sb_internal#2

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(sb_internal#2);
                                lock(&sbi->cp_mutex);
                                lock(sb_internal#2);
   lock(&sbi->gc_mutex);

  *** DEADLOCK ***

 1 lock held by f2fs_gc-250:0/22186:
  #0:  (sb_internal#2){++++.-}, at: [<f8fb5609>] gc_thread_func+0x159/0x4a0 [f2fs]

 stack backtrace:
 CPU: 2 PID: 22186 Comm: f2fs_gc-250:0 Tainted: G           O    4.13.0-rc1+ #32
 Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
 Call Trace:
  dump_stack+0x5f/0x92
  print_circular_bug+0x1b3/0x1bd
  validate_chain.isra.36+0xc50/0xdb0
  ? __this_cpu_preempt_check+0xf/0x20
  __lock_acquire+0x405/0x7b0
  lock_acquire+0xae/0x220
  ? f2fs_sync_fs+0x7b/0x1b0 [f2fs]
  __mutex_lock+0x4f/0x830
  ? f2fs_sync_fs+0x7b/0x1b0 [f2fs]
  mutex_lock_nested+0x25/0x30
  ? f2fs_sync_fs+0x7b/0x1b0 [f2fs]
  f2fs_sync_fs+0x7b/0x1b0 [f2fs]
  f2fs_balance_fs_bg+0xb9/0x200 [f2fs]
  gc_thread_func+0x302/0x4a0 [f2fs]
  ? preempt_schedule_common+0x2f/0x4d
  ? f2fs_gc+0x540/0x540 [f2fs]
  kthread+0xe9/0x120
  ? f2fs_gc+0x540/0x540 [f2fs]
  ? kthread_create_on_node+0x30/0x30
  ret_from_fork+0x19/0x24

The deadlock occurs in below condition:
GC Thread Thread B
- sb_start_intwrite
- f2fs_sync_file
 - f2fs_sync_fs
  - mutex_lock(&sbi->gc_mutex)
   - write_checkpoint
    - block_operations
     - f2fs_sync_inode_meta
      - iput
       - sb_start_intwrite
 - mutex_lock(&sbi->gc_mutex)

Fix this by altering sb_start_intwrite to sb_start_write_trylock.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: don't give partially written atomic data from process crash
Jaegeuk Kim [Tue, 25 Jul 2017 02:46:29 +0000 (19:46 -0700)]
f2fs: don't give partially written atomic data from process crash

This patch resolves the below scenario.

== Process 1 ==     == Process 2 ==
open(w)             open(rw)
begin
write(new_#1)
process_crash
  f_op->flush
  locks_remove_posix
  f_op>release
                    read (new_#1)

In order to avoid corrupted database caused by new_#1, we must do roll-back
at process_crash time. In order to check that, this patch keeps task which
triggers transaction begin, and does roll-back in f_op->flush before removing
file locks.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: give a try to do atomic write in -ENOMEM case
Jaegeuk Kim [Wed, 19 Jul 2017 17:59:55 +0000 (10:59 -0700)]
f2fs: give a try to do atomic write in -ENOMEM case

It'd be better to retry writing atomic pages when we get -ENOMEM.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: preserve i_mode if __f2fs_set_acl() fails
Ernesto A. Fernández [Mon, 24 Jul 2017 01:32:54 +0000 (22:32 -0300)]
f2fs: preserve i_mode if __f2fs_set_acl() fails

When changing a file's acl mask, __f2fs_set_acl() will first set the
group bits of i_mode to the value of the mask, and only then set the
actual extended attribute representing the new acl.

If the second part fails (due to lack of space, for example) and the
file had no acl attribute to begin with, the system will from now on
assume that the mask permission bits are actual group permission bits,
potentially granting access to the wrong users.

Prevent this by only changing the inode mode after the acl has been set.

Signed-off-by: Ernesto A. Fernández <ernesto.mnd.fernandez@gmail.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: alloc new nids for xattr block in recovery
Yunlei He [Tue, 18 Jul 2017 01:48:12 +0000 (09:48 +0800)]
f2fs: alloc new nids for xattr block in recovery

recovery file A: recovery file B:
-get_dnode_of_data
-alloc_nid
-recover_xattr_data
-set_node_addr(sbi, &ni, NEW_ADDR, false);
--->bug_on for nid has been used by file A

In recovery process, new allocated node blocks may "reuse" xattr block
nids, this patch alloc new nids for xattr blocks in recovery process to
avoid this problem.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: spread struct f2fs_dentry_ptr for inline path
Chao Yu [Sun, 16 Jul 2017 07:08:54 +0000 (15:08 +0800)]
f2fs: spread struct f2fs_dentry_ptr for inline path

Use f2fs_dentry_ptr structure to indicate inline dentry structure as
much as possible, so we can wrap inline dentry with size-fixed fields
to the one with size-changeable fields. With this change, we can
handle size-changeable inline dentry more easily.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
6 years agof2fs: remove unused input parameter
Yunlei He [Mon, 17 Jul 2017 11:16:11 +0000 (19:16 +0800)]
f2fs: remove unused input parameter

