When committing inmem pages is successful, we revoke already committed
blocks in __revoke_inmem_pages() and finally replace the committed
ones with the old blocks using f2fs_replace_block(). However, if
the committed block was newly created one, the address of the old
block is NEW_ADDR and __f2fs_replace_block() cannot handle NEW_ADDR
as new_blkaddr properly and a kernel panic occurrs.
Signed-off-by: NDaeho Jeong <daeho.jeong@samsung.com>
Tested-by: NShu Tan <shu.tan@samsung.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

We take very long time to finish generic/476, this is because we will
check consistence of all discard entries in global rb tree while
traversing all different granularity pending lists, even when the list
is empty, in order to avoid that unneeded overhead, we have to skip
the check when coming up an empty list.

generic/476 time consumption:
					cost
Before patch & w/o consistence check	57s
Before patch & w/ consistence check	1426s
After patch				78s
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Fixes the following sparse warnings:

fs/f2fs/segment.c:887:6: warning:
 symbol '__check_sit_bitmap' was not declared. Should it be static?
fs/f2fs/segment.c:1327:6: warning:
 symbol 'f2fs_wait_discard_bio' was not declared. Should it be static?
fs/f2fs/super.c:1661:5: warning:
 symbol 'f2fs_get_projid' was not declared. Should it be static?
Signed-off-by: NWei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In some case, the node blocks has wrong blkaddr whose segment type is
NODE, e.g., recover inode has missing xattr flag and the blkaddr is in
the xattr range. Since fsck.f2fs does not check the recovery nodes, this
will cause __f2fs_replace_block change the curseg of node and do the
update_sit_entry(sbi, new_blkaddr, 1) with no next_blkoff refresh, as a
result, when recovery process write checkpoint and sync nodes, the
next_blkoff of curseg is used in the segment bit map, then it will
cause f2fs_bug_on. So let's check segment type in __f2fs_replace_block.
Signed-off-by: NYunlong Song <yunlong.song@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Let's avoid BUG_ON during fill_super, when on-disk was totall corrupted.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

No need return value in restore summary process
Signed-off-by: NYunlei He <heyunlei@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Use f2fs_k{m,z}alloc as much as possible to increase fault injection
points.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch supports to inject fault into kvmalloc/kvzalloc.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch introduces f2fs_kzalloc based on f2fs_kmalloc in order to
support error injection for kzalloc().
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Write hints helps F2FS to determine which type of segments would be
selected for buffered write.

This patch implements the mapping from write hints to segment types
as shown below.

  hints               segment type
  -----               ------------
  WRITE_LIFE_SHORT    CURSEG_HOT_DATA
  WRITE_LIFE_EXTREME  CURSEG_COLD_DATA
  others              CURSEG_WARM_DATA

the F2FS poliy for hot/cold seperation has precedence over this hints.
And hints are not applied in in-place update.
Signed-off-by: NHyunchul Lee <cheol.lee@lge.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Sometimes, after running generic/270 of fstest, fsck reports summary
info and actual position of block address in direct node becoming
inconsistent.

The root cause is race in between __f2fs_replace_block and change_curseg
as below:

Thread A				Thread B
- __clone_blkaddrs
 - f2fs_replace_block
  - __f2fs_replace_block
   - segnoA = GET_SEGNO(sbi, blkaddrA);
   - type = se->type:=CURSEG_HOT_DATA
   - if (!IS_CURSEG(sbi, segnoA))
         type = CURSEG_WARM_DATA
					- allocate_data_block
					 - allocate_segment
					  - get_ssr_segment
					  - change_curseg(segnoA, CURSEG_HOT_DATA)
   - change_curseg(segnoA, CURSEG_WARM_DATA)
    - reset_curseg
     - __set_sit_entry_type
      - change se->type from CURSEG_HOT_DATA to CURSEG_WARM_DATA

So finally, hot curseg locates in segnoA, but type of segnoA becomes
CURSEG_WARM_DATA.

Then if we invoke __f2fs_replace_block(blkaddrB, blkaddrA, true, false),
as blkaddrA locates in segnoA, so we will move warm type curseg to segnoA,
then change its summary cache and writeback it to summary block.

But segnoA is used by hot type curseg too, once it moves or persist, it
will cover summary block content with inner old summary cache, result in
inconsistent status.

