The current scrub implementation reuses bios and pages as often as possible,
allocating them only on start and releasing them when finished. This leads
to more problems with the block layer than it's worth. The elevator gets
confused when there are more pages added to the bio than bi_size suggests.
This patch completely rips out the reuse of bios and pages and allocates
them freshly for each submit.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Maosn <chris.mason@oracle.com>
write_dev_supers was changed to use RCU to protect the list of
devices, but it was then sleeping while it actually wrote the supers.
This fixes it to just use the mutex, since we really don't any
concurrency in write_dev_supers anyway.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
For a filesystem that has lots of files in it, the first time we mount
it with free ino caching support, it can take quite a long time to
setup the caching before we can create new files.
Here we fill the cache with [highest_ino, BTRFS_LAST_FREE_OBJECTID]
before we start the caching thread to search through the extent tree.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
scrub_page collects several pages into one bio as long as they are physically
contiguous. As we only save one logical address for the whole bio, don't
collect pages that are physically contiguous but logically discontiguous.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
This will detect small random writes into files and
queue the up for an auto defrag process. It isn't well suited to
database workloads yet, but works for smaller files such as rpm, sqlite
or bdb databases.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs_devices->devices is only updated on remove and add device paths, so we can
use rcu to protect it in the reader side
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Drop device_list_mutex for the reader side on clone_fs_devices and
btrfs_rm_device pathes since the fs_info->volume_mutex can ensure the device
list is not updated
btrfs_close_extra_devices is the initialized path, we can not add or remove
device at this time, so we can simply drop the mutex safely, like other
initialized function does(add_missing_dev, __find_device, __btrfs_open_devices
...).
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
On remove device path, it updates device->dev_alloc_list but does not hold
chunk lock
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
On btrfs_congested_fn and __unplug_io_fn paths, we should hold
device_list_mutex to avoid remove/add device path to
update fs_devices->devices
On __btrfs_close_devices and btrfs_prepare_sprout paths, the devices in
fs_devices->devices or fs_devices->devices is updated, so we should hold
the mutex to avoid the reader side to reach them
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
'bh' is forgot to release if no error is detected
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
merge_state can free the current state if it can be merged with the next node,
but in set_extent_bit(), after merge_state, we still use the current extent to
get the next node and cache it into cached_state
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
It doesn't allocate extent_state and check the result properly:
- in set_extent_bit, it doesn't allocate extent_state if the path is not
allowed wait
- in clear_extent_bit, it doesn't check the result after atomic-ly allocate,
we trigger BUG_ON() if it's fail
- if allocate fail, we trigger BUG_ON instead of returning -ENOMEM since
the return value of clear_extent_bit() is ignored by many callers
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Btrfs_alloc_path should be matched with btrfs_free_path in error-handling code.
A simplified version of the semantic match that finds this problem is as
follows: (http://coccinelle.lip6.fr/)
// <smpl>
@r exists@
local idexpression struct btrfs_path * x;
expression ra,rb;
position p1,p2;
@@
x = btrfs_alloc_path@p1(...)
... when != btrfs_free_path(x,...)
when != if (...) { ... btrfs_free_path(x,...) ...}
when != x = ra
if(...) { ... when != x = rb
when forall
when != btrfs_free_path(x,...)
\(return <+...x...+>; \| return@p2...; \) }
@script:python@
p1 << r.p1;
p2 << r.p2;
@@
cocci.print_main("alloc",p1)
cocci.print_secs("return",p2)
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
If return value of btrfs_inc_extent_ref() is not 0, BUG() is called.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When read_one_inode() fails, error code is returned to caller instead
of BUG_ON().
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Currently, btrfs_truncate_item and btrfs_extend_item returns only 0.
So, the check by BUG_ON in the caller is unnecessary.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The error code is returned instead of calling BUG_ON when
btrfs_del_item returns the error.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The error code is returned instead of calling BUG_ON when
btrfs_previous_item returns the error.
Signed-off-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Observed as a large delay when --mixed filesystem is filled up.