This patch remove unused input parameter in function
new_node_page.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Yong Sheng <shengyong1@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: avoid cpu lockup
Jaegeuk Kim [Fri, 14 Jul 2017 18:45:21 +0000 (11:45 -0700)]
f2fs: avoid cpu lockup

Before retrying to flush data or dentry pages, we need to release cpu in order
to prevent watchdog.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: include seq_file.h for sysfs.c
Jaegeuk Kim [Fri, 14 Jul 2017 00:45:21 +0000 (17:45 -0700)]
f2fs: include seq_file.h for sysfs.c

This patch includes seq_file.h to avoid compile error.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: Don't clear SGID when inheriting ACLs
Jaegeuk Kim [Tue, 11 Jul 2017 21:56:49 +0000 (14:56 -0700)]
f2fs: Don't clear SGID when inheriting ACLs

This patch copies commit b7f8a09f80:
"btrfs: Don't clear SGID when inheriting ACLs" written by Jan.

Fixes: 073931017b49d9458aa351605b43a7e34598caef
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: remove extra inode_unlock() in error path
Luis Henriques [Tue, 11 Jul 2017 16:30:33 +0000 (17:30 +0100)]
f2fs: remove extra inode_unlock() in error path

This commit removes an extra inode_unlock() that is being done in function
f2fs_ioc_setflags error path.  While there, get rid of a useless 'out'
label as well.

Fixes: 0abd675e97e6 ("f2fs: support plain user/group quota")
Signed-off-by: Luis Henriques <lhenriques@suse.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agofscrypt: add support for AES-128-CBC
Daniel Walter [Mon, 19 Jun 2017 07:27:58 +0000 (09:27 +0200)]
fscrypt: add support for AES-128-CBC

fscrypt provides facilities to use different encryption algorithms which
are selectable by userspace when setting the encryption policy. Currently,
only AES-256-XTS for file contents and AES-256-CBC-CTS for file names are
implemented. This is a clear case of kernel offers the mechanism and
userspace selects a policy. Similar to what dm-crypt and ecryptfs have.

This patch adds support for using AES-128-CBC for file contents and
AES-128-CBC-CTS for file name encryption. To mitigate watermarking
attacks, IVs are generated using the ESSIV algorithm. While AES-CBC is
actually slightly less secure than AES-XTS from a security point of view,
there is more widespread hardware support. Using AES-CBC gives us the
acceptable performance while still providing a moderate level of security
for persistent storage.

Especially low-powered embedded devices with crypto accelerators such as
CAAM or CESA often only support AES-CBC. Since using AES-CBC over AES-XTS
is basically thought of a last resort, we use AES-128-CBC over AES-256-CBC
since it has less encryption rounds and yields noticeable better
performance starting from a file size of just a few kB.

Signed-off-by: Daniel Walter <dwalter@sigma-star.at>
[david@sigma-star.at: addressed review comments]
Signed-off-by: David Gstir <david@sigma-star.at>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
7 years agofscrypt: inline fscrypt_free_filename()
Eric Biggers [Tue, 23 May 2017 01:14:06 +0000 (18:14 -0700)]
fscrypt: inline fscrypt_free_filename()

fscrypt_free_filename() only needs to do a kfree() of crypto_buf.name,
which works well as an inline function.  We can skip setting the various
pointers to NULL, since no user cares about it (the name is always freed
just before it goes out of scope).

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: David Gstir <david@sigma-star.at>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
7 years agof2fs: make more close to v4.13-rc1
Jaegeuk Kim [Tue, 11 Jul 2017 02:16:28 +0000 (19:16 -0700)]
f2fs: make more close to v4.13-rc1

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: support plain user/group quota
Chao Yu [Sat, 8 Jul 2017 16:13:07 +0000 (00:13 +0800)]
f2fs: support plain user/group quota

This patch adds to support plain user/group quota.

Change Note by Jaegeuk Kim.

- Use f2fs page cache for quota files in order to consider garbage collection.
  so, quota files are not tolerable for sudden power-cuts, so user needs to do
  quotacheck.

- setattr() calls dquot_transfer which will transfer inode->i_blocks.
  We can't reclaim that during f2fs_evict_inode(). So, we need to count
  node blocks as well in order to match i_blocks with dquot's space.

  Note that, Chao wrote a patch to count inode->i_blocks without inode block.
  (f2fs: don't count inode block in in-memory inode.i_blocks)

- in f2fs_remount, we need to make RW in prior to dquot_resume.

