fix missing skip pages info in f2fs_writepages trace event.
Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
f2fs use single bio buffer per type data (META/NODE/DATA) for caching
writes locating in continuous block address as many as possible, after
submitting, these writes may be still cached in bio buffer, so we have
to flush cached writes in bio buffer by calling f2fs_submit_merged_bio.
Unfortunately, in the scenario of high concurrency, bio buffer could be
flushed by someone else before we submit it as below reasons:
a) there is no space in bio buffer.
b) add a request of different type (SYNC, ASYNC).
c) add a discontinuous block address.
For this condition, f2fs_submit_merged_bio will be devastating, because
it could break the following merging of writes in bio buffer, split one
big bio into two smaller one.
This patch introduces f2fs_submit_merged_bio_cond which can do a
conditional submitting with bio buffer, before submitting it will judge
whether:
- page in DATA type bio buffer is matching with specified page;
- page in DATA type bio buffer is belong to specified inode;
- page in NODE type bio buffer is belong to specified inode;
If there is no eligible page in bio buffer, we will skip submitting step,
result in gaining more chance to merge consecutive block IOs in bio cache.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Sometimes, if cp_error is set, there remains under-writeback pages, resulting in
kernel hang in put_super.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Likewise f2fs_write_cache_pages, let's do for node and meta pages too.
Especially, for node blocks, we should do this before marking its fsync
and dentry flags.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch makes f2fs_map_blocks supporting returning next potential
page offset which skips hole region in indirect tree of inode, and
use it to speed up fiemap in handling big hole case.
Test method:
xfs_io -f /mnt/f2fs/file -c "pwrite 1099511627776 4096"
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"
Before:
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"
/mnt/f2fs/file:
EXT: FILE-OFFSET BLOCK-RANGE TOTAL FLAGS
0: [0..2147483647]: hole 2147483648
1: [2147483648..2147483655]: 81920..81927 8 0x1
real 3m3.518s
user 0m0.000s
sys 3m3.456s
After:
time xfs_io -f /mnt/f2fs/file -c "fiemap -v"
/mnt/f2fs/file:
EXT: FILE-OFFSET BLOCK-RANGE TOTAL FLAGS
0: [0..2147483647]: hole 2147483648
1: [2147483648..2147483655]: 81920..81927 8 0x1
real 0m0.008s
user 0m0.000s
sys 0m0.008s
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When seeking data in ->llseek, if we encounter a big hole which covers
several dnode pages, we will try to seek data from index of page which
is the first page of next dnode page, at most we could skip searching
(ADDRS_PER_BLOCK - 1) pages.
However it's still not efficient, because if our indirect/double-indirect
pointer are NULL, there are no dnode page locate in the tree indirect/
double-indirect pointer point to, it's not necessary to search the whole
region.
This patch introduces get_next_page_offset to calculate next page offset
based on current searching level and max searching level returned from
get_dnode_of_data, with this, we could skip searching the entire area
indirect or double-indirect node block is not exist.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
There are redundant pointer conversion in following call stack:
- at position a, inode was been converted to f2fs_file_info.
- at position b, f2fs_file_info was been converted to inode again.
- truncate_blocks(inode,..)
- fi = F2FS_I(inode) ---a
- ADDRS_PER_PAGE(node_page, fi)
- addrs_per_inode(fi)
- inode = &fi->vfs_inode ---b
- f2fs_has_inline_xattr(inode)
- fi = F2FS_I(inode)
- is_inode_flag_set(fi,..)
In order to avoid unneeded conversion, alter ADDRS_PER_PAGE and
addrs_per_inode to acept parameter with type of inode pointer.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch uses existing function f2fs_map_block to simplify implementation
of __allocate_data_blocks.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_map_blocks, we use duplicated codes to handle first block mapping
and the following blocks mapping, it's unnecessary. This patch simplifies
f2fs_map_blocks to avoid using copied codes.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces lifetime IO write statistics exposed to the sysfs interface.
The write IO amount is obtained from block layer, accumulated in the file system and
stored in the hot node summary of checkpoint.
Signed-off-by: Shuoran Liu <liushuoran@huawei.com>
Signed-off-by: Pengyang Hou <houpengyang@huawei.com>
[Jaegeuk Kim: add sysfs documentation]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
On the worst case, we need to scan the whole radix tree and every rb-tree to
free the victimed extent_nodes when shrinking.
Pengyang initially introduced a victim_list to record the victimed extent_nodes,
and free these extent_nodes by just scanning a list.
Later, Chao Yu enhances the original patch to improve memory footprint by
removing victim list.
