2018-04-03 20:16:55 +03:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0 */
|
2009-03-13 17:10:06 +03:00
|
|
|
/*
|
|
|
|
* Copyright (C) 2008 Oracle. All rights reserved.
|
|
|
|
*/
|
2018-04-03 20:16:55 +03:00
|
|
|
|
|
|
|
#ifndef BTRFS_DELAYED_REF_H
|
|
|
|
#define BTRFS_DELAYED_REF_H
|
2009-03-13 17:10:06 +03:00
|
|
|
|
2017-03-03 11:55:15 +03:00
|
|
|
#include <linux/refcount.h>
|
|
|
|
|
2012-09-21 04:21:01 +04:00
|
|
|
/* these are the possible values of struct btrfs_delayed_ref_node->action */
|
2009-03-13 17:10:06 +03:00
|
|
|
#define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */
|
|
|
|
#define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */
|
|
|
|
#define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
|
2009-03-25 16:55:11 +03:00
|
|
|
#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
|
2009-03-13 17:10:06 +03:00
|
|
|
|
|
|
|
struct btrfs_delayed_ref_node {
|
2017-10-19 21:16:00 +03:00
|
|
|
struct rb_node ref_node;
|
btrfs: improve delayed refs iterations
This issue was found when I tried to delete a heavily reflinked file,
when deleting such files, other transaction operation will not have a
chance to make progress, for example, start_transaction() will blocked
in wait_current_trans(root) for long time, sometimes it even triggers
soft lockups, and the time taken to delete such heavily reflinked file
is also very large, often hundreds of seconds. Using perf top, it reports
that:
PerfTop: 7416 irqs/sec kernel:99.8% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
---------------------------------------------------------------------------------------
84.37% [btrfs] [k] __btrfs_run_delayed_refs.constprop.80
11.02% [kernel] [k] delay_tsc
0.79% [kernel] [k] _raw_spin_unlock_irq
0.78% [kernel] [k] _raw_spin_unlock_irqrestore
0.45% [kernel] [k] do_raw_spin_lock
0.18% [kernel] [k] __slab_alloc
It seems __btrfs_run_delayed_refs() took most cpu time, after some debug
work, I found it's select_delayed_ref() causing this issue, for a delayed
head, in our case, it'll be full of BTRFS_DROP_DELAYED_REF nodes, but
select_delayed_ref() will firstly try to iterate node list to find
BTRFS_ADD_DELAYED_REF nodes, obviously it's a disaster in this case, and
waste much time.
To fix this issue, we introduce a new ref_add_list in struct btrfs_delayed_ref_head,
then in select_delayed_ref(), if this list is not empty, we can directly use
nodes in this list. With this patch, it just took about 10~15 seconds to
delte the same file. Now using perf top, it reports that:
PerfTop: 2734 irqs/sec kernel:99.5% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
----------------------------------------------------------------------------------------
20.74% [kernel] [k] _raw_spin_unlock_irqrestore
16.33% [kernel] [k] __slab_alloc
5.41% [kernel] [k] lock_acquired
4.42% [kernel] [k] lock_acquire
4.05% [kernel] [k] lock_release
3.37% [kernel] [k] _raw_spin_unlock_irq
For normal files, this patch also gives help, at least we do not need to
iterate whole list to found BTRFS_ADD_DELAYED_REF nodes.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2016-10-26 13:07:33 +03:00
|
|
|
/*
|
|
|
|
* If action is BTRFS_ADD_DELAYED_REF, also link this node to
|
|
|
|
* ref_head->ref_add_list, then we do not need to iterate the
|
|
|
|
* whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes.
|
|
|
|
*/
|
|
|
|
struct list_head add_list;
|
2015-03-30 12:03:00 +03:00
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/* the starting bytenr of the extent */
|
|
|
|
u64 bytenr;
|
|
|
|
|
|
|
|
/* the size of the extent */
|
|
|
|
u64 num_bytes;
|
|
|
|
|
2011-09-14 14:37:00 +04:00
|
|
|
/* seq number to keep track of insertion order */
|
|
|
|
u64 seq;
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/* ref count on this data structure */
|
2017-03-03 11:55:15 +03:00
|
|
|
refcount_t refs;
|
2009-03-13 17:10:06 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* how many refs is this entry adding or deleting. For
|
|
|
|
* head refs, this may be a negative number because it is keeping
|
|
|
|
* track of the total mods done to the reference count.
