For below two cases, we can't guarantee data consistence:

a)
1. xfs_io "pwrite 0 4195328" "fsync"
2. xfs_io "pwrite 4195328 1024" "fdatasync"
3. godown
4. umount & mount
--> isize we updated before fdatasync won't be recovered

b)
1. xfs_io "pwrite -S 0xcc 0 4202496" "fsync"
2. xfs_io "fpunch 4194304 4096" "fdatasync"
3. godown
4. umount & mount
--> dnode we punched before fdatasync won't be recovered

The reason is that normally fdatasync won't be aware of modification
of metadata in file, e.g. isize changing, dnode updating, so in ->fsync
we will skip flushing node pages for above cases, result in making
fdatasynced file being lost during recovery.

Currently we have introduced DIRTY_META global list in sbi for tracking
dirty inode selectively, so in fdatasync we can choose to flush nodes
depend on dirty state of current inode in the list.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Thread A		Thread B		Thread C
- f2fs_create
 - f2fs_new_inode
  - f2fs_lock_op
   - alloc_nid
    alloc last nid
  - f2fs_unlock_op
			- f2fs_create
			 - f2fs_new_inode
			  - f2fs_lock_op
			   - alloc_nid
			    as node count still not
			    be increased, we will
			    loop in alloc_nid
						- f2fs_write_node_pages
						 - f2fs_balance_fs_bg
						  - f2fs_sync_fs
						   - write_checkpoint
						    - block_operations
						     - f2fs_lock_all
 - f2fs_lock_op

While creating new inode, we do not allocate and account nid atomically,
so that when there is almost no free nids left, we may encounter deadloop
like above stack.

In order to avoid that, reuse nm_i::available_nids for accounting free nids
and make nid allocation and counting being atomical during node creation.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Thread A			Thread B

- write_checkpoint
 - block_operations
   -blk_start_plug
    -sync_node_pages		- f2fs_do_sync_file
				 - fsync_node_pages
				  - f2fs_wait_on_page_writeback

Thread A wait for global F2FS_DIRTY_NODES decreased to zero,
it start a plug list, some requests have been added to this list.
Thread B lock one dirty node page, and wait this page write back.
But this page has been in plug list of thread A with PG_writeback flag.
Thread A keep on running and its plug list has no chance to finish,
so it seems a deadlock between cp and fsync path.

This patch add a wait on page write back before set node page dirty
to avoid this problem.
Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Pengyang Hou <houpengyang@huawei.com>
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Normally, while committing checkpoint, we will wait on all pages to be
writebacked no matter the page is data or metadata, so in scenario where
there are lots of data IO being submitted with metadata, we may suffer
long latency for waiting writeback during checkpoint.

Indeed, we only care about persistence for pages with metadata, but not
pages with data, as file system consistent are only related to metadate,
so in order to avoid encountering long latency in above scenario, let's
recognize and reference metadata in submitted IOs, wait writeback only
for metadatas.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Previously, written_valid_blocks was got by ckpt->valid_block_count. But if
the last checkpoint has some NEW_ADDR due to power-cut, we can get wrong value.
Fix it to get the number from actual written block count from sit entries.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

If many threads hit has_not_enough_free_secs() in f2fs_balance_fs() at the same
time, all the threads would do FG_GC or BG_GC.
In this critical path, we totally don't need to do BG_GC at all.
Let's avoid that.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

In direct_IO path of f2fs_file_write_iter(),
1. f2fs_preallocate_blocks(F2FS_GET_BLOCK_PRE_DIO)
   -> allocate LBA X
2. f2fs_direct_IO()
   -> return 0;

Then,
f2fs_write_data_page() will allocate another LBA X+1.

This makes EIO triggered by HM-SMR.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch has no functional change.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch implements multiple devices support for f2fs.
Given multiple devices by mkfs.f2fs, f2fs shows them entirely as one big
volume under one f2fs instance.