This patch tries to fix this issue by introduce global curseg lock to avoid
race in between __f2fs_replace_block and change_curseg.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

After commit a468f0ef ("f2fs: use crc and cp version to determine
roll-forward recovery"), last caller of update_meta_page passing @src
with NULL is gone, so remove related dead code there.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

There are some cases user didn't update SIT cache under this lock,
so let's use rw_semaphore instead of mutex to enhance concurrently
accessing.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This reverts commit 5e443818

The commit should be reverted because call sequence of below two parts
of code must be kept:
a. update sit information, it needs to be updated before segment
allocation since latter allocation may trigger SSR, and SSR allocation
needs latest valid block information of all segments.
b. update segment status, it needs to be updated after segment allocation
since we can skip updating current opened segment status.

Fixes: 5e443818 ("f2fs: handle dirty segments inside refresh_sit_entry")
Suggested-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NYunlong Song <yunlong.song@huawei.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
[Jaegeuk Kim: remove refresh_sit_entry function]
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch exports min_ssr_segments threshold in sysfs to let user
control triggering SSR allocation flexibly.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

We have supported to issue discard in specified range during fstrim,
it needs to return caller with successfully trimmed bytes in that
range instead of bytes of invalid blocks which are scanned in
checkpoint.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch adds to support bio allocation error injection to simulate
out-of-memory test scenario.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

There are several assignments to variables that are redundant
as the values are never read when the variables are updated later
and so the redundant statements can be safely removed.

Cleans up clang warnings:
fs/f2fs/segment.c:923:19: warning: Value stored to 'p' during its initialization is never read
fs/f2fs/segment.c:2060:2: warning: Value stored to 'hint' is never read
fs/f2fs/segment.c:2353:2: warning: Value stored to 'start_block' is never read
fs/f2fs/segment.c:2354:2: warning: Value stored to 'end_block' is never read
Signed-off-by: NColin Ian King <colin.king@canonical.com>
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

If some abnormal users try lots of atomic write operations, f2fs is able to
produce pinned pages in the main memory which affects system performance.
This patch limits that as 20% over total memory size, and if f2fs reaches
to the limit, it will drop all the inmemory pages.
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In ->umount, once we drop remained discard entries, we should not
set CP_TRIMMED_FLAG with another checkpoint.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch adds tracepoint to trace f2fs_remove_discard.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

__submit_discard_cmd may lead long latency due to exhaustion of I/O
request resource in block layer, so issuing all discard under cmd_lock
may lead to hangtask, in order to avoid that, let's reduce it's coverage.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

There are many different scenarios such as fstrim, umount, urgent or
background where we will issue discards, actually, they need use
different policy in aspect of io aware, discard granularity, delay
interval and so on. But now they just share one common discard policy,
so there will be race when changing policy in between these scenarios,
the interference of changing discard policy will be very serious.

This patch changes to split discard policy for different scenarios.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch wraps scattered optional parameters into discard policy as
below, later, with it we expect that we can adjust these parameters with
proper strategy in different scenario.

struct discard_policy {
	unsigned int min_interval;	/* used for candidates exist */
	unsigned int max_interval;	/* used for candidates not exist */
	unsigned int max_requests;	/* # of discards issued per round */
	unsigned int io_aware_gran;	/* minimum granularity discard not be aware of I/O */
	bool io_aware;			/* issue discard in idle time */
	bool sync;			/* submit discard with REQ_SYNC flag */
};

This patch doesn't change any logic of codes.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Fstrim intends to trim invalid blocks of filesystem only with specified
range and granularity, but actually, it will issue all previous cached
discard commands which may be out-of-range and be with unmatched
granularity, it's unneeded.

In order to fix above issues, this patch introduces new helps to support
to issue and wait discard in range and adds a new fstrim_list for tracking
in-flight discard from ->fstrim.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

If f2fs manages multiple devices, in checkpoint, we need to issue flush
in those devices which contain dirty data/node in their cache before
we write checkpoint region, otherwise, filesystem metadata could be
corrupted if hitting SPO after checkpoint.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

When multiple device feature is enabled, during ->fsync we will issue
flush in all devices to make sure node/data of the file being persisted
into storage. But some flushes of device could be unneeded as file's
data may be not writebacked into those devices. So this patch adds and
manage bitmap per inode in global cache to indicate which device is
dirty and it needs to issue flush during ->fsync, hence, we could improve
performance of fsync in scenario of multiple device.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

As Ju Hyung Park reported:

"When 'fstrim' is called for manual trim, a BUG() can be triggered
randomly with this patch.

