2018-04-03 20:16:55 +03:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0 */
|
2008-06-26 00:01:30 +04:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2008 Oracle. All rights reserved.
|
|
|
|
*/
|
|
|
|
|
2018-04-03 20:16:55 +03:00
|
|
|
#ifndef BTRFS_LOCKING_H
|
|
|
|
#define BTRFS_LOCKING_H
|
2008-06-26 00:01:30 +04:00
|
|
|
|
2020-01-30 15:59:44 +03:00
|
|
|
#include <linux/atomic.h>
|
|
|
|
#include <linux/wait.h>
|
|
|
|
#include <linux/percpu_counter.h>
|
2019-09-24 19:44:24 +03:00
|
|
|
#include "extent_io.h"
|
|
|
|
|
2011-07-16 23:23:14 +04:00
|
|
|
#define BTRFS_WRITE_LOCK 1
|
|
|
|
#define BTRFS_READ_LOCK 2
|
|
|
|
#define BTRFS_WRITE_LOCK_BLOCKING 3
|
|
|
|
#define BTRFS_READ_LOCK_BLOCKING 4
|
|
|
|
|
2020-08-20 18:46:02 +03:00
|
|
|
/*
|
|
|
|
* We are limited in number of subclasses by MAX_LOCKDEP_SUBCLASSES, which at
|
|
|
|
* the time of this patch is 8, which is how many we use. Keep this in mind if
|
|
|
|
* you decide you want to add another subclass.
|
|
|
|
*/
|
|
|
|
enum btrfs_lock_nesting {
|
|
|
|
BTRFS_NESTING_NORMAL,
|
|
|
|
|
2020-08-20 18:46:03 +03:00
|
|
|
/*
|
|
|
|
* When we COW a block we are holding the lock on the original block,
|
|
|
|
* and since our lockdep maps are rootid+level, this confuses lockdep
|
|
|
|
* when we lock the newly allocated COW'd block. Handle this by having
|
|
|
|
* a subclass for COW'ed blocks so that lockdep doesn't complain.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_COW,
|
|
|
|
|
2020-08-20 18:46:04 +03:00
|
|
|
/*
|
|
|
|
* Oftentimes we need to lock adjacent nodes on the same level while
|
|
|
|
* still holding the lock on the original node we searched to, such as
|
|
|
|
* for searching forward or for split/balance.
|
|
|
|
*
|
|
|
|
* Because of this we need to indicate to lockdep that this is
|
|
|
|
* acceptable by having a different subclass for each of these
|
|
|
|
* operations.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_LEFT,
|
|
|
|
BTRFS_NESTING_RIGHT,
|
|
|
|
|
2020-08-20 18:46:05 +03:00
|
|
|
/*
|
|
|
|
* When splitting we will be holding a lock on the left/right node when
|
|
|
|
* we need to cow that node, thus we need a new set of subclasses for
|
|
|
|
* these two operations.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_LEFT_COW,
|
|
|
|
BTRFS_NESTING_RIGHT_COW,
|
|
|
|
|
2020-08-20 18:46:06 +03:00
|
|
|
/*
|
|
|
|
* When splitting we may push nodes to the left or right, but still use
|
|
|
|
* the subsequent nodes in our path, keeping our locks on those adjacent
|
|
|
|
* blocks. Thus when we go to allocate a new split block we've already
|
|
|
|
* used up all of our available subclasses, so this subclass exists to
|
|
|
|
* handle this case where we need to allocate a new split block.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_SPLIT,
|
|
|
|
|
2020-08-20 18:46:07 +03:00
|
|
|
/*
|
|
|
|
* When promoting a new block to a root we need to have a special
|
|
|
|
* subclass so we don't confuse lockdep, as it will appear that we are
|
|
|
|
* locking a higher level node before a lower level one. Copying also
|
|
|
|
* has this problem as it appears we're locking the same block again
|
|
|
|
* when we make a snapshot of an existing root.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_NEW_ROOT,
|
|
|
|
|
2020-08-20 18:46:02 +03:00
|
|
|
/*
|
|
|
|
* We are limited to MAX_LOCKDEP_SUBLCLASSES number of subclasses, so
|
|
|
|
* add this in here and add a static_assert to keep us from going over
|
|
|
|
* the limit. As of this writing we're limited to 8, and we're
|
|
|
|
* definitely using 8, hence this check to keep us from messing up in
|
|
|
|
* the future.