Internal block management is very simple, but we will modify block allocation
and background GC policy to boost IO speed by exploiting them accoording to
each device speed.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

We can allow dio reads for LFS mode, while doing buffered writes for dio writes.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Now we don't need to be too much careful about storage alignment for dio, since
its speed becomes quite fast and we'd better avoid any misalignment first.

Revert: 38aa0889 (f2fs: align direct_io'ed data to section)
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

If one block has been to written to a new place, just return
in move data process. This patch check it again with holding
page lock.
Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

i_times of inode will be set with current system time which can be
configured through 'date', so it's not safe to judge dnode block as
garbage data or unchanged inode depend on i_times.

Now, we have used enhanced 'cp_ver + cp' crc method to verify valid
dnode block, so I expect recoverying invalid dnode is almost not
possible.

This reverts commit 807b1e1c.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

The generic freeze_super() calls sync_filesystems() before f2fs_freeze().
So, basically we don't need to do checkpoint in f2fs_freeze(). But, in xfs/068,
it triggers circular locking problem below due to gc_mutex for checkpoint.

======================================================
[ INFO: possible circular locking dependency detected ]
4.9.0-rc1+ #132 Tainted: G           OE
-------------------------------------------------------

1. wait for __sb_start_write() by

 [<ffffffff9845f353>] dump_stack+0x85/0xc2
 [<ffffffff980e80bf>] print_circular_bug+0x1cf/0x230
 [<ffffffff980eb4d0>] __lock_acquire+0x19e0/0x1bc0
 [<ffffffff980ebdcb>] lock_acquire+0x11b/0x220
 [<ffffffffc08c7c3b>] ? f2fs_drop_inode+0x9b/0x160 [f2fs]
 [<ffffffff9826bdd0>] __sb_start_write+0x130/0x200
 [<ffffffffc08c7c3b>] ? f2fs_drop_inode+0x9b/0x160 [f2fs]
 [<ffffffffc08c7c3b>] f2fs_drop_inode+0x9b/0x160 [f2fs]
 [<ffffffff98289991>] iput+0x171/0x2c0
 [<ffffffffc08cfccf>] f2fs_sync_inode_meta+0x3f/0xf0 [f2fs]
 [<ffffffffc08cfe04>] block_operations+0x84/0x110 [f2fs]
 [<ffffffffc08cff78>] write_checkpoint+0xe8/0xf20 [f2fs]
 [<ffffffff980e979d>] ? trace_hardirqs_on+0xd/0x10
 [<ffffffffc08c6de9>] ? f2fs_sync_fs+0x79/0x190 [f2fs]
 [<ffffffff9803e9d9>] ? sched_clock+0x9/0x10
 [<ffffffffc08c6de9>] ? f2fs_sync_fs+0x79/0x190 [f2fs]
 [<ffffffffc08c6df5>] f2fs_sync_fs+0x85/0x190 [f2fs]
 [<ffffffff982a4f90>] ? do_fsync+0x70/0x70
 [<ffffffff982a4f90>] ? do_fsync+0x70/0x70
 [<ffffffff982a4fb0>] sync_fs_one_sb+0x20/0x30
 [<ffffffff9826ca3e>] iterate_supers+0xae/0x100
 [<ffffffff982a50b5>] sys_sync+0x55/0x90
 [<ffffffff9890b345>] entry_SYSCALL_64_fastpath+0x23/0xc6

2. wait for sbi->gc_mutex by

 [<ffffffff980ebdcb>] lock_acquire+0x11b/0x220
 [<ffffffff989063d6>] mutex_lock_nested+0x76/0x3f0
 [<ffffffffc08c6de9>] f2fs_sync_fs+0x79/0x190 [f2fs]
 [<ffffffffc08c7a6c>] f2fs_freeze+0x1c/0x20 [f2fs]
 [<ffffffff9826b6ef>] freeze_super+0xcf/0x190
 [<ffffffff9827eebc>] do_vfs_ioctl+0x53c/0x6a0
 [<ffffffff9827f099>] SyS_ioctl+0x79/0x90
 [<ffffffff9890b345>] entry_SYSCALL_64_fastpath+0x23/0xc6
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Previously, we assigned CURSEG_WARM_DATA for direct_io, but if we have two or
four logs, we do not use that type at all.
Let's fix it.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Shouldn't update in-memory i_atime with on-disk i_mtime of inode when
recovering inode.