The policy of lru list shrinking becomes:
1) lock lru list's lock
2) trylock extent tree's lock
3) remove extent node from lru list
4) unlock lru list's lock
5) do shrink
6) repeat 1) to 5)
Signed-off-by: Hou Pengyang <houpengyang@huawei.com>
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
There are three steps to free an extent node:
1) list_del_init, 2)__detach_extent_node, 3) kmem_cache_free
In path f2fs_destroy_extent_tree, 1->2->3 to free a node,
But in path f2fs_update_extent_tree_range, it is 2->1->3.
This patch makes all the order to be: 1->2->3
It makes sense, since in the next patch, we import a victim list in the
path shrink_extent_tree, we could check if the extent_node is in the victim
list by checking the list_empty(). So it is necessary to put 1) first.
Signed-off-by: Hou Pengyang <houpengyang@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
variable nsearched in get_victim_by_default() indicates the number of
dirty segments we already checked. There are 2 problems about the way
it updates:
1. When p.ofs_unit is greater than 1, the victim we find consists
of multiple segments, possibly more than 1 dirty segment.
But nsearched always increases by 1.
2. If segments have been found but not been chosen, nsearched won't
increase. So even we have checked all dirty segments, nsearched
may still less than p.max_search.
All these problems could cause unnecessary search after all dirty
segments have already been checked.
Signed-off-by: Fan li <fanofcode.li@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
no need to wait inline file page writeback for no one
use it, so this patch delete unnecessary wait.
Signed-off-by: Yunlei He <heyunlei@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In write_begin, if storage supports stable_page, we don't need to wait for
writeback to update its contents.
This patch introduces to use wait_for_stable_page instead of
wait_on_page_writeback.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If we configure section consist of multiple segments, foreground GC will
do the garbage collection with following approach:
for each segment in victim section
blk_start_plug
for each valid block in segment
write out by OPU method
submit bio cache <---
blk_finish_plug <---
There are two issue:
1) for most of the time, 'submit bio cache' will break the merging in
current bio buffer from writes of next segments, making a smaller bio
submitting.
2) block plug only cover IO submitting in one segment, which reduce
opportunity of merging IOs in plug with multiple segments.
So refactor the code as below structure to strive for biggest
opportunity of merging IOs:
blk_start_plug
for each segment in victim section
for each valid block in segment
write out by OPU method
submit bio cache
blk_finish_plug
Test method:
1. mkfs.f2fs -s 8 /dev/sdX
2. touch 32 files
3. write 2M data into each file
4. punch 1.5M data from offset 0 for each file
5. trigger foreground gc through ioctl
Before patch, there are totoally 40 bios submitted.
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 65536, size = 122880
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 65776, size = 122880
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 66016, size = 122880
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 66256, size = 122880
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 66496, size = 32768
----repeat for 8 times
After patch, there are totally 35 bios submitted.
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 65536, size = 122880
----repeat 34 times
f2fs_submit_write_bio: dev = (8,32), WRITE_SYNC, DATA, sector = 73696, size = 16384
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If end_io gets an error, we don't need to set the page as dirty, since we
already set f2fs_stop_checkpoint which will not flush any data.
This will resolve the following warning.
======================================================
[ INFO: HARDIRQ-safe -> HARDIRQ-unsafe lock order detected ]
4.4.0+ #9 Tainted: G O
------------------------------------------------------
xfs_io/26773 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
(&(&sbi->inode_lock[i])->rlock){+.+...}, at: [<ffffffffc025483f>] update_dirty_page+0x6f/0xd0 [f2fs]
and this task is already holding:
(&(&q->__queue_lock)->rlock){-.-.-.}, at: [<ffffffff81396ea2>] blk_queue_bio+0x422/0x490
which would create a new lock dependency:
(&(&q->__queue_lock)->rlock){-.-.-.} -> (&(&sbi->inode_lock[i])->rlock){+.+...}
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In write_begin, if there is an inline_data, f2fs loads it into 0'th data page.
Since it's the read path, we don't need to sync its inode page.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In write_end, we don't need to sync inode page at every time.
Instead, we can expect f2fs_write_inode will update later.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The sceanrio is:
1. create fully node blocks
2. flush node blocks
3. write inline_data for all the node blocks again
4. flush node blocks redundantly
So, this patch tries to flush inline_data when flushing node blocks.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The scenario is:
1. create lots of node blocks
2. sync
3. write lots of inline_data
-> got panic due to no free space
In that case, we should flush node blocks when writing inline_data in #3,
and trigger gc as well.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If there are many writepages calls by multiple threads in background, we don't
need to serialize to merge all the bios, since it's background.
In such the case, it'd better to run writepages concurrently.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
get_new_segment starts from current segment position, tries to search a
free segment among its right neighbors locate in same section.