I'm seeing this issue on both x86 Desktop and arm64 Android phone.

On x86 Desktop, this was caused during Ubuntu boot-up. I have a
cronjob installed which calls 'fstrim -v /' during boot. On arm64
Android, this was caused during GC looping with 1ms gc_min_sleep_time
& gc_max_sleep_time."

Root cause of this issue is that f2fs_wait_discard_bios can only be
used by f2fs_put_super, because during put_super there must be no
other referrers, so it can ignore discard entry's reference count
when removing the entry, otherwise in other caller we will hit bug_on
in __remove_discard_cmd as there may be other issuer added reference
count in discard entry.

Thread A				Thread B
					- issue_discard_thread
- f2fs_ioc_fitrim
 - f2fs_trim_fs
  - f2fs_wait_discard_bios
   - __issue_discard_cmd
    - __submit_discard_cmd
					 - __wait_discard_cmd
					  - dc->ref++
					  - __wait_one_discard_bio
   - __wait_discard_cmd
    - __remove_discard_cmd
     - f2fs_bug_on(sbi, dc->ref)

Fixes: 969d1b18Reported-by: NJu Hyung Park <qkrwngud825@gmail.com>
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Once we encounter I/O interruption during issuing discards, we will delay
long time before next round, but if system status is I/O idle during the
time, it may loses opportunity to issue discards. So this patch changes
to hurry up to issue discard after io interruption.

Besides, this patch also fixes to issue discards accurately with assigned
rate.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch activates SSR in gc_urgent mode.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In android, we'd better wait for fstrim completion instead of issuing the
discard commands asynchronous.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch renames functions regarding to buffer management via META_MAPPING
used for encrypted blocks especially. We can actually use them in generic way.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

In scenario of remount_ro vs flush, after flush_thread exits in
->remount_fs, flusher will only clean up golbal issue_list, but
without waking up flushers waiting on that list, result in hang
related user threads.

In order to fix this issue, this patch enables the flusher to
take charge of issue_flush thread: executes merged flush command,
and wake up all sleeping flushers.

Fixes: 5eba8c5d ("f2fs: fix to access nullified flush_cmd_control pointer")
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

Previously, we will miss merging flush command during fsync due to below
race condition:

Thread A 		Thread B		Thread C
- f2fs_issue_flush
 - atomic_read(&issing_flush)
			- f2fs_issue_flush
			 - atomic_read(&issing_flush)
						- f2fs_issue_flush
						 - atomic_read(&issing_flush)
  - atomic_inc(&issing_flush)
			  - atomic_inc(&issing_flush)
						  - atomic_inc(&issing_flush)
   - submit_flush_wait
			   - submit_flush_wait
						   - submit_flush_wait

It needs to use atomic_inc_return instead to avoid such race.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

allocate_segment_by_default is the only caller of change_curseg passing
@reuse with 'false', but commit 763bfe1b ("f2fs: remove reusing any
prefree segments") removes the calling, after that, @reuse in
change_curseg always be true, so, let's clean up the unneeded parameter.
Signed-off-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This patch fixes to avoid needless wake ups.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

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: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

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: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

It's time to issue all the discard commands, if user sets the idle time.
Reviewed-by: NChao Yu <yuchao0@huawei.com>
Signed-off-by: NJaegeuk Kim <jaegeuk@kernel.org>

This way we don't need a block_device structure to submit I/O.  The
block_device has different life time rules from the gendisk and
request_queue and is usually only available when the block device node
is open.  Other callers need to explicitly create one (e.g. the lightnvm
passthrough code, or the new nvme multipathing code).

For the actual I/O path all that we need is the gendisk, which exists
once per block device.  But given that the block layer also does
partition remapping we additionally need a partition index, which is
used for said remapping in generic_make_request.

Note that all the block drivers generally want request_queue or
sometimes the gendisk, so this removes a layer of indirection all
over the stack.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>