Shuoran found this bug which is hidden for a long time, honour is belong
to him.
Signed-off-by: Shuoran Liu <liushuoran@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

We should record updating status of inode only for living inode, for those
unlinked inode it needs to clear its ino cache, otherwise after the ino
was been reused, it will cause unneeded node page writing during ->fsync.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Similarly to the regular discard, trace zone reset events.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

When a zoned block device is mounted, discarding sections
contained in sequential zones must reset the zone write pointer.
For sections contained in conventional zones, the regular discard
is used if the drive supports it.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

With the zoned block device feature enabled, section discard
need to do a zone reset for sections contained in sequential
zones, and a regular discard (if supported) for sections
stored in conventional zones. Avoid the need for a costly
report zones to obtain a section zone type when discarding it
by caching the types of the device zones in the super block
information. This cache is initialized at mount time for mounts
with the zoned block device feature enabled.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

The LFS mode is mandatory for host-managed zoned block devices as
update in place optimizations are not possible for segments in
sequential zones.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Zone write pointer reset acts as discard for zoned block
devices. So if the zoned block device feature is enabled,
always declare that discard is enabled, even if the device
does not actually support the command.
For the same reason, prevent the use the "nodicard" mount
option.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

For zoned block devices, discard is replaced by zone reset. So
do not warn if the device does not supports discard.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

The F2FS_FEATURE_BLKZONED feature indicates that the drive was formatted
 with zone alignment optimization. This is optional for host-aware
devices, but mandatory for host-managed zoned block devices.
So check that the feature is set in this latter case.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

SMR stands for "Shingled Magnetic Recording" which makes sense
only for hard disk drives (spinning rust). The ZBC/ZAC standards
enable management of SMR disks, but solid state drives may also
support those standards. So rename the HMSMR feature to BLKZONED
to avoid a HDD centric terminology. For the same reason, rename
f2fs_sb_mounted_hmsmr to f2fs_sb_mounted_blkzoned.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch should fix an infinite loop case below.

F2FS-fs : inject IO error in f2fs_read_end_io+0xf3/0x120 [f2fs]
F2FS-fs (nvme0n1p1): recover_orphan_inode: orphan failed (ino=39ac1a), run fsck to fix.
...
[<ffffffffc0b11ede>] sync_meta_pages+0xae/0x270 [f2fs]
[<ffffffffc0b288dd>] ? flush_sit_entries+0x8d/0x960 [f2fs]
[<ffffffffc0b13801>] write_checkpoint+0x361/0xf20 [f2fs]
[<ffffffffb40e979d>] ? trace_hardirqs_on+0xd/0x10
[<ffffffffc0b0a199>] ? f2fs_sync_fs+0x79/0x190 [f2fs]
[<ffffffffc0b0a1a5>] f2fs_sync_fs+0x85/0x190 [f2fs]
[<ffffffffc0b2560e>] f2fs_balance_fs_bg+0x7e/0x1c0 [f2fs]
[<ffffffffc0b216c4>] f2fs_write_node_pages+0x34/0x320 [f2fs]
[<ffffffffb41dff21>] do_writepages+0x21/0x30
[<ffffffffb429edb1>] __writeback_single_inode+0x61/0x760
[<ffffffffb490a937>] ? _raw_spin_unlock+0x27/0x40
[<ffffffffb42a0805>] writeback_single_inode+0xd5/0x190
[<ffffffffb42a0959>] write_inode_now+0x99/0xc0
[<ffffffffb4289a16>] iput+0x1f6/0x2c0
[<ffffffffc0b0e3be>] f2fs_fill_super+0xe0e/0x1300 [f2fs]
[<ffffffffb426c394>] ? sget_userns+0x4f4/0x530
[<ffffffffb426c692>] mount_bdev+0x182/0x1b0
[<ffffffffc0b0d5b0>] ? f2fs_commit_super+0x100/0x100 [f2fs]
[<ffffffffc0b0a375>] f2fs_mount+0x15/0x20 [f2fs]
[<ffffffffb426d038>] mount_fs+0x38/0x170
[<ffffffffb428ec9b>] vfs_kern_mount+0x6b/0x160
[<ffffffffb4291d9e>] do_mount+0x1be/0xd60
[<ffffffffb4291a57>] ? copy_mount_options+0xb7/0x220
[<ffffffffb4292c54>] SyS_mount+0x94/0xd0
[<ffffffffb490b345>] entry_SYSCALL_64_fastpath+0x23/0xc6
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Report error of f2fs_fill_dentries to ->iterate_shared, otherwise when
error ocurrs, user may just list part of dirents in target directory
without any hints.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