|
|
|
|
* For individual refs, this will always be a positive number
|
|
|
|
*
|
|
|
|
* It may be more than one, since it is possible for a single
|
|
|
|
* parent to have more than one ref on an extent
|
|
|
|
*/
|
|
|
|
int ref_mod;
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
unsigned int action:8;
|
|
|
|
unsigned int type:8;
|
2009-03-13 17:10:06 +03:00
|
|
|
/* is this node still in the rbtree? */
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
unsigned int is_head:1;
|
2009-03-13 17:10:06 +03:00
|
|
|
unsigned int in_tree:1;
|
|
|
|
};
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
struct btrfs_delayed_extent_op {
|
|
|
|
struct btrfs_disk_key key;
|
2015-11-30 18:51:29 +03:00
|
|
|
u8 level;
|
|
|
|
bool update_key;
|
|
|
|
bool update_flags;
|
|
|
|
bool is_data;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
u64 flags_to_set;
|
|
|
|
};
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/*
|
|
|
|
* the head refs are used to hold a lock on a given extent, which allows us
|
|
|
|
* to make sure that only one process is running the delayed refs
|
|
|
|
* at a time for a single extent. They also store the sum of all the
|
|
|
|
* reference count modifications we've queued up.
|
|
|
|
*/
|
|
|
|
struct btrfs_delayed_ref_head {
|
2017-09-29 22:43:57 +03:00
|
|
|
u64 bytenr;
|
|
|
|
u64 num_bytes;
|
|
|
|
refcount_t refs;
|
2009-03-13 17:10:06 +03:00
|
|
|
/*
|
|
|
|
* the mutex is held while running the refs, and it is also
|
|
|
|
* held when checking the sum of reference modifications.
|
|
|
|
*/
|
|
|
|
struct mutex mutex;
|
|
|
|
|
2014-01-23 18:21:38 +04:00
|
|
|
spinlock_t lock;
|
2018-08-22 22:51:50 +03:00
|
|
|
struct rb_root_cached ref_tree;
|
btrfs: improve delayed refs iterations
This issue was found when I tried to delete a heavily reflinked file,
when deleting such files, other transaction operation will not have a
chance to make progress, for example, start_transaction() will blocked
in wait_current_trans(root) for long time, sometimes it even triggers
soft lockups, and the time taken to delete such heavily reflinked file
is also very large, often hundreds of seconds. Using perf top, it reports
that:
PerfTop: 7416 irqs/sec kernel:99.8% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
---------------------------------------------------------------------------------------
84.37% [btrfs] [k] __btrfs_run_delayed_refs.constprop.80
11.02% [kernel] [k] delay_tsc
0.79% [kernel] [k] _raw_spin_unlock_irq
0.78% [kernel] [k] _raw_spin_unlock_irqrestore
0.45% [kernel] [k] do_raw_spin_lock
0.18% [kernel] [k] __slab_alloc
It seems __btrfs_run_delayed_refs() took most cpu time, after some debug
work, I found it's select_delayed_ref() causing this issue, for a delayed
head, in our case, it'll be full of BTRFS_DROP_DELAYED_REF nodes, but
select_delayed_ref() will firstly try to iterate node list to find
BTRFS_ADD_DELAYED_REF nodes, obviously it's a disaster in this case, and
waste much time.