|
|
|
|
*/
|
|
|
|
BTRFS_NESTING_MAX,
|
|
|
|
};
|
|
|
|
|
|
|
|
static_assert(BTRFS_NESTING_MAX <= MAX_LOCKDEP_SUBCLASSES,
|
|
|
|
"too many lock subclasses defined");
|
|
|
|
|
2020-01-30 15:59:44 +03:00
|
|
|
struct btrfs_path;
|
|
|
|
|
2020-08-20 18:46:02 +03:00
|
|
|
void __btrfs_tree_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest);
|
2012-03-01 17:56:26 +04:00
|
|
|
void btrfs_tree_lock(struct extent_buffer *eb);
|
|
|
|
void btrfs_tree_unlock(struct extent_buffer *eb);
|
Btrfs: Change btree locking to use explicit blocking points
Most of the btrfs metadata operations can be protected by a spinlock,
but some operations still need to schedule.
So far, btrfs has been using a mutex along with a trylock loop,
most of the time it is able to avoid going for the full mutex, so
the trylock loop is a big performance gain.
This commit is step one for getting rid of the blocking locks entirely.
btrfs_tree_lock takes a spinlock, and the code explicitly switches
to a blocking lock when it starts an operation that can schedule.
We'll be able get rid of the blocking locks in smaller pieces over time.
Tracing allows us to find the most common cause of blocking, so we
can start with the hot spots first.
The basic idea is:
btrfs_tree_lock() returns with the spin lock held
btrfs_set_lock_blocking() sets the EXTENT_BUFFER_BLOCKING bit in
the extent buffer flags, and then drops the spin lock. The buffer is
still considered locked by all of the btrfs code.
If btrfs_tree_lock gets the spinlock but finds the blocking bit set, it drops
the spin lock and waits on a wait queue for the blocking bit to go away.
Much of the code that needs to set the blocking bit finishes without actually
blocking a good percentage of the time. So, an adaptive spin is still
used against the blocking bit to avoid very high context switch rates.
btrfs_clear_lock_blocking() clears the blocking bit and returns
with the spinlock held again.
btrfs_tree_unlock() can be called on either blocking or spinning locks,
it does the right thing based on the blocking bit.
ctree.c has a helper function to set/clear all the locked buffers in a
path as blocking.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-02-04 17:25:08 +03:00
|
|
|
|
2020-08-20 18:46:02 +03:00
|
|
|
void __btrfs_tree_read_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest,
|
|
|
|
bool recurse);
|
2011-07-16 23:23:14 +04:00
|
|
|
void btrfs_tree_read_lock(struct extent_buffer *eb);
|
|
|
|
void btrfs_tree_read_unlock(struct extent_buffer *eb);
|
|
|
|
void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb);
|
2018-04-04 02:43:05 +03:00
|
|
|
void btrfs_set_lock_blocking_read(struct extent_buffer *eb);
|
|
|
|
void btrfs_set_lock_blocking_write(struct extent_buffer *eb);
|
2011-07-16 23:23:14 +04:00
|
|
|
int btrfs_try_tree_read_lock(struct extent_buffer *eb);
|
|
|
|
int btrfs_try_tree_write_lock(struct extent_buffer *eb);
|
2014-11-19 21:25:09 +03:00
|
|
|
int btrfs_tree_read_lock_atomic(struct extent_buffer *eb);
|
2020-08-20 18:46:01 +03:00
|
|
|
struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root);
|
|
|
|
struct extent_buffer *__btrfs_read_lock_root_node(struct btrfs_root *root,
|
|
|
|
bool recurse);
|
|
|
|
|
|
|
|
static inline struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
|
|
|
|
{
|
|
|
|
return __btrfs_read_lock_root_node(root, false);
|
|
|
|
}
|
2014-11-19 21:25:09 +03:00
|
|
|
|
2019-09-24 19:44:24 +03:00
|
|
|
#ifdef CONFIG_BTRFS_DEBUG
|
|
|
|
static inline void btrfs_assert_tree_locked(struct extent_buffer *eb) {
|
btrfs: switch extent buffer tree lock to rw_semaphore
Historically we've implemented our own locking because we wanted to be
able to selectively spin or sleep based on what we were doing in the
tree. For instance, if all of our nodes were in cache then there's
rarely a reason to need to sleep waiting for node locks, as they'll
likely become available soon. At the time this code was written the
rw_semaphore didn't do adaptive spinning, and thus was orders of
magnitude slower than our home grown locking.