gcc is unsure about the use of last_ofs_in_node, which might happen
without a prior initialization:

fs/f2fs//git/arm-soc/fs/f2fs/data.c: In function ‘f2fs_map_blocks’:
fs/f2fs/data.c:799:54: warning: ‘last_ofs_in_node’ may be used uninitialized in this function [-Wmaybe-uninitialized]
   if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {

As pointed out by Chao Yu, the code is actually correct as 'prealloc'
is only set if the last_ofs_in_node has been set, the two always
get updated together.

This initializes last_ofs_in_node to dn.ofs_in_node for each
new dnode at the start of the 'next_block' loop, which at that
point is a correct initialization as well. I assume that compilers
that correctly track the contents of the variables and do not
warn about the condition also figure out that they can eliminate
the extra assignment here.

Fixes: 46008c6d ("f2fs: support in batch multi blocks preallocation")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch removes percpu_count usage due to performance regression in iozone.

Fixes: 523be8a6 ("f2fs: use percpu_counter for page counters")
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch tries to make more clean inodes when flushing dirty inodes in
checkpoint.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This is to avoid no free segment bug during checkpoint caused by a number of
dirty inodes.

The case was reported by Chao like this.
1. mount with lazytime option
2. fill 4k file until disk is full
3. sync filesystem
4. read all files in the image
5. umount

In this case, we actually don't need to flush dirty inode to inode page during
checkpoint.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

We don't need to allocate bio partially in order to maximize sequential writes.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch avoids build warning.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

If inode becomes dirty, we need to check the # of dirty inodes whether or not
further checkpoint would be required.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

If there are a lot of dirty inodes, we need to flush all of them when doing
checkpoint. So, we need to count this for enough free space.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch makes sure it returns a positive value instead of a probable
casted negative value as shrink count.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

Let build_free_nids support sync/async methods, in allocation flow of nids,
we use synchronuous method, so that we can avoid looping in alloc_nid when
free memory is low; in unblock_operations and f2fs_balance_fs_bg we use
asynchronuous method in where low memory condition can interrupt us.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

This patch cleans up to use consistent free nid list ops.
Reviewed-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

During free nid allocation, in order to do preallocation, we will tag free
nid entry as allocated one and still leave it in free nid list, for other
allocators who want to grab free nids, it needs to traverse the free nid
list for lookup. It becomes overhead in scenario of allocating free nid
intensively by multithreads.

This patch splits free nid list to two list: {free,alloc}_nid_list, to
keep free nids and preallocated free nids separately, after that, traverse
latency will be gone, besides split nid_cnt for separate statistic.

Additionally, introduce __insert_nid_to_list and __remove_nid_from_list for
cleanup.
Signed-off-by: Chao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: modify f2fs_bug_on to avoid needless branches]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>