- handle fault_injection case during f2fs_quota_off_umount

- TODO: Project quota

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: avoid deadlock caused by lock order of page and lock_op
Jaegeuk Kim [Thu, 22 Jun 2017 00:52:39 +0000 (17:52 -0700)]
f2fs: avoid deadlock caused by lock order of page and lock_op

- punch_hole
 - fill_zero
  - f2fs_lock_op
  - get_new_data_page
   - lock_page

- f2fs_write_data_pages
 - lock_page
 - do_write_data_page
  - f2fs_lock_op

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: use spin_{,un}lock_irq{save,restore}
Chao Yu [Fri, 7 Jul 2017 06:10:15 +0000 (14:10 +0800)]
f2fs: use spin_{,un}lock_irq{save,restore}

generic/361 reports below warning, this is because: once, there is
someone entering into critical region of sbi.cp_lock, if write_end_io.
f2fs_stop_checkpoint is invoked from an triggered IRQ, we will encounter
deadlock.

So this patch changes to use spin_{,un}lock_irq{save,restore} to create
critical region without IRQ enabled to avoid potential deadlock.

 irq event stamp: 83391573
 loop: Write error at byte offset 438729728, length 1024.
 hardirqs last  enabled at (83391573): [<c1809752>] restore_all+0xf/0x65
 hardirqs last disabled at (83391572): [<c1809eac>] reschedule_interrupt+0x30/0x3c
 loop: Write error at byte offset 438860288, length 1536.
 softirqs last  enabled at (83389244): [<c180cc4e>] __do_softirq+0x1ae/0x476
 softirqs last disabled at (83389237): [<c101ca7c>] do_softirq_own_stack+0x2c/0x40
 loop: Write error at byte offset 438990848, length 2048.
 ================================
 WARNING: inconsistent lock state
 4.12.0-rc2+ #30 Tainted: G           O
 --------------------------------
 inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
 xfs_io/7959 [HC1[1]:SC0[0]:HE0:SE1] takes:
  (&(&sbi->cp_lock)->rlock){?.+...}, at: [<f96f96cc>] f2fs_stop_checkpoint+0x1c/0x50 [f2fs]
 {HARDIRQ-ON-W} state was registered at:
   __lock_acquire+0x527/0x7b0
   lock_acquire+0xae/0x220
   _raw_spin_lock+0x42/0x50
   do_checkpoint+0x165/0x9e0 [f2fs]
   write_checkpoint+0x33f/0x740 [f2fs]
   __f2fs_sync_fs+0x92/0x1f0 [f2fs]
   f2fs_sync_fs+0x12/0x20 [f2fs]
   sync_filesystem+0x67/0x80
   generic_shutdown_super+0x27/0x100
   kill_block_super+0x22/0x50
   kill_f2fs_super+0x3a/0x40 [f2fs]
   deactivate_locked_super+0x3d/0x70
   deactivate_super+0x40/0x60
   cleanup_mnt+0x39/0x70
   __cleanup_mnt+0x10/0x20
   task_work_run+0x69/0x80
   exit_to_usermode_loop+0x57/0x85
   do_fast_syscall_32+0x18c/0x1b0
   entry_SYSENTER_32+0x4c/0x7b
 irq event stamp: 1957420
 hardirqs last  enabled at (1957419): [<c1808f37>] _raw_spin_unlock_irq+0x27/0x50
 hardirqs last disabled at (1957420): [<c1809f9c>] call_function_single_interrupt+0x30/0x3c
 softirqs last  enabled at (1953784): [<c180cc4e>] __do_softirq+0x1ae/0x476
 softirqs last disabled at (1953773): [<c101ca7c>] do_softirq_own_stack+0x2c/0x40

 other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(&(&sbi->cp_lock)->rlock);
   <Interrupt>
     lock(&(&sbi->cp_lock)->rlock);

  *** DEADLOCK ***

 2 locks held by xfs_io/7959:
  #0:  (sb_writers#13){.+.+.+}, at: [<c11fd7ca>] vfs_write+0x16a/0x190
  #1:  (&sb->s_type->i_mutex_key#16){+.+.+.}, at: [<f96e33f5>] f2fs_file_write_iter+0x25/0x140 [f2fs]