To fix this issue, we introduce a new ref_add_list in struct btrfs_delayed_ref_head,
then in select_delayed_ref(), if this list is not empty, we can directly use
nodes in this list. With this patch, it just took about 10~15 seconds to
delte the same file. Now using perf top, it reports that:
PerfTop: 2734 irqs/sec kernel:99.5% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
----------------------------------------------------------------------------------------
20.74% [kernel] [k] _raw_spin_unlock_irqrestore
16.33% [kernel] [k] __slab_alloc
5.41% [kernel] [k] lock_acquired
4.42% [kernel] [k] lock_acquire
4.05% [kernel] [k] lock_release
3.37% [kernel] [k] _raw_spin_unlock_irq
For normal files, this patch also gives help, at least we do not need to
iterate whole list to found BTRFS_ADD_DELAYED_REF nodes.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2016-10-26 13:07:33 +03:00
|
|
|
/* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */
|
|
|
|
struct list_head ref_add_list;
|
2009-03-13 17:17:05 +03:00
|
|
|
|
2013-10-14 08:59:45 +04:00
|
|
|
struct rb_node href_node;
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
struct btrfs_delayed_extent_op *extent_op;
|
2015-02-03 18:50:16 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* This is used to track the final ref_mod from all the refs associated
|
|
|
|
* with this head ref, this is not adjusted as delayed refs are run,
|
|
|
|
* this is meant to track if we need to do the csum accounting or not.
|
|
|
|
*/
|
|
|
|
int total_ref_mod;
|
|
|
|
|
2017-09-29 22:43:57 +03:00
|
|
|
/*
|
|
|
|
* This is the current outstanding mod references for this bytenr. This
|
|
|
|
* is used with lookup_extent_info to get an accurate reference count
|
|
|
|
* for a bytenr, so it is adjusted as delayed refs are run so that any
|
|
|
|
* on disk reference count + ref_mod is accurate.
|
|
|
|
*/
|
|
|
|
int ref_mod;
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/*
|
|
|
|
* when a new extent is allocated, it is just reserved in memory
|
|
|
|
* The actual extent isn't inserted into the extent allocation tree
|
|
|
|
* until the delayed ref is processed. must_insert_reserved is
|
|
|
|
* used to flag a delayed ref so the accounting can be updated
|
|
|
|
* when a full insert is done.
|
|
|
|
*
|
|
|
|
* It is possible the extent will be freed before it is ever
|
|
|
|
* inserted into the extent allocation tree. In this case
|
|
|
|
* we need to update the in ram accounting to properly reflect
|
|
|
|
* the free has happened.
|
|
|
|
*/
|
|
|
|
unsigned int must_insert_reserved:1;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
unsigned int is_data:1;
|
2018-04-18 09:41:54 +03:00
|
|
|
unsigned int is_system:1;
|
2014-01-23 18:21:38 +04:00
|
|
|
unsigned int processing:1;
|
2009-03-13 17:10:06 +03:00
|
|
|
};
|
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
struct btrfs_delayed_tree_ref {
|
2009-03-13 17:10:06 +03:00
|
|
|
struct btrfs_delayed_ref_node node;
|
2011-09-14 16:01:24 +04:00
|
|
|
u64 root;
|
|
|
|
u64 parent;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
int level;
|
|
|
|
};
|
2009-03-13 17:10:06 +03:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
struct btrfs_delayed_data_ref {
|
|
|
|
struct btrfs_delayed_ref_node node;
|
2011-09-14 16:01:24 +04:00
|
|
|
u64 root;
|
|
|
|
u64 parent;
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
u64 objectid;
|
|
|
|
u64 offset;
|
2009-03-13 17:10:06 +03:00
|
|
|
};
|
|
|
|
|
btrfs: only let one thread pre-flush delayed refs in commit
I've been running a stress test that runs 20 workers in their own
subvolume, which are running an fsstress instance with 4 threads per
worker, which is 80 total fsstress threads. In addition to this I'm
running balance in the background as well as creating and deleting
snapshots. This test takes around 12 hours to run normally, going
slower and slower as the test goes on.
The reason for this is because fsstress is running fsync sometimes, and
because we're messing with block groups we often fall through to
btrfs_commit_transaction, so will often have 20-30 threads all calling
btrfs_commit_transaction at the same time.
These all get stuck contending on the extent tree while they try to run
delayed refs during the initial part of the commit.
This is suboptimal, really because the extent tree is a single point of
failure we only want one thread acting on that tree at once to reduce
lock contention.
Fix this by making the flushing mechanism a bit operation, to make it
easy to use test_and_set_bit() in order to make sure only one task does
this initial flush.
