Patchwork [RFC,02/10,v1,RESEND] ext4: add operations on extent status tree

login
register
mail settings
Submitter Zheng Liu
Date July 22, 2012, 7:59 a.m.
Message ID <1342943986-12413-3-git-send-email-wenqing.lz@taobao.com>
Download mbox | patch
Permalink /patch/172493/
State Superseded
Headers show

Comments

Zheng Liu - July 22, 2012, 7:59 a.m.
From: Zheng Liu <wenqing.lz@taobao.com>

This patch adds operations on a extent status tree.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
---
 fs/ext4/Makefile         |    2 +-
 fs/ext4/extents_status.c |  418 ++++++++++++++++++++++++++++++++++++++++++++++
 fs/ext4/extents_status.h |   17 ++
 3 files changed, 436 insertions(+), 1 deletions(-)
 create mode 100644 fs/ext4/extents_status.c
Lukas Czerner - July 31, 2012, 11:55 a.m.
On Sun, 22 Jul 2012, Zheng Liu wrote:

> Date: Sun, 22 Jul 2012 15:59:38 +0800
> From: Zheng Liu <gnehzuil.liu@gmail.com>
> To: linux-ext4@vger.kernel.org, linux-fsdevel@vger.kernel.org
> Cc: xiaoqiangnk@gmail.com, achender@linux.vnet.ibm.com, wenqing.lz@taobao.com
> Subject: [RFC][PATCH 02/10 v1][RESEND] ext4: add operations on extent status
>     tree
> 
> From: Zheng Liu <wenqing.lz@taobao.com>
> 
> This patch adds operations on a extent status tree.
> 
> Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
> Signed-off-by: Allison Henderson <achender@linux.vnet.ibm.com>
> Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
> ---
>  fs/ext4/Makefile         |    2 +-
>  fs/ext4/extents_status.c |  418 ++++++++++++++++++++++++++++++++++++++++++++++
>  fs/ext4/extents_status.h |   17 ++
>  3 files changed, 436 insertions(+), 1 deletions(-)
>  create mode 100644 fs/ext4/extents_status.c
> 
> diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile
> index 56fd8f86..41f22be 100644
> --- a/fs/ext4/Makefile
> +++ b/fs/ext4/Makefile
> @@ -7,7 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
>  ext4-y	:= balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
>  		ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
>  		ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
> -		mmp.o indirect.o
> +		mmp.o indirect.o extents_status.o
>  
>  ext4-$(CONFIG_EXT4_FS_XATTR)		+= xattr.o xattr_user.o xattr_trusted.o
>  ext4-$(CONFIG_EXT4_FS_POSIX_ACL)	+= acl.o
> diff --git a/fs/ext4/extents_status.c b/fs/ext4/extents_status.c
> new file mode 100644
> index 0000000..bd4e589
> --- /dev/null
> +++ b/fs/ext4/extents_status.c
> @@ -0,0 +1,418 @@
> +/*
> + *  fs/ext4/extents_status.c
> + *
> + * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
> + *
> + * Ext4 extents status tree core functions.
> + */
> +#include <linux/rbtree.h>
> +#include "ext4.h"
> +#include "extents_status.h"
> +#include "ext4_extents.h"
> +
> +/*
> + * extents status tree implementation for ext4.
> + *
> + *
> + * ==========================================================================
> + * Extents status encompass delayed extents and extent locks
> + *
> + * 1. Why delayed extent implementation ?
> + *
> + * Without delayed extent, ext4 identifies a delayed extent by looking up
> + * page cache, this has several deficiencies - complicated, buggy, and
> + * inefficient code.
> + *
> + * FIEMAP, SEEK_HOLE/DATA, bigalloc, punch hole and writeout all need to know if
> + * a block or a range of blocks are belonged to a delayed extent.
> + *
> + * Let us have a look at how they do without delayed extents implementation.
> + *   --	FIEMAP
> + *	FIEMAP looks up page cache to identify delayed allocations from holes.
> + *
> + *   --	SEEK_HOLE/DATA
> + *	SEEK_HOLE/DATA has the same problem as FIEMAP.
> + *
> + *   --	bigalloc
> + *	bigalloc looks up page cache to figure out if a block is already
> + *	under delayed allocation or not to determine whether quota reserving
> + *	is needed for the cluster.
> + *
> + *   -- punch hole
> + *	punch hole looks up page cache to identify a delayed extent.
> + *
> + *   --	writeout
> + *	Writeout looks up whole page cache to see if a buffer is mapped, If
> + *	there are not very many delayed buffers, then it is time comsuming.
> + *
> + * With delayed extents implementation, FIEMAP, SEEK_HOLE/DATA, bigalloc and
> + * writeout can figure out if a block or a range of blocks is under delayed
> + * allocation(belonged to a delayed extent) or not by searching the delayed
> + * extent tree.
> + *
> + *
> + * ==========================================================================
> + * 2. ext4 delayed extents impelmentation
> + *
> + *   --	delayed extent
> + *	A delayed extent is a range of blocks which are contiguous logically and
> + *	under delayed allocation.  Unlike extent in ext4, delayed extent in ext4
> + *	is a in-memory struct, there is no corresponding on-disk data.  There is
> + *	no limit on length of delayed extent, so a delayed extent can contain as
> + *	many blocks as they are contiguous logically.
> + *
> + *   --	delayed extent tree
> + *	Every inode has a delayed extent tree and all under delayed allocation
> + *	blocks are added to the tree as dealyed extents.  Delayed extents in
> + *	the tree are ordered by logical block no.
> + *
> + *   --	operations on a delayed extent tree
> + *	There are three operations on a delayed extent tree: find next delayed
> + *	extent, adding a space(a range of blocks) and removing a space.
> + *
> + *   --	race on a delayed extent tree
> + *	Delayed extent tree is protected inode->i_data_sem like extent tree.
> + *
> + *
> + * ==========================================================================
> + * 3. performance analysis
> + *   --	overhead
> + *	1. Apart from operations on a delayed extent tree, we need to
> + *	down_write(inode->i_data_sem) in delayed write path to maintain delayed
> + *	extent tree, this can have impact on parallel read-write and write-write
> + *
> + *	2. There is a cache extent for write access, so if writes are not very
> + *	random, adding space operaions are in O(1) time.
> + *
> + *   --	gain
> + *	3. Code is much simpler, more readable, more maintainable and
> + *      more efficient.
> + */
> +
> +static struct kmem_cache *ext4_es_cachep;
> +
> +int __init ext4_init_es(void)
> +{
> +	ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
> +	if (ext4_es_cachep == NULL)
> +		return -ENOMEM;
> +	return 0;
> +}
> +
> +void ext4_exit_es(void)
> +{
> +	if (ext4_es_cachep)
> +		kmem_cache_destroy(ext4_es_cachep);
> +}
> +
> +void ext4_es_init_tree(struct ext4_es_tree *tree)
> +{
> +	tree->root = RB_ROOT;
> +	tree->cache_es = NULL;
> +}
> +
> +#ifdef SE_DEBUG
> +static void ext4_es_print_tree(struct inode *inode)
> +{
> +	struct ext4_es_tree *tree;
> +	struct rb_node *node;
> +
> +	printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
> +	tree = &EXT4_I(inode)->i_es_tree;
> +	node = rb_first(&tree->root);
> +	while (node) {
> +		struct extent_status *es;
> +		es = rb_entry(node, struct extent_status, rb_node);
> +		printk(KERN_DEBUG " [%u/%u)", es->start, es->len);
> +		node = rb_next(node);
> +	}
> +	printk(KERN_DEBUG "\n");
> +}
> +#else
> +#define ext4_es_print_tree(inode)
> +#endif
> +
> +static inline ext4_lblk_t extent_status_end(struct extent_status *es)
> +{
> +	if (es->start + es->len < es->start)
> +		return (ext4_lblk_t)-1;