Test example:
1. create tiny --mixed FS:
$ dd if=/dev/zero of=2G.img seek=$((2048 * 1024 * 1024 - 1)) count=1 bs=1
$ mkfs.btrfs --mixed 2G.img
$ mount -oloop 2G.img /mnt/ut/
2. Try to fill it up:
$ dd if=/dev/urandom of=10M.file bs=10240 count=1024
$ seq 1 256 | while read file_no; do echo $file_no; time cp 10M.file ${file_no}.copy; done
Up to '200.copy' it goes fast, but when disk fills-up each -ENOSPC
message takes 3 seconds to pop-up _every_ ENOSPC (and in usermode linux
it's even more: 30-60 seconds!). (Maybe, time depends on kernel's timer resolution).
No IO, no CPU load, just rescheduling. Some debugging revealed busy spinning
in shrink_delalloc.
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
Reviewed-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
In 2008, commit b4f6c45dfb dropped the use
of fs/btrfs/version.sh, but left the script behind. Kill it.
Commit by Jamey Sharp and Josh Triplett.
Signed-off-by: Jamey Sharp <jamey@minilop.net>
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Cc: Chris Mason <chris.mason@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Btrfs's tree search ioctl has a field to indicate that no more than a
given number of records should be returned. The ioctl doesn't honour
this, as the tested value is not incremented until the end of the
copy_to_sk function. This patch removes an unnecessary local variable,
and updates the num_found counter as each key is found in the tree.
Signed-off-by: Hugo Mills <hugo@carfax.org.uk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
240f62c875 replaced the node_lock with rcu_read_lock, but forgot
to remove the actual lock in the data structure. Remove it here.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
On lookup we only want to read the inode item, so leave the path spinning. Also
we're just wholesale reading the leaf off, so map the leaf so we don't do a
bunch of kmap/kunmaps. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
If there are duplicate entries in the free space cache, discard the entire cache
and load it the old fashioned way. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
If we have a very large filesystem, we can spend a lot of time in
find_free_extent just trying to allocate from empty block groups. So instead
check to see if the block group even has enough space for the allocation, and if
not go on to the next block group.
Signed-off-by: Josef Bacik <josef@redhat.com>
Our readahead is sort of sloppy, and really isn't always needed. For example if
ls is doing a stating ls (which is the default) it's going to stat in non-disk
order, so if say you have a directory with a stupid amount of files, readahead
is going to do nothing but waste time in the case of doing the stat. Taking the
unconditional readahead out made my test go from 57 minutes to 36 minutes. This
means that everywhere we do loop through the tree we want to make sure we do set
path->reada properly, so I went through and found all of the places where we
loop through the path and set reada to 1. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
When the fs is super full and we unmount the fs, we could get stuck in this
thing where unmount is waiting for the caching kthread to make progress and the
caching kthread keeps scheduling because we're in the middle of a commit. So
instead just let the caching kthread keep going and only yeild if
need_resched(). This makes my horrible umount case go from taking up to 10
minutes to taking less than 20 seconds. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Originally this was going to be used as a way to give hints to the allocator,
but frankly we can get much better hints elsewhere and it's not even used at all
for anything usefull. In addition to be completely useless, when we initialize
an inode we try and find a freeish block group to set as the inodes block group,
and with a completely full 40gb fs this takes _forever_, so I imagine with say
1tb fs this is just unbearable. So just axe the thing altoghether, we don't
need it and it saves us 8 bytes in the inode and saves us 500 microseconds per
inode lookup in my testcase. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We have a bit of debugging in btrfs_search_slot to make sure the level of the
cow block is the same as the original block we were cow'ing. I don't think I've
ever seen this tripped, so kill it. This saves us 2 kmap's per level in our
search. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
If we have particularly full nodes, we could call btrfs_node_blockptr up to 32
times, which is 32 pairs of kmap/kunmap, which _sucks_. So go ahead and map the
extent buffer while we look for readahead targets. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
In count_range_bits we are adjusting total_bytes based on the range we are
searching for, but we don't adjust the range start according to the range we are
searching for, which makes for weird results. For example, if the range
[0-8192]
is set DELALLOC, but I search for 4096-8192, I will get back 4096 for the number
of bytes found, but the range_start will be 0, which makes it look like the
range is [0-4096]. So instead set range_start = max(cur_start, state->start).