Once we're into the transaction commit we only have one thread doing
delayed ref running, it's just this initial pre-flush that is
problematic. With this patch my stress test takes around 90 minutes to
run, instead of 12 hours.
The memory barrier is not necessary for the flushing bit as it's
ordered, unlike plain int. The transaction state accessed in
btrfs_should_end_transaction could be affected by that too as it's not
always used under transaction lock. Upon Nikolay's analysis in [1]
it's not necessary:
In should_end_transaction it's read without holding any locks. (U)
It's modified in btrfs_cleanup_transaction without holding the
fs_info->trans_lock (U), but the STATE_ERROR flag is going to be set.
set in cleanup_transaction under fs_info->trans_lock (L)
set in btrfs_commit_trans to COMMIT_START under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_DOING under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_UNBLOCK under
fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_COMPLETED without locks but at this
point the transaction is finished and fs_info->running_trans is NULL (U
but irrelevant).
So by the looks of it we can have a concurrent READ race with a WRITE,
due to reads not taking a lock. In this case what we want to ensure is
we either see new or old state. I consulted with Will Deacon and he said
that in such a case we'd want to annotate the accesses to ->state with
(READ|WRITE)_ONCE so as to avoid a theoretical tear, in this case I
don't think this could happen but I imagine at some point KCSAN would
flag such an access as racy (which it is).
[1] https://lore.kernel.org/linux-btrfs/e1fd5cc1-0f28-f670-69f4-e9958b4964e6@suse.com
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ add comments regarding memory barrier ]
Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-18 22:24:20 +03:00
|
|
|
enum btrfs_delayed_ref_flags {
|
|
|
|
/* Indicate that we are flushing delayed refs for the commit */
|
|
|
|
BTRFS_DELAYED_REFS_FLUSHING,
|
|
|
|
};
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
struct btrfs_delayed_ref_root {
|
2013-10-14 08:59:45 +04:00
|
|
|
/* head ref rbtree */
|
2018-08-22 22:51:49 +03:00
|
|
|
struct rb_root_cached href_root;
|
2013-10-14 08:59:45 +04:00
|
|
|
|
2015-04-16 09:34:17 +03:00
|
|
|
/* dirty extent records */
|
|
|
|
struct rb_root dirty_extent_root;
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/* this spin lock protects the rbtree and the entries inside */
|
|
|
|
spinlock_t lock;
|
|
|
|
|
|
|
|
/* how many delayed ref updates we've queued, used by the
|
|
|
|
* throttling code
|
|
|
|
*/
|
2014-01-23 18:21:38 +04:00
|
|
|
atomic_t num_entries;
|
2009-03-13 17:10:06 +03:00
|
|
|
|
2009-03-13 17:17:05 +03:00
|
|
|
/* total number of head nodes in tree */
|
|
|
|
unsigned long num_heads;
|
|
|
|
|
|
|
|
/* total number of head nodes ready for processing */
|
|
|
|
unsigned long num_heads_ready;
|
|
|
|
|
2015-02-03 18:50:16 +03:00
|
|
|
u64 pending_csums;
|
|
|
|
|
btrfs: only let one thread pre-flush delayed refs in commit
I've been running a stress test that runs 20 workers in their own
subvolume, which are running an fsstress instance with 4 threads per
worker, which is 80 total fsstress threads. In addition to this I'm
running balance in the background as well as creating and deleting
snapshots. This test takes around 12 hours to run normally, going
slower and slower as the test goes on.
The reason for this is because fsstress is running fsync sometimes, and
because we're messing with block groups we often fall through to
btrfs_commit_transaction, so will often have 20-30 threads all calling
btrfs_commit_transaction at the same time.
These all get stuck contending on the extent tree while they try to run
delayed refs during the initial part of the commit.
This is suboptimal, really because the extent tree is a single point of
failure we only want one thread acting on that tree at once to reduce
lock contention.
Fix this by making the flushing mechanism a bit operation, to make it
easy to use test_and_set_bit() in order to make sure only one task does
this initial flush.
Once we're into the transaction commit we only have one thread doing
delayed ref running, it's just this initial pre-flush that is
problematic. With this patch my stress test takes around 90 minutes to
run, instead of 12 hours.