Is this possible to happen ? It seems to me that if so, it means
that the extent is corrupted and we should do something about this
rather than ignore the problem by returning -1 (callers does not
actually check for it), at least warning should be in place.

> +	return es->start + es->len;

This seems like it should rather be es->start + es->len - 1, since
the function name suggests that we should get the last block of the
extent. Having it this way seems error prone to the off by one
errors.

> +}
> +
> +/*
> + * search through the tree for an delayed_extent with a given offset.  If
> + * it can't be found, try to find next extent.

This helper could be used by any other extent tree in the ext4 not
just for delayed extents right (if we're going to add some in the
future)? So we should change the comment, because it is generic
enough.

> + */
> +static struct extent_status *__es_tree_search(struct rb_root *root,
> +						ext4_lblk_t offset)
> +{
> +	struct rb_node *node = root->rb_node;
> +	struct extent_status *es = NULL;
> +
> +	while (node) {
> +		es = rb_entry(node, struct extent_status, rb_node);
> +		if (offset < es->start)
> +			node = node->rb_left;
> +		else if (offset >= extent_status_end(es))
> +			node = node->rb_right;
> +		else
> +			return es;
> +	}
> +
> +	if (es && offset < es->start)
> +		return es;
> +
> +	if (es && offset >= extent_status_end(es)) {
> +		node = rb_next(&es->rb_node);
> +		return node ? rb_entry(node, struct extent_status, rb_node) :
> +			      NULL;
> +	}
> +
> +	return NULL;
> +}
> +
> +/*
> + * ext4_es_find_extent: find the 1st delayed extent covering @es->start
> + * if it exists, otherwise, the next extent after @es->start.
> + *
> + * @inode: the inode which owns delayed extents
> + * @es: delayed extent that we found
> + *
> + * Returns next block beyond the found extent.