This makes everything come out right. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
The ceph guys keep running into problems where we have space reserved in our
orphan block rsv when freeing it up. This is because they tend to do snapshots
alot, so their truncates tend to use a bunch of space, so when we go to do
things like update the inode we have to steal reservation space in order to make
the reservation happen. This happens because truncate can use as much space as
it freaking feels like, but we still have to hold space for removing the orphan
item and updating the inode, which will definitely always happen. So in order
to fix this we need to split all of the reservation stuf up. So with this patch
we have
1) The orphan block reserve which only holds the space for deleting our orphan
item when everything is over.
2) The truncate block reserve which gets allocated and used specifically for the
space that the truncate will use on a per truncate basis.
3) The transaction will always have 1 item's worth of data reserved so we can
update the inode normally.
Hopefully this will make the ceph problem go away. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We use trans_mutex for lots of things, here's a basic list
1) To serialize trans_handles joining the currently running transaction
2) To make sure that no new trans handles are started while we are committing
3) To protect the dead_roots list and the transaction lists
Really the serializing trans_handles joining is not too hard, and can really get
bogged down in acquiring a reference to the transaction. So replace the
trans_mutex with a trans_lock spinlock and use it to do the following
1) Protect fs_info->running_transaction. All trans handles have to do is check
this, and then take a reference of the transaction and keep on going.
2) Protect the fs_info->trans_list. This doesn't get used too much, basically
it just holds the current transactions, which will usually just be the currently
committing transaction and the currently running transaction at most.
3) Protect the dead roots list. This is only ever processed by splicing the
list so this is relatively simple.
4) Protect the fs_info->reloc_ctl stuff. This is very lightweight and was using
the trans_mutex before, so this is a pretty straightforward change.
5) Protect fs_info->no_trans_join. Because we don't hold the trans_lock over
the entirety of the commit we need to have a way to block new people from
creating a new transaction while we're doing our work. So we set no_trans_join
and in join_transaction we test to see if that is set, and if it is we do a
wait_on_commit.
6) Make the transaction use count atomic so we don't need to take locks to
modify it when we're dropping references.
7) Add a commit_lock to the transaction to make sure multiple people trying to
commit the same transaction don't race and commit at the same time.
8) Make open_ioctl_trans an atomic so we don't have to take any locks for ioctl
trans.
I have tested this with xfstests, but obviously it is a pretty hairy change so
lots of testing is greatly appreciated. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
We currently track trans handles in current->journal_info, but we don't actually
use it. This patch fixes it. This will cover the case where we have multiple
people starting transactions down the call chain. This keeps us from having to
allocate a new handle and all of that, we just increase the use count of the
current handle, save the old block_rsv, and return. I tested this with xfstests
and it worked out fine. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
I keep forgetting that btrfs_join_transaction() just ignores the num_items
argument, which leads me to sending pointless patches and looking stupid :). So
just kill the num_items argument from btrfs_join_transaction and
btrfs_start_ioctl_transaction, since neither of them use it. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
In the prealloc filling code and compressed code we don't set trans->block_rsv
to the delalloc block reserve properly, which is going to make us use metadata
from the wrong pool, this patch fixes that. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
The current code relogs the entire inode every time during fsync log,
and it is much better suited to small files rather than large ones.
During my performance test, the fsync performace of large files sucks,
and we can ascribe this to the tremendous amount of csum infos of the
large ones, cause we have to flush all of these csum infos into log trees
even when there are only _one_ change in the whole file data. Apparently,
to optimize fsync, we need to create a filter to skip the unnecessary csum
ones, that is, the corresponding file data remains unchanged before this fsync.
Here I have some test results to show, I use sysbench to do "random write + fsync".
===
sysbench --test=fileio --num-threads=1 --file-num=2 --file-block-size=4K --file-total-size=8G --file-test-mode=rndwr --file-io-mode=sync --file-extra-flags= [prepare, run]
===
Sysbench args:
- Number of threads: 1
- Extra file open flags: 0
- 2 files, 4Gb each
- Block size 4Kb
- Number of random requests for random IO: 10000
- Read/Write ratio for combined random IO test: 1.50
- Periodic FSYNC enabled, calling fsync() each 100 requests.
- Calling fsync() at the end of test, Enabled.
- Using synchronous I/O mode
- Doing random write test
Sysbench results:
===
Operations performed: 0 Read, 10000 Write, 200 Other = 10200 Total
Read 0b Written 39.062Mb Total transferred 39.062Mb
===
a) without patch: (*SPEED* : 451.01Kb/sec)
112.75 Requests/sec executed
b) with patch: (*SPEED* : 4.7533Mb/sec)
1216.84 Requests/sec executed
PS: I've made a _sub transid_ stuff patch, but it does not perform as effectively as this patch,
and I'm wanderring where the problem is and trying to improve it more.
Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Changelog V5 -> V6:
- Fix oom when the memory load is high, by storing the delayed nodes into the
root's radix tree, and letting btrfs inodes go.
Changelog V4 -> V5:
- Fix the race on adding the delayed node to the inode, which is spotted by
Chris Mason.
- Merge Chris Mason's incremental patch into this patch.
- Fix deadlock between readdir() and memory fault, which is reported by
Itaru Kitayama.
Changelog V3 -> V4:
- Fix nested lock, which is reported by Itaru Kitayama, by updating space cache
inode in time.
Changelog V2 -> V3:
- Fix the race between the delayed worker and the task which does delayed items
balance, which is reported by Tsutomu Itoh.
- Modify the patch address David Sterba's comment.
- Fix the bug of the cpu recursion spinlock, reported by Chris Mason
Changelog V1 -> V2:
- break up the global rb-tree, use a list to manage the delayed nodes,
which is created for every directory and file, and used to manage the
delayed directory name index items and the delayed inode item.
- introduce a worker to deal with the delayed nodes.
Compare with Ext3/4, the performance of file creation and deletion on btrfs
is very poor. the reason is that btrfs must do a lot of b+ tree insertions,
such as inode item, directory name item, directory name index and so on.
If we can do some delayed b+ tree insertion or deletion, we can improve the
performance, so we made this patch which implemented delayed directory name
index insertion/deletion and delayed inode update.
Implementation:
- introduce a delayed root object into the filesystem, that use two lists to
manage the delayed nodes which are created for every file/directory.
One is used to manage all the delayed nodes that have delayed items. And the
other is used to manage the delayed nodes which is waiting to be dealt with
by the work thread.
- Every delayed node has two rb-tree, one is used to manage the directory name
index which is going to be inserted into b+ tree, and the other is used to
manage the directory name index which is going to be deleted from b+ tree.
- introduce a worker to deal with the delayed operation. This worker is used
to deal with the works of the delayed directory name index items insertion
and deletion and the delayed inode update.
When the delayed items is beyond the lower limit, we create works for some
delayed nodes and insert them into the work queue of the worker, and then
go back.
When the delayed items is beyond the upper bound, we create works for all
the delayed nodes that haven't been dealt with, and insert them into the work
queue of the worker, and then wait for that the untreated items is below some
threshold value.
- When we want to insert a directory name index into b+ tree, we just add the
information into the delayed inserting rb-tree.
And then we check the number of the delayed items and do delayed items
balance. (The balance policy is above.)
- When we want to delete a directory name index from the b+ tree, we search it
in the inserting rb-tree at first. If we look it up, just drop it. If not,
add the key of it into the delayed deleting rb-tree.
Similar to the delayed inserting rb-tree, we also check the number of the
delayed items and do delayed items balance.
(The same to inserting manipulation)
- When we want to update the metadata of some inode, we cached the data of the
inode into the delayed node. the worker will flush it into the b+ tree after
dealing with the delayed insertion and deletion.
- We will move the delayed node to the tail of the list after we access the
delayed node, By this way, we can cache more delayed items and merge more
inode updates.
- If we want to commit transaction, we will deal with all the delayed node.
- the delayed node will be freed when we free the btrfs inode.
- Before we log the inode items, we commit all the directory name index items
and the delayed inode update.
I did a quick test by the benchmark tool[1] and found we can improve the
performance of file creation by ~15%, and file deletion by ~20%.
Before applying this patch:
Create files:
Total files: 50000
Total time: 1.096108
Average time: 0.000022
Delete files:
Total files: 50000
Total time: 1.510403
Average time: 0.000030
After applying this patch:
Create files:
Total files: 50000
Total time: 0.932899
Average time: 0.000019
Delete files:
Total files: 50000
Total time: 1.215732
Average time: 0.000024
[1] http://marc.info/?l=linux-btrfs&m=128212635122920&q=p3
Many thanks for Kitayama-san's help!
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Reviewed-by: David Sterba <dave@jikos.cz>
Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Tested-by: Itaru Kitayama <kitayama@cl.bb4u.ne.jp>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