The memory barrier is not necessary for the flushing bit as it's
ordered, unlike plain int. The transaction state accessed in
btrfs_should_end_transaction could be affected by that too as it's not
always used under transaction lock. Upon Nikolay's analysis in [1]
it's not necessary:
In should_end_transaction it's read without holding any locks. (U)
It's modified in btrfs_cleanup_transaction without holding the
fs_info->trans_lock (U), but the STATE_ERROR flag is going to be set.
set in cleanup_transaction under fs_info->trans_lock (L)
set in btrfs_commit_trans to COMMIT_START under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_DOING under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_UNBLOCK under
fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_COMPLETED without locks but at this
point the transaction is finished and fs_info->running_trans is NULL (U
but irrelevant).
So by the looks of it we can have a concurrent READ race with a WRITE,
due to reads not taking a lock. In this case what we want to ensure is
we either see new or old state. I consulted with Will Deacon and he said
that in such a case we'd want to annotate the accesses to ->state with
(READ|WRITE)_ONCE so as to avoid a theoretical tear, in this case I
don't think this could happen but I imagine at some point KCSAN would
flag such an access as racy (which it is).
[1] https://lore.kernel.org/linux-btrfs/e1fd5cc1-0f28-f670-69f4-e9958b4964e6@suse.com
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ add comments regarding memory barrier ]
Signed-off-by: David Sterba <dsterba@suse.com>
2020-12-18 22:24:20 +03:00
|
|
|
unsigned long flags;
|
2009-03-13 17:17:05 +03:00
|
|
|
|
|
|
|
u64 run_delayed_start;
|
2015-04-20 04:53:50 +03:00
|
|
|
|
|
|
|
/*
|
|
|
|
* To make qgroup to skip given root.
|
2016-05-20 04:18:45 +03:00
|
|
|
* This is for snapshot, as btrfs_qgroup_inherit() will manually
|
2015-04-20 04:53:50 +03:00
|
|
|
* modify counters for snapshot and its source, so we should skip
|
|
|
|
* the snapshot in new_root/old_roots or it will get calculated twice
|
|
|
|
*/
|
|
|
|
u64 qgroup_to_skip;
|
2009-03-13 17:10:06 +03:00
|
|
|
};
|
|
|
|
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
enum btrfs_ref_type {
|
|
|
|
BTRFS_REF_NOT_SET,
|
|
|
|
BTRFS_REF_DATA,
|
|
|
|
BTRFS_REF_METADATA,
|
|
|
|
BTRFS_REF_LAST,
|
|
|
|
};
|
|
|
|
|
|
|
|
struct btrfs_data_ref {
|
|
|
|
/* For EXTENT_DATA_REF */
|
|
|
|
|
2021-10-12 11:21:33 +03:00
|
|
|
/* Original root this data extent belongs to */
|
|
|
|
u64 owning_root;
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
|
|
|
|
/* Inode which refers to this data extent */
|
|
|
|
u64 ino;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* file_offset - extent_offset
|
|
|
|
*
|
|
|
|
* file_offset is the key.offset of the EXTENT_DATA key.
|
|
|
|
* extent_offset is btrfs_file_extent_offset() of the EXTENT_DATA data.
|
|
|
|
*/
|
|
|
|
u64 offset;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct btrfs_tree_ref {
|
|
|
|
/*
|
|
|
|
* Level of this tree block
|
|
|
|
*
|
|
|
|
* Shared for skinny (TREE_BLOCK_REF) and normal tree ref.
|
|
|
|
*/
|
|
|
|
int level;
|
|
|
|
|
|
|
|
/*
|
2021-10-12 11:21:33 +03:00
|
|
|
* Root which owns this tree block.