 stack backtrace:
 CPU: 2 PID: 7959 Comm: xfs_io Tainted: G           O    4.12.0-rc2+ #30
 Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
 Call Trace:
  dump_stack+0x5f/0x92
  print_usage_bug+0x1d3/0x1dd
  ? check_usage_backwards+0xe0/0xe0
  mark_lock+0x23d/0x280
  __lock_acquire+0x699/0x7b0
  ? __this_cpu_preempt_check+0xf/0x20
  ? trace_hardirqs_off_caller+0x91/0xe0
  lock_acquire+0xae/0x220
  ? f2fs_stop_checkpoint+0x1c/0x50 [f2fs]
  _raw_spin_lock+0x42/0x50
  ? f2fs_stop_checkpoint+0x1c/0x50 [f2fs]
  f2fs_stop_checkpoint+0x1c/0x50 [f2fs]
  f2fs_write_end_io+0x147/0x150 [f2fs]
  bio_endio+0x7a/0x1e0
  blk_update_request+0xad/0x410
  blk_mq_end_request+0x16/0x60
  lo_complete_rq+0x3c/0x70
  __blk_mq_complete_request_remote+0x11/0x20
  flush_smp_call_function_queue+0x6d/0x120
  ? debug_smp_processor_id+0x12/0x20
  generic_smp_call_function_single_interrupt+0x12/0x30
  smp_call_function_single_interrupt+0x25/0x40
  call_function_single_interrupt+0x37/0x3c
 EIP: _raw_spin_unlock_irq+0x2d/0x50
 EFLAGS: 00000296 CPU: 2
 EAX: 00000001 EBX: d2ccc51c ECX: 00000001 EDX: c1aacebd
 ESI: 00000000 EDI: 00000000 EBP: c96c9d1c ESP: c96c9d18
  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
  ? inherit_task_group.isra.98.part.99+0x6b/0xb0
  __add_to_page_cache_locked+0x1d4/0x290
  add_to_page_cache_lru+0x38/0xb0
  pagecache_get_page+0x8e/0x200
  f2fs_write_begin+0x96/0xf00 [f2fs]
  ? trace_hardirqs_on_caller+0xdd/0x1c0
  ? current_time+0x17/0x50
  ? trace_hardirqs_on+0xb/0x10
  generic_perform_write+0xa9/0x170
  __generic_file_write_iter+0x1a2/0x1f0
  ? f2fs_preallocate_blocks+0x137/0x160 [f2fs]
  f2fs_file_write_iter+0x6e/0x140 [f2fs]
  ? __lock_acquire+0x429/0x7b0
  __vfs_write+0xc1/0x140
  vfs_write+0x9b/0x190
  SyS_pwrite64+0x63/0xa0
  do_fast_syscall_32+0xa1/0x1b0
  entry_SYSENTER_32+0x4c/0x7b
 EIP: 0xb7786c61
 EFLAGS: 00000293 CPU: 2
 EAX: ffffffda EBX: 00000003 ECX: 08416000 EDX: 00001000
 ESI: 18b24000 EDI: 00000000 EBP: 00000003 ESP: bf9b36b0
  DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 007b

Fixes: aaec2b1d1879 ("f2fs: introduce cp_lock to protect updating of ckpt_flags")
Cc: stable@vger.kernel.org
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: relax migratepage for atomic written page
Jaegeuk Kim [Thu, 6 Jul 2017 21:46:01 +0000 (14:46 -0700)]
f2fs: relax migratepage for atomic written page

In order to avoid lock contention for atomic written pages, we'd better give
EBUSY in f2fs_migrate_page when mode is asynchronous. We expect it will be
released soon as transaction commits.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: don't count inode block in in-memory inode.i_blocks
Chao Yu [Wed, 5 Jul 2017 17:11:31 +0000 (01:11 +0800)]
f2fs: don't count inode block in in-memory inode.i_blocks

Previously, we count all inode consumed blocks including inode block,
xattr block, index block, data block into i_blocks, for other generic
filesystems, they won't count inode block into i_blocks, so for
userspace applications or quota system, they may detect incorrect block
count according to i_blocks value in inode.

This patch changes to count all blocks into inode.i_blocks excluding
inode block, for on-disk i_blocks, we keep counting inode block for
backward compatibility.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agoRevert "f2fs: fix to clean previous mount option when remount_fs"
Chao Yu [Wed, 5 Jul 2017 04:17:24 +0000 (12:17 +0800)]
Revert "f2fs: fix to clean previous mount option when remount_fs"

Don't clear old mount option before parse new option during ->remount_fs
like other generic filesystems.

This reverts commit 26666c8a4366debae30ae37d0688b2bec92d196a.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: do not set LOST_PINO for renamed dir
Sheng Yong [Mon, 26 Jun 2017 02:41:36 +0000 (10:41 +0800)]
f2fs: do not set LOST_PINO for renamed dir

After renaming a directory, fsck could detect unmatched pino. The scenario
can be reproduced as the following:

$ mkdir /bar/subbar /foo
$ rename /bar/subbar /foo

Then fsck will report:
[ASSERT] (__chk_dots_dentries:1182)  --> Bad inode number[0x3] for '..', parent parent ino is [0x4]

Rename sets LOST_PINO for old_inode. However, the flag cannot be cleared,
since dir is written back with CP. So, let's get rid of LOST_PINO for a
renamed dir and fix the pino directly at the end of rename.

Signed-off-by: Sheng Yong <shengyong1@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: do not set LOST_PINO for newly created dir
Sheng Yong [Mon, 26 Jun 2017 02:41:35 +0000 (10:41 +0800)]
f2fs: do not set LOST_PINO for newly created dir

Since directories will be written back with checkpoint and fsync a
directory will always write CP, there is no need to set LOST_PINO
after creating a directory.