But previously our search area was set as [current segment, max segment],
which means we have to search to more bits in free_segmap bitmap for some
worse cases. So here we correct the search area to [current segment, last
segment in section] to avoid unnecessary searching.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch exports a new sysfs entry 'dirty_nat_ratio' to control threshold
of dirty nat entries, if current ratio exceeds configured threshold,
checkpoint will be triggered in f2fs_balance_fs_bg for flushing dirty nats.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When testing f2fs with xfstest, generic/251 is stuck for long time,
the case uses below serials to obtain fresh released space in device,
in order to prepare for following fstrim test.
1. rm -rf /mnt/dir
2. mkdir /mnt/dir/
3. cp -axT `pwd`/ /mnt/dir/
4. goto 1
During preparing step, all nat entries will be cached in nat cache,
most of them are dirty entries with invalid blkaddr, which means
nodes related to these entries have been truncated, and they could
be reused after the dirty entries been checkpointed.
However, there was no checkpoint been triggered, so nid allocators
(e.g. mkdir, creat) will run into long journey of iterating all NAT
pages, looking for free nids in alloc_nid->build_free_nids.
Here, in f2fs_balance_fs_bg we give another chance to do checkpoint
to flush nat entries for reusing them in free nid cache when dirty
entry count exceeds 10% of max count.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Operations in is_merged_page is related to inner bio cache, move it to
data.c.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Currently we maintain perfect LRU list by moving entry to the tail of
the list when it gets used. However these operations on cache-global
list are relatively expensive.
In this patch we switch to lazy updates of LRU list. Whenever entry gets
used, we set a referenced bit in it. When reclaiming entries, we give
referenced entries another round in the LRU. Since the list is not a
real LRU anymore, rename it to just 'list'.
In my testing this logic gives about 30% boost to workloads with mostly
unique xattr blocks (e.g. xattr-bench with 10 files and 10000 unique
xattr values).
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
* multipath:
NFS add callback_ops to nfs4_proc_bind_conn_to_session_callback
pnfs/NFSv4.1: Add multipath capabilities to pNFS flexfiles servers over NFSv3
SUNRPC: Allow addition of new transports to a struct rpc_clnt
NFSv4.1: nfs4_proc_bind_conn_to_session must iterate over all connections
SUNRPC: Make NFS swap work with multipath
SUNRPC: Add a helper to apply a function to all the rpc_clnt's transports
SUNRPC: Allow caller to specify the transport to use
SUNRPC: Use the multipath iterator to assign a transport to each task
SUNRPC: Make rpc_clnt store the multipath iterators
SUNRPC: Add a structure to track multiple transports
SUNRPC: Make freeing of struct xprt rcu-safe
SUNRPC: Uninline xprt_get(); It isn't performance critical.
SUNRPC: Reorder rpc_task to put waitqueue related info in same cachelines
SUNRPC: Remove unused function rpc_task_reset_client
* nfsv41_cb:
NFSv4.x: Fix NFS4ERR_RETRY_UNCACHED_REP in nfs4_callback_sequence
NFSv4.x: Allow multiple callbacks in flight
NFSv4.x: Fix wraparound issues when validing the callback sequence id
NFSv4.x: Enforce the ca_maxresponsesize_cached on the back channel
NFSv4.x: CB_SEQUENCE should return NFS4ERR_DELAY if still executing
NFSv4.x: Remove hard coded slotids in callback channel
Replace another case where the layout 'plh_block_lgets' can trigger
infinite loops in send_layoutget().
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
If the server reboots while there is a layoutget outstanding, then
the call to pnfs_choose_layoutget_stateid() will fail with an EAGAIN
error, which causes an infinite loop in send_layoutget(). The reason
why we never break out of the loop is that the layout 'plh_block_lgets'
field is never cleared.
Fix is to replace plh_block_lgets with NFS_LAYOUT_INVALID_STID, which
can be reset after a new layoutget.
Fixes: ab7d763e47 ("pNFS: Ensure nfs4_layoutget_prepare returns...")
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
This patch fixes the problems with patch b3a5bbfd7.
1. It removes a return statement from lowcomms_error_report
because it needs to call the original error report in all paths
through the function.
2. All socket callbacks are saved and restored, not just the
sk_error_report, and that's done so with proper locking like
sunrpc does.
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
This patch replaces the call to nodeid_to_addr with a call to
kernel_getpeername. This avoids taking a spinlock because it may
potentially be called from a softirq context.
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: David Teigland <teigland@redhat.com>
So far number of entries in mbcache is limited only by the pressure from
the shrinker. Since too many entries degrade the hash table and
generally we expect that caching more entries has diminishing returns,
limit number of entries the same way as in the old mbcache to 16 * hash
table size.