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
*
|
|
|
|
* For TREE_BLOCK_REF (skinny metadata, either inline or keyed)
|
|
|
|
*/
|
2021-10-12 11:21:33 +03:00
|
|
|
u64 owning_root;
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
|
|
|
|
/* For non-skinny metadata, no special member needed */
|
|
|
|
};
|
|
|
|
|
|
|
|
struct btrfs_ref {
|
|
|
|
enum btrfs_ref_type type;
|
|
|
|
int action;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Whether this extent should go through qgroup record.
|
|
|
|
*
|
|
|
|
* Normally false, but for certain cases like delayed subtree scan,
|
|
|
|
* setting this flag can hugely reduce qgroup overhead.
|
|
|
|
*/
|
|
|
|
bool skip_qgroup;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Optional. For which root is this modification.
|
|
|
|
* Mostly used for qgroup optimization.
|
|
|
|
*
|
|
|
|
* When unset, data/tree ref init code will populate it.
|
|
|
|
* In certain cases, we're modifying reference for a different root.
|
|
|
|
* E.g. COW fs tree blocks for balance.
|
|
|
|
* In that case, tree_ref::root will be fs tree, but we're doing this
|
|
|
|
* for reloc tree, then we should set @real_root to reloc tree.
|
|
|
|
*/
|
|
|
|
u64 real_root;
|
|
|
|
u64 bytenr;
|
|
|
|
u64 len;
|
|
|
|
|
|
|
|
/* Bytenr of the parent tree block */
|
|
|
|
u64 parent;
|
|
|
|
union {
|
|
|
|
struct btrfs_data_ref data_ref;
|
|
|
|
struct btrfs_tree_ref tree_ref;
|
|
|
|
};
|
|
|
|
};
|
|
|
|
|
2012-11-21 06:21:28 +04:00
|
|
|
extern struct kmem_cache *btrfs_delayed_ref_head_cachep;
|
|
|
|
extern struct kmem_cache *btrfs_delayed_tree_ref_cachep;
|
|
|
|
extern struct kmem_cache *btrfs_delayed_data_ref_cachep;
|
|
|
|
extern struct kmem_cache *btrfs_delayed_extent_op_cachep;
|
|
|
|
|
2017-11-03 02:21:50 +03:00
|
|
|
int __init btrfs_delayed_ref_init(void);
|
2018-02-19 19:24:18 +03:00
|
|
|
void __cold btrfs_delayed_ref_exit(void);
|
2012-11-21 06:21:28 +04:00
|
|
|
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
static inline void btrfs_init_generic_ref(struct btrfs_ref *generic_ref,
|
|
|
|
int action, u64 bytenr, u64 len, u64 parent)
|
|
|
|
{
|
|
|
|
generic_ref->action = action;
|
|
|
|
generic_ref->bytenr = bytenr;
|
|
|
|
generic_ref->len = len;
|
|
|
|
generic_ref->parent = parent;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void btrfs_init_tree_ref(struct btrfs_ref *generic_ref,
|
|
|
|
int level, u64 root)
|
|
|
|
{
|
|
|
|
/* If @real_root not set, use @root as fallback */
|
|
|
|
if (!generic_ref->real_root)
|
|
|
|
generic_ref->real_root = root;
|
|
|
|
generic_ref->tree_ref.level = level;
|
2021-10-12 11:21:33 +03:00
|
|
|
generic_ref->tree_ref.owning_root = root;
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
generic_ref->type = BTRFS_REF_METADATA;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void btrfs_init_data_ref(struct btrfs_ref *generic_ref,
|
|
|
|
u64 ref_root, u64 ino, u64 offset)
|
|
|
|
{
|
|
|
|
/* If @real_root not set, use @root as fallback */
|
|
|
|
if (!generic_ref->real_root)
|
|
|
|
generic_ref->real_root = ref_root;
|
2021-10-12 11:21:33 +03:00
|
|
|
generic_ref->data_ref.owning_root = ref_root;
|
btrfs: delayed-ref: Introduce better documented delayed ref structures
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
2019-04-04 09:45:29 +03:00
|
|
|
generic_ref->data_ref.ino = ino;
|
|
|
|
generic_ref->data_ref.offset = offset;
|
|
|
|
generic_ref->type = BTRFS_REF_DATA;
|
|
|
|
}
|
|
|
|
|
2012-11-21 06:21:28 +04:00
|
|
|
static inline struct btrfs_delayed_extent_op *
|
|
|
|
btrfs_alloc_delayed_extent_op(void)
|
|
|
|
{
|
|
|
|
return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void
|
|
|
|
btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op)
|
|
|
|
{
|
|
|
|
if (op)
|
|
|
|
kmem_cache_free(btrfs_delayed_extent_op_cachep, op);