Signed-off-by: Sheng Yong <shengyong1@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: skip ->writepages for {mete,node}_inode during recovery
Chao Yu [Thu, 29 Jun 2017 15:20:45 +0000 (23:20 +0800)]
f2fs: skip ->writepages for {mete,node}_inode during recovery

Skip ->writepages in prior to ->writepage for {meta,node}_inode during
recovery, hence unneeded loop in ->writepages can be avoided.

Moreover, check SBI_POR_DOING earlier while writebacking pages.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: introduce __check_sit_bitmap
Chao Yu [Fri, 30 Jun 2017 09:19:02 +0000 (17:19 +0800)]
f2fs: introduce __check_sit_bitmap

After we introduce discard thread, discard command can be issued
concurrently with data allocating, this patch adds new function to
heck sit bitmap to ensure that userdata was invalid in which on-going
discard command covered.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
 Conflicts:
fs/f2fs/segment.c

7 years agof2fs: stop gc/discard thread in prior during umount
Chao Yu [Thu, 29 Jun 2017 15:17:45 +0000 (23:17 +0800)]
f2fs: stop gc/discard thread in prior during umount

This patch resolves kernel panic for xfstests/081, caused by recent f2fs_bug_on

 f2fs: add f2fs_bug_on in __remove_discard_cmd

For fixing, we will stop gc/discard thread in prior in ->kill_sb in order to
avoid referring and releasing race among them.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: introduce reserved_blocks in sysfs
Chao Yu [Mon, 26 Jun 2017 08:24:41 +0000 (16:24 +0800)]
f2fs: introduce reserved_blocks in sysfs

In this patch, we add a new sysfs interface, with it, we can control
number of reserved blocks in system which could not be used by user,
it enable f2fs to let user to configure for adjusting over-provision
ratio dynamically instead of changing it by mkfs.

So we can expect it will help to reserve more free space for relieving
GC in both filesystem and flash device.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
 Conflicts:
Documentation/ABI/testing/sysfs-fs-f2fs

7 years agof2fs: avoid redundant f2fs_flush after remount
Yunlong Song [Sat, 24 Jun 2017 07:57:19 +0000 (15:57 +0800)]
f2fs: avoid redundant f2fs_flush after remount

create_flush_cmd_control will create redundant issue_flush_thread after each
remount with flush_merge option.

Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: report # of free inodes more precisely
Jaegeuk Kim [Thu, 22 Jun 2017 03:55:55 +0000 (20:55 -0700)]
f2fs: report # of free inodes more precisely

If the partition is small, we don't need to report total # of inodes including
hidden free nodes.

Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: add ioctl to do gc with target block address
Jaegeuk Kim [Thu, 15 Jun 2017 23:44:42 +0000 (16:44 -0700)]
f2fs: add ioctl to do gc with target block address

This patch adds f2fs_ioc_gc_range() to move blocks located in the given
range.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: don't need to check encrypted inode for partial truncation
Jaegeuk Kim [Wed, 14 Jun 2017 15:05:32 +0000 (08:05 -0700)]
f2fs: don't need to check encrypted inode for partial truncation

The cache_only is always false, if inode is encrypted.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: measure inode.i_blocks as generic filesystem
Chao Yu [Wed, 14 Jun 2017 15:00:56 +0000 (23:00 +0800)]
f2fs: measure inode.i_blocks as generic filesystem

Both in memory or on disk, generic filesystems record i_blocks with
512bytes sized sector count, also VFS sub module such as disk quota
follows this rule, but f2fs records it with 4096bytes sized block
count, this difference leads to that once we use dquota's function
which inc/dec iblocks, it will make i_blocks of f2fs being inconsistent
between in memory and on disk.

In order to resolve this issue, this patch changes to make in-memory
i_blocks of f2fs recording sector count instead of block count,
meanwhile leaving on-disk i_blocks recording block count.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: set CP_TRIMMED_FLAG correctly
Chao Yu [Wed, 14 Jun 2017 15:00:55 +0000 (23:00 +0800)]
f2fs: set CP_TRIMMED_FLAG correctly

Don't set CP_TRIMMED_FLAG for non-zoned block device or discard
unsupported device, it can avoid to trigger unneeded checkpoint for
that kind of device.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: require key for truncate(2) of encrypted file
Eric Biggers [Tue, 13 Jun 2017 23:47:54 +0000 (16:47 -0700)]
f2fs: require key for truncate(2) of encrypted file

Currently, filesystems allow truncate(2) on an encrypted file without
the encryption key.  However, it's impossible to correctly handle the
case where the size being truncated to is not a multiple of the
filesystem block size, because that would require decrypting the final
block, zeroing the part beyond i_size, then encrypting the block.