However now the opposite is the case. There are a few problems with how
we implement blocking locks, namely that we use a normal waitqueue and
simply wake everybody up in reverse sleep order. This leads to some
suboptimal performance behavior, and a lot of context switches in highly
contended cases. The rw_semaphores actually do this properly, and also
have adaptive spinning that works relatively well.
The locking code is also a bit of a bear to understand, and we lose the
benefit of lockdep for the most part because the blocking states of the
lock are simply ad-hoc and not mapped into lockdep.
So rework the locking code to drop all of this custom locking stuff, and
simply use a rw_semaphore for everything. This makes the locking much
simpler for everything, as we can now drop a lot of cruft and blocking
transitions. The performance numbers vary depending on the workload,
because generally speaking there doesn't tend to be a lot of contention
on the btree. However, on my test system which is an 80 core single
socket system with 256GiB of RAM and a 2TiB NVMe drive I get the
following results (with all debug options off):
dbench 200 baseline
Throughput 216.056 MB/sec 200 clients 200 procs max_latency=1471.197 ms
dbench 200 with patch
Throughput 737.188 MB/sec 200 clients 200 procs max_latency=714.346 ms
Previously we also used fs_mark to test this sort of contention, and
those results are far less impressive, mostly because there's not enough
tasks to really stress the locking
fs_mark -d /d[0-15] -S 0 -L 20 -n 100000 -s 0 -t 16
baseline
Average Files/sec: 160166.7
p50 Files/sec: 165832
p90 Files/sec: 123886
p99 Files/sec: 123495
real 3m26.527s
user 2m19.223s
sys 48m21.856s
patched
Average Files/sec: 164135.7
p50 Files/sec: 171095
p90 Files/sec: 122889
p99 Files/sec: 113819
real 3m29.660s
user 2m19.990s
sys 44m12.259s
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2020-08-20 18:46:09 +03:00
|
|
|
lockdep_assert_held(&eb->lock);
|
2019-09-24 19:44:24 +03:00
|
|
|
}
|
|
|
|
#else
|
|
|
|
static inline void btrfs_assert_tree_locked(struct extent_buffer *eb) { }
|
|
|
|
#endif
|
2011-07-16 23:23:14 +04:00
|
|
|
|
2019-09-24 20:17:17 +03:00
|
|
|
void btrfs_set_path_blocking(struct btrfs_path *p);
|
2019-09-24 20:17:17 +03:00
|
|
|
void btrfs_unlock_up_safe(struct btrfs_path *path, int level);
|
2019-09-24 20:17:17 +03:00
|
|
|
|
2011-07-16 23:23:14 +04:00
|
|
|
static inline void btrfs_tree_unlock_rw(struct extent_buffer *eb, int rw)
|
|
|
|
{
|
|
|
|
if (rw == BTRFS_WRITE_LOCK || rw == BTRFS_WRITE_LOCK_BLOCKING)
|
|
|
|
btrfs_tree_unlock(eb);
|
|
|
|
else if (rw == BTRFS_READ_LOCK_BLOCKING)
|
|
|
|
btrfs_tree_read_unlock_blocking(eb);
|
|
|
|
else if (rw == BTRFS_READ_LOCK)
|
|
|
|
btrfs_tree_read_unlock(eb);
|
|
|
|
else
|
|
|
|
BUG();
|
|
|
|
}
|
|
|
|
|
2020-01-30 15:59:44 +03:00
|
|
|
struct btrfs_drew_lock {
|
|
|
|
atomic_t readers;
|
|
|
|
struct percpu_counter writers;
|
|
|
|
wait_queue_head_t pending_writers;
|
|
|
|
wait_queue_head_t pending_readers;
|
|
|
|
};
|
|
|
|
|
|
|
|
int btrfs_drew_lock_init(struct btrfs_drew_lock *lock);
|
|
|
|
void btrfs_drew_lock_destroy(struct btrfs_drew_lock *lock);
|
|
|
|
void btrfs_drew_write_lock(struct btrfs_drew_lock *lock);
|
|
|
|
bool btrfs_drew_try_write_lock(struct btrfs_drew_lock *lock);
|
|
|
|
void btrfs_drew_write_unlock(struct btrfs_drew_lock *lock);
|
|
|
|
void btrfs_drew_read_lock(struct btrfs_drew_lock *lock);
|
|
|
|
void btrfs_drew_read_unlock(struct btrfs_drew_lock *lock);
|
|
|
|
|
2008-06-26 00:01:30 +04:00
|
|
|
#endif
|