|
|
|
|
}
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
|
|
|
|
{
|
2017-03-03 11:55:15 +03:00
|
|
|
WARN_ON(refcount_read(&ref->refs) == 0);
|
|
|
|
if (refcount_dec_and_test(&ref->refs)) {
|
2009-03-13 17:10:06 +03:00
|
|
|
WARN_ON(ref->in_tree);
|
2012-11-21 06:21:28 +04:00
|
|
|
switch (ref->type) {
|
|
|
|
case BTRFS_TREE_BLOCK_REF_KEY:
|
|
|
|
case BTRFS_SHARED_BLOCK_REF_KEY:
|
|
|
|
kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref);
|
|
|
|
break;
|
|
|
|
case BTRFS_EXTENT_DATA_REF_KEY:
|
|
|
|
case BTRFS_SHARED_DATA_REF_KEY:
|
|
|
|
kmem_cache_free(btrfs_delayed_data_ref_cachep, ref);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
BUG();
|
|
|
|
}
|
2009-03-13 17:10:06 +03:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-01-16 00:48:55 +03:00
|
|
|
static inline u64 btrfs_ref_head_to_space_flags(
|
|
|
|
struct btrfs_delayed_ref_head *head_ref)
|
|
|
|
{
|
|
|
|
if (head_ref->is_data)
|
|
|
|
return BTRFS_BLOCK_GROUP_DATA;
|
|
|
|
else if (head_ref->is_system)
|
|
|
|
return BTRFS_BLOCK_GROUP_SYSTEM;
|
|
|
|
return BTRFS_BLOCK_GROUP_METADATA;
|
|
|
|
}
|
|
|
|
|
2017-09-29 22:43:57 +03:00
|
|
|
static inline void btrfs_put_delayed_ref_head(struct btrfs_delayed_ref_head *head)
|
|
|
|
{
|
|
|
|
if (refcount_dec_and_test(&head->refs))
|
|
|
|
kmem_cache_free(btrfs_delayed_ref_head_cachep, head);
|
|
|
|
}
|
|
|
|
|
2018-06-20 15:48:53 +03:00
|
|
|
int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
|
2019-04-04 09:45:31 +03:00
|
|
|
struct btrfs_ref *generic_ref,
|
2021-01-16 00:48:55 +03:00
|
|
|
struct btrfs_delayed_extent_op *extent_op);
|
2018-06-20 15:48:54 +03:00
|
|
|
int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
|
2019-04-04 09:45:32 +03:00
|
|
|
struct btrfs_ref *generic_ref,
|
2021-01-16 00:48:55 +03:00
|
|
|
u64 reserved);
|
2019-03-20 13:42:34 +03:00
|
|
|
int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
u64 bytenr, u64 num_bytes,
|
|
|
|
struct btrfs_delayed_extent_op *extent_op);
|
2012-08-08 00:00:32 +04:00
|
|
|
void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans,
|
|
|
|
struct btrfs_delayed_ref_root *delayed_refs,
|
|
|
|
struct btrfs_delayed_ref_head *head);
|
2009-03-13 17:10:06 +03:00
|
|
|
|
2009-03-13 17:11:24 +03:00
|
|
|
struct btrfs_delayed_ref_head *
|
2017-01-30 23:24:37 +03:00
|
|
|
btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
|
|
|
|
u64 bytenr);
|
2018-10-11 08:40:34 +03:00
|
|
|
int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs,
|
2009-03-13 17:17:05 +03:00
|
|
|
struct btrfs_delayed_ref_head *head);
|
2012-12-19 12:10:10 +04:00
|
|
|
static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head)
|
|
|
|
{
|
|
|
|
mutex_unlock(&head->mutex);
|
|
|
|
}
|
2018-12-03 18:20:29 +03:00
|
|
|
void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
|
|
|
|
struct btrfs_delayed_ref_head *head);
|
2014-01-23 18:21:38 +04:00
|
|
|
|
2018-10-11 08:40:33 +03:00
|
|
|
struct btrfs_delayed_ref_head *btrfs_select_ref_head(
|
|
|
|
struct btrfs_delayed_ref_root *delayed_refs);
|
2011-09-14 14:37:00 +04:00
|
|
|
|
2018-04-04 15:57:42 +03:00
|
|
|
int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq);
|
2011-09-14 14:37:00 +04:00
|
|
|
|
2019-06-19 22:11:58 +03:00
|
|
|
void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
|
|
|
|
void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
|
|
|
|
int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
|
|
|
|
enum btrfs_reserve_flush_enum flush);
|
|
|
|
void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
|
|
|
|
struct btrfs_block_rsv *src,
|
|
|
|
u64 num_bytes);
|
|
|
|
int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans);
|
|
|
|
bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info);
|
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