As other modifications to encrypted file contents are prohibited without
the key, just prohibit truncate(2) as well, making it fail with ENOKEY.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: move sysfs code from super.c to fs/f2fs/sysfs.c
Chao Yu [Wed, 14 Jun 2017 09:39:47 +0000 (17:39 +0800)]
f2fs: move sysfs code from super.c to fs/f2fs/sysfs.c

Codes related to sysfs and procfs are dispersive and mixed with sb
related codes, but actually these codes are independent from others,
so split them from super.c, and reorgnize and manger them in sysfs.c.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
 Conflicts:
fs/f2fs/super.c

7 years agof2fs: clean up sysfs codes
Chao Yu [Wed, 14 Jun 2017 09:39:46 +0000 (17:39 +0800)]
f2fs: clean up sysfs codes

Just cleanup.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: fix wrong error number of fill_super
Chao Yu [Mon, 12 Jun 2017 01:44:27 +0000 (09:44 +0800)]
f2fs: fix wrong error number of fill_super

This patch fixes incorrect error number in error path of fill_super.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: fix to show injection rate in ->show_options
Chao Yu [Mon, 12 Jun 2017 01:44:24 +0000 (09:44 +0800)]
f2fs: fix to show injection rate in ->show_options

If fault injection functionality is enabled, show additional injection
rate in ->show_options.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: Fix a return value in case of error in 'f2fs_fill_super'
Christophe JAILLET [Sun, 11 Jun 2017 07:21:11 +0000 (09:21 +0200)]
f2fs: Fix a return value in case of error in 'f2fs_fill_super'

err must be set to -ENOMEM, otherwise we return 0.

Fixes: a912b54d3aaa0 ("f2fs: split bio cache")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: use proper variable name
Tiezhu Yang [Thu, 8 Jun 2017 22:32:54 +0000 (06:32 +0800)]
f2fs: use proper variable name

It is better to use variable name "inline_dentry" instead of "dentry_blk"
when data type is "struct f2fs_inline_dentry". This patch has no functional
changes, just to make code more readable especially when call the function
make_dentry_ptr_inline() and f2fs_convert_inline_dir().

Signed-off-by: Tiezhu Yang <kernelpatch@126.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: fix to avoid panic when encountering corrupt node
Chao Yu [Wed, 7 Jun 2017 03:17:35 +0000 (11:17 +0800)]
f2fs: fix to avoid panic when encountering corrupt node

With fault_injection option, generic/361 of fstests will complain us
with below message:

Call Trace:
 get_node_page+0x12/0x20 [f2fs]
 f2fs_iget+0x92/0x7d0 [f2fs]
 f2fs_fill_super+0x10fb/0x15e0 [f2fs]
 mount_bdev+0x184/0x1c0
 f2fs_mount+0x15/0x20 [f2fs]
 mount_fs+0x39/0x150
 vfs_kern_mount+0x67/0x110
 do_mount+0x1bb/0xc70
 SyS_mount+0x83/0xd0
 do_syscall_64+0x6e/0x160
 entry_SYSCALL64_slow_path+0x25/0x25

Since mkfs loop device in f2fs partition can be failed silently due to
checkpoint error injection, so root inode page can be corrupted, in order
to avoid needless panic, in get_node_page, it's better to leave message
and return error to caller, and let fsck repaire it later.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: don't track newly allocated nat entry in list
Chao Yu [Mon, 5 Jun 2017 10:29:08 +0000 (18:29 +0800)]
f2fs: don't track newly allocated nat entry in list

We will never persist newly allocated nat entries during checkpoint(), so
we don't need to track such nat entries in nat dirty list in order to
avoid:
- more latency during traversing dirty list;
- sorting nat sets incorrectly due to recording wrong entry_cnt in nat
entry set.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: add f2fs_bug_on in __remove_discard_cmd
Chao Yu [Mon, 5 Jun 2017 10:29:07 +0000 (18:29 +0800)]
f2fs: add f2fs_bug_on in __remove_discard_cmd

Recently, discard related codes have changed a lot, so add f2fs_bug_on to
detect potential bug.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: introduce __wait_one_discard_bio
Chao Yu [Mon, 5 Jun 2017 10:29:06 +0000 (18:29 +0800)]
f2fs: introduce __wait_one_discard_bio

In order to avoid copied codes.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: dax: fix races between page faults and truncating pages
Qiuyang Sun [Thu, 18 May 2017 03:06:45 +0000 (11:06 +0800)]
f2fs: dax: fix races between page faults and truncating pages

Currently in F2FS, page faults and operations that truncate the pagecahe
or data blocks, are completely unsynchronized. This can result in page
fault faulting in a page into a range that we are changing after
truncating, and thus we can end up with a page mapped to disk blocks that
will be shortly freed. Filesystem corruption will shortly follow.

This patch fixes the problem by creating new rw semaphore i_mmap_sem in
f2fs_inode_info and grab it for functions removing blocks from extent tree
and for read over page faults. The mechanism is similar to that in ext4.