Once we exceed the desired maximum number of entries, we schedule a
backround work to reclaim entries. If the background work cannot keep up
and the number of entries exceeds two times the desired maximum, we
reclaim some entries directly when allocating a new entry.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Both ext2 and ext4 are now converted to mbcache2. Remove the old mbcache
code.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The conversion is generally straightforward. We convert filesystem from
a global cache to per-fs one. Similarly to ext4 the tricky part is that
xattr block corresponding to found mbcache entry can get freed before we
get buffer lock for that block. So we have to check whether the entry is
still valid after getting the buffer lock.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The conversion is generally straightforward. The only tricky part is
that xattr block corresponding to found mbcache entry can get freed
before we get buffer lock for that block. So we have to check whether
the entry is still valid after getting buffer lock.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Original mbcache was designed to have more features than what ext?
filesystems ended up using. It supported entry being in more hashes, it
had a home-grown rwlocking of each entry, and one cache could cache
entries from multiple filesystems. This genericity also resulted in more
complex locking, larger cache entries, and generally more code
complexity.
This is reimplementation of the mbcache functionality to exactly fit the
purpose ext? filesystems use it for. Cache entries are now considerably
smaller (7 instead of 13 longs), the code is considerably smaller as
well (414 vs 913 lines of code), and IMO also simpler. The new code is
also much more lightweight.
I have measured the speed using artificial xattr-bench benchmark, which
spawns P processes, each process sets xattr for F different files, and
the value of xattr is randomly chosen from a pool of V values. Averages
of runtimes for 5 runs for various combinations of parameters are below.
The first value in each cell is old mbache, the second value is the new
mbcache.
V=10
F\P 1 2 4 8 16 32 64
10 0.158,0.157 0.208,0.196 0.500,0.277 0.798,0.400 3.258,0.584 13.807,1.047 61.339,2.803
100 0.172,0.167 0.279,0.222 0.520,0.275 0.825,0.341 2.981,0.505 12.022,1.202 44.641,2.943
1000 0.185,0.174 0.297,0.239 0.445,0.283 0.767,0.340 2.329,0.480 6.342,1.198 16.440,3.888
V=100
F\P 1 2 4 8 16 32 64
10 0.162,0.153 0.200,0.186 0.362,0.257 0.671,0.496 1.433,0.943 3.801,1.345 7.938,2.501
100 0.153,0.160 0.221,0.199 0.404,0.264 0.945,0.379 1.556,0.485 3.761,1.156 7.901,2.484
1000 0.215,0.191 0.303,0.246 0.471,0.288 0.960,0.347 1.647,0.479 3.916,1.176 8.058,3.160
V=1000
F\P 1 2 4 8 16 32 64
10 0.151,0.129 0.210,0.163 0.326,0.245 0.685,0.521 1.284,0.859 3.087,2.251 6.451,4.801
100 0.154,0.153 0.211,0.191 0.276,0.282 0.687,0.506 1.202,0.877 3.259,1.954 8.738,2.887
1000 0.145,0.179 0.202,0.222 0.449,0.319 0.899,0.333 1.577,0.524 4.221,1.240 9.782,3.579
V=10000
F\P 1 2 4 8 16 32 64
10 0.161,0.154 0.198,0.190 0.296,0.256 0.662,0.480 1.192,0.818 2.989,2.200 6.362,4.746
100 0.176,0.174 0.236,0.203 0.326,0.255 0.696,0.511 1.183,0.855 4.205,3.444 19.510,17.760
1000 0.199,0.183 0.240,0.227 1.159,1.014 2.286,2.154 6.023,6.039 ---,10.933 ---,36.620
V=100000
F\P 1 2 4 8 16 32 64
10 0.171,0.162 0.204,0.198 0.285,0.230 0.692,0.500 1.225,0.881 2.990,2.243 6.379,4.771
100 0.151,0.171 0.220,0.210 0.295,0.255 0.720,0.518 1.226,0.844 3.423,2.831 19.234,17.544
1000 0.192,0.189 0.249,0.225 1.162,1.043 2.257,2.093 5.853,4.997 ---,10.399 ---,32.198
We see that the new code is faster in pretty much all the cases and
starting from 4 processes there are significant gains with the new code
resulting in upto 20-times shorter runtimes. Also for large numbers of
cached entries all values for the old code could not be measured as the
kernel started hitting softlockups and died before the test completed.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In commit bcff24887d ("ext4: don't read blocks from disk after extents
being swapped") bh is not updated correctly in the for loop and wrong
data has been written to disk. generic/324 catches this on sub-page
block size ext4.
Fixes: bcff24887d ("ext4: don't read blocks from disk after extentsbeing swapped")
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org
Now, ext4_free_blocks() doesn't revoke data blocks of per-file data
journalled inode and it can cause file data inconsistency problems.
Even though data blocks of per-file data journalled inode are already
forgotten by jbd2_journal_invalidatepage() in advance of invoking
ext4_free_blocks(), we still need to revoke the data blocks here.
Moreover some of the metadata blocks, which are not found by
sb_find_get_block(), are still needed to be revoked, but this is also
missing here.
Signed-off-by: Daeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