/*
|
|
|
|
* helper functions to cast a node into its container
|
|
|
|
*/
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
static inline struct btrfs_delayed_tree_ref *
|
|
|
|
btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node)
|
2009-03-13 17:10:06 +03:00
|
|
|
{
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
return container_of(node, struct btrfs_delayed_tree_ref, node);
|
|
|
|
}
|
2009-03-13 17:10:06 +03:00
|
|
|
|
Btrfs: Mixed back reference (FORWARD ROLLING FORMAT CHANGE)
This commit introduces a new kind of back reference for btrfs metadata.
Once a filesystem has been mounted with this commit, IT WILL NO LONGER
BE MOUNTABLE BY OLDER KERNELS.
When a tree block in subvolume tree is cow'd, the reference counts of all
extents it points to are increased by one. At transaction commit time,
the old root of the subvolume is recorded in a "dead root" data structure,
and the btree it points to is later walked, dropping reference counts
and freeing any blocks where the reference count goes to 0.
The increments done during cow and decrements done after commit cancel out,
and the walk is a very expensive way to go about freeing the blocks that
are no longer referenced by the new btree root. This commit reduces the
transaction overhead by avoiding the need for dead root records.
When a non-shared tree block is cow'd, we free the old block at once, and the
new block inherits old block's references. When a tree block with reference
count > 1 is cow'd, we increase the reference counts of all extents
the new block points to by one, and decrease the old block's reference count by
one.
This dead tree avoidance code removes the need to modify the reference
counts of lower level extents when a non-shared tree block is cow'd.
But we still need to update back ref for all pointers in the block.
This is because the location of the block is recorded in the back ref
item.
We can solve this by introducing a new type of back ref. The new
back ref provides information about pointer's key, level and in which
tree the pointer lives. This information allow us to find the pointer
by searching the tree. The shortcoming of the new back ref is that it
only works for pointers in tree blocks referenced by their owner trees.
This is mostly a problem for snapshots, where resolving one of these
fuzzy back references would be O(number_of_snapshots) and quite slow.
The solution used here is to use the fuzzy back references in the common
case where a given tree block is only referenced by one root,
and use the full back references when multiple roots have a reference
on a given block.
This commit adds per subvolume red-black tree to keep trace of cached
inodes. The red-black tree helps the balancing code to find cached
inodes whose inode numbers within a given range.
This commit improves the balancing code by introducing several data
structures to keep the state of balancing. The most important one
is the back ref cache. It caches how the upper level tree blocks are
referenced. This greatly reduce the overhead of checking back ref.
The improved balancing code scales significantly better with a large
number of snapshots.
This is a very large commit and was written in a number of
pieces. But, they depend heavily on the disk format change and were
squashed together to make sure git bisect didn't end up in a
bad state wrt space balancing or the format change.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
2009-06-10 18:45:14 +04:00
|
|
|
static inline struct btrfs_delayed_data_ref *
|
|
|
|
btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node)
|
|
|
|
{
|
|
|
|
return container_of(node, struct btrfs_delayed_data_ref, node);
|
2009-03-13 17:10:06 +03:00
|
|
|
}
|
2018-04-03 20:16:55 +03:00
|
|
|
|
2009-03-13 17:10:06 +03:00
|
|
|
#endif
|