Signed-off-by: Qiuyang Sun <sunqiuyang@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
 Conflicts:
fs/f2fs/file.c

7 years agof2fs: simplify the way of calulating next nat address
Fan Li [Fri, 2 Jun 2017 07:45:42 +0000 (15:45 +0800)]
f2fs: simplify the way of calulating next nat address

The index of segment which the next nat block is in has only one different
bit than the current one, so to get the next nat address, we can simply
alter that one bit.

Signed-off-by: Fan Li <fanofcode.li@samsung.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: sanity check size of nat and sit cache
Jin Qian [Thu, 1 Jun 2017 18:18:30 +0000 (11:18 -0700)]
f2fs: sanity check size of nat and sit cache

Make sure number of entires doesn't exceed max journal size.

Cc: stable@vger.kernel.org
Signed-off-by: Jin Qian <jinqian@android.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: fix a panic caused by NULL flush_cmd_control
Yunlei He [Thu, 1 Jun 2017 08:43:51 +0000 (16:43 +0800)]
f2fs: fix a panic caused by NULL flush_cmd_control

Mount fs with option noflush_merge, boot failed for illegal address
fcc in function f2fs_issue_flush:

        if (!test_opt(sbi, FLUSH_MERGE)) {
                ret = submit_flush_wait(sbi);
                atomic_inc(&fcc->issued_flush);   ->  Here, fcc illegal
                return ret;
        }

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: remove the unnecessary cast for PTR_ERR
Zhang Shengju [Thu, 1 Jun 2017 08:50:10 +0000 (16:50 +0800)]
f2fs: remove the unnecessary cast for PTR_ERR

It's not necessary to specify 'int' casting for PTR_ERR.

Signed-off-by: Zhang Shengju <zhangshengju@cmss.chinamobile.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: remove false-positive bug_on
Jaegeuk Kim [Thu, 1 Jun 2017 22:39:27 +0000 (15:39 -0700)]
f2fs: remove false-positive bug_on

For example,

f2fs_create
 - new_node_page is failed
 - handle_failed_inode
  - skip to add it into orphan list, since ni.blk_addr == NULL_ADDR
   : set_inode_flag(inode, FI_FREE_NID)

f2fs_evict_inode
 - EIO due to fault injection
 - f2fs_bug_on() is triggered

So, we don't need to call f2fs_bug_on in this case.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: Do not issue small discards in LFS mode
Damien Le Moal [Fri, 26 May 2017 08:04:40 +0000 (17:04 +0900)]
f2fs: Do not issue small discards in LFS mode

clear_prefree_segments() issues small discards after discarding full
segments. These small discards may not be section aligned, so not zone
aligned on a zoned block device, causing __f2fs_iissue_discard_zone() to fail.
Fix this by not issuing small discards for a volume mounted with the BLKZONED
feature enabled.

Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: don't bother checking for encryption key in ->write_iter()
Eric Biggers [Tue, 23 May 2017 00:39:45 +0000 (17:39 -0700)]
f2fs: don't bother checking for encryption key in ->write_iter()

Since only an open file can be written to, and we only allow open()ing
an encrypted file when its key is available, there is no need to check
for the key again before permitting each ->write_iter().

This code was also broken in that it wouldn't actually have failed if
the key was in fact unavailable.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: David Gstir <david@sigma-star.at>
Acked-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: don't bother checking for encryption key in ->mmap()
Eric Biggers [Tue, 23 May 2017 00:39:43 +0000 (17:39 -0700)]
f2fs: don't bother checking for encryption key in ->mmap()

Since only an open file can be mmap'ed, and we only allow open()ing an
encrypted file when its key is available, there is no need to check for
the key again before permitting each mmap().

This f2fs copy of this code was also broken in that it wouldn't actually
have failed if the key was in fact unavailable.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: David Gstir <david@sigma-star.at>
Acked-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: wait discard IO completion without cmd_lock held
Chao Yu [Fri, 19 May 2017 15:46:45 +0000 (23:46 +0800)]
f2fs: wait discard IO completion without cmd_lock held

Wait discard IO completion outside cmd_lock to avoid long latency
of holding cmd_lock in IO busy scenario.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: wake up all waiters in f2fs_submit_discard_endio
Chao Yu [Fri, 19 May 2017 15:46:44 +0000 (23:46 +0800)]
f2fs: wake up all waiters in f2fs_submit_discard_endio

There could be more than one waiter waiting discard IO completion, so we
need use complete_all() instead of complete() in f2fs_submit_discard_endio
to avoid hungtask.

Fixes:  ec9895add2c5 ("f2fs: don't hold cmd_lock during waiting discard
command")
Cc: <stable@vger.kernel.org>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: show more info if fail to issue discard
Chao Yu [Fri, 19 May 2017 15:46:43 +0000 (23:46 +0800)]
f2fs: show more info if fail to issue discard

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: introduce io_list for serialize data/node IOs
Chao Yu [Fri, 19 May 2017 15:37:01 +0000 (23:37 +0800)]
f2fs: introduce io_list for serialize data/node IOs

Serialize data/node IOs by using fifo list instead of mutex lock,
it will help to enhance concurrency of f2fs, meanwhile keeping LFS
IO semantics.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: split wio_mutex
Chao Yu [Fri, 19 May 2017 15:37:00 +0000 (23:37 +0800)]
f2fs: split wio_mutex

Split wio_mutex to adjust different temperature bio cache.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: combine huge num of discard rb tree consistence checks
Yunlei He [Fri, 19 May 2017 06:42:12 +0000 (14:42 +0800)]
f2fs: combine huge num of discard rb tree consistence checks

Came across a hungtask caused by huge number of rb tree traversing
during adding discard addrs in cp. This patch combine these consistence
checks and move it to discard thread.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: fix a bug caused by NULL extent tree
Yunlei He [Fri, 19 May 2017 07:06:12 +0000 (15:06 +0800)]
f2fs: fix a bug caused by NULL extent tree

Thread A: Thread B:

-f2fs_remount
    -sbi->mount_opt.opt = 0;
<--- -f2fs_iget
         -do_read_inode
     -f2fs_init_extent_tree
         -F2FS_I(inode)->extent_tree is NULL
        -default_options && parse_options
    -remount return
<---  -f2fs_map_blocks
          -f2fs_lookup_extent_tree
                                                              -f2fs_bug_on(sbi, !et);

The same problem with f2fs_new_inode.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: try to freeze in gc and discard threads
Jaegeuk Kim [Wed, 17 May 2017 17:36:58 +0000 (10:36 -0700)]
f2fs: try to freeze in gc and discard threads

This allows to freeze gc and discard threads.

Cc: stable@vger.kernel.org
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: add a new function get_ssr_cost
Yunlei He [Wed, 17 May 2017 09:22:51 +0000 (17:22 +0800)]
f2fs: add a new function get_ssr_cost

This patch add a new method get_ssr_cost to select
SSR segment more accurately.

Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: declare load_free_nid_bitmap static
Hou Pengyang [Wed, 17 May 2017 02:48:48 +0000 (02:48 +0000)]
f2fs: declare load_free_nid_bitmap static

Signed-off-by: Hou Pengyang <houpengyang@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: avoid f2fs_lock_op for IPU writes
Jaegeuk Kim [Fri, 12 May 2017 20:51:34 +0000 (13:51 -0700)]
f2fs: avoid f2fs_lock_op for IPU writes

Currently, if we do get_node_of_data before f2fs_lock_op, there may be dead lock
as follows, where process A would be in infinite loop, and B will NOT be awaked.

Process A(cp):            Process B:
f2fs_lock_all(sbi)
                        get_dnode_of_data <---- lock dn.node_page
flush_nodes             f2fs_lock_op

So, this patch adds f2fs_trylock_op to avoid f2fs_lock_op done by IPU.

Signed-off-by: Hou Pengyang <houpengyang@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: split bio cache
Jaegeuk Kim [Wed, 10 May 2017 18:18:25 +0000 (11:18 -0700)]
f2fs: split bio cache

Split DATA/NODE type bio cache according to different temperature,
so write IOs with the same temperature can be merged in corresponding
bio cache as much as possible, otherwise, different temperature write
IOs submitting into one bio cache will always cause split of bio.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
 Conflicts:
include/trace/events/f2fs.h

7 years agof2fs: use fio instead of multiple parameters
Jaegeuk Kim [Wed, 10 May 2017 21:19:54 +0000 (14:19 -0700)]
f2fs: use fio instead of multiple parameters

This patch just changes using fio instead of parameters.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: remove unnecessary read cases in merged IO flow
Jaegeuk Kim [Wed, 10 May 2017 18:28:38 +0000 (11:28 -0700)]
f2fs: remove unnecessary read cases in merged IO flow

Merged IO flow doesn't need to care about read IOs.

f2fs_submit_merged_bio -> f2fs_submit_merged_write
f2fs_submit_merged_bios -> f2fs_submit_merged_writes
f2fs_submit_merged_bio_cond -> f2fs_submit_merged_write_cond

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: use f2fs_submit_page_bio for ra_meta_pages
Jaegeuk Kim [Wed, 10 May 2017 18:23:36 +0000 (11:23 -0700)]
f2fs: use f2fs_submit_page_bio for ra_meta_pages

This patch avoids to use f2fs_submit_merged_bio for read, which was the only
read case.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
7 years agof2fs: make sure f2fs_gc returns consistent errno
Weichao Guo [Wed, 10 May 2017 20:28:00 +0000 (04:28 +0800)]
f2fs: make sure f2fs_gc returns consistent errno

By default, f2fs_gc returns -EINVAL in general error cases, e.g., no victim
was selected. However, the default errno may be overwritten in two cases:
gc_more and BG_GC -> FG_GC. We should return consistent errno in such cases.

Signed-off-by: Weichao Guo <guoweichao@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>