This is not exactly right isn't it ? From what I can see it returns
the first block of the next extent after the one we're returning in
es, or EXT_MAX_BLOCKS if no extent is found.

> + * Delayed extent is returned via @es.
> + */
> +ext4_lblk_t ext4_es_find_extent(struct inode *inode, struct extent_status *es)
> +{
> +	struct ext4_es_tree *tree;
> +	struct extent_status *es1;
> +	struct rb_node *node;
> +	ext4_lblk_t ret = EXT_MAX_BLOCKS;

Would not it make sense to try the cache first ? Maybe we do not
need to search the tree at all ?

> +
> +	es->len = 0;
> +	tree = &EXT4_I(inode)->i_es_tree;
> +	es1 = __es_tree_search(&tree->root, es->start);
> +	if (es1) {
> +		tree->cache_es = es1;
> +		es->start = es1->start;
> +		es->len = es1->len;
> +		node = rb_next(&es1->rb_node);
> +		if (node) {
> +			es1 = rb_entry(node, struct extent_status, rb_node);
> +			ret = es1->start;
> +		}
> +	}
> +
> +	return ret;
> +}
> +
> +static struct extent_status *
> +ext4_es_alloc_extent(ext4_lblk_t start, ext4_lblk_t len)
> +{
> +	struct extent_status *es;
> +	es = kmem_cache_alloc(ext4_es_cachep, GFP_NOFS);
> +	if (es == NULL)
> +		return NULL;
> +	es->start = start;
> +	es->len = len;
> +	return es;
> +}
> +
> +static void ext4_es_free_extent(struct extent_status *es)
> +{
> +	kmem_cache_free(ext4_es_cachep, es);
> +}
> +
> +static void ext4_es_try_to_merge_left(struct ext4_es_tree *tree,
> +				      struct extent_status *es)
> +{
> +	struct extent_status *es1;
> +	struct rb_node *node;
> +
> +	node = rb_prev(&es->rb_node);
> +	if (!node)
> +		return;
> +
> +	es1 = rb_entry(node, struct extent_status, rb_node);
> +	if (extent_status_end(es1) == es->start) {
> +		es1->len += es->len;
> +		rb_erase(&es->rb_node, &tree->root);
> +		if (es == tree->cache_es)
> +			tree->cache_es = es1;
> +		ext4_es_free_extent(es);
> +	}
> +}
> +
> +static void ext4_es_try_to_merge_right(struct ext4_es_tree *tree,
> +				       struct extent_status *es)
> +{
> +	struct extent_status *es1;
> +	struct rb_node *node;
> +
> +	node = rb_next(&es->rb_node);
> +	if (!node)
> +		return;
> +
> +	es1 = rb_entry(node, struct extent_status, rb_node);
> +	if (es1->start == extent_status_end(es)) {
> +		es->len += es1->len;
> +		rb_erase(node, &tree->root);
> +		if (es1 == tree->cache_es)
> +			tree->cache_es = es;
> +		ext4_es_free_extent(es1);
> +	}
> +}
> +
> +/*
> + * ext4_es_add_space: adds a space to a delayed extent tree.
> + * Caller holds inode->i_data_sem.
> + *
> + * ext4_es_add_space is callyed by ext4_dealyed_write_begin and
> + * ext4_es_remove_space.
> + *
> + * Return 0 on success, error code on failure.
> + */
> +int ext4_es_add_space(struct inode *inode, ext4_lblk_t offset, ext4_lblk_t len)

I would rather call the function ext4_es_add_extent(), but that may
be subjective.

> +{
> +	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
> +	struct rb_node **p = &tree->root.rb_node;
> +	struct rb_node *parent = NULL;
> +	struct extent_status *es;
> +	ext4_lblk_t end = offset + len;
> +
> +	BUG_ON(end <= offset);
> +
> +	es = tree->cache_es;
> +	es_debug("add [%u/%u) to extent status tree of inode %lu\n",
> +		 offset, len, inode->i_ino);
> +
> +	if (es && extent_status_end(es) == offset) {
> +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> +		es->len += len;
> +		ext4_es_try_to_merge_right(tree, es);
> +		goto out;
> +	} else if (es && es->start == end) {
> +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> +		es->start = offset;
> +		es->len += len;
> +		ext4_es_try_to_merge_left(tree, es);
> +		goto out;
> +	} else if (es && es->start <= offset &&
> +		   extent_status_end(es) >= end) {
> +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> +		goto out;
> +	}
> +
> +	while (*p) {
> +		parent = *p;
> +		es = rb_entry(parent, struct extent_status, rb_node);
> +
> +		if (offset < es->start) {
> +			if (end == es->start) {
> +				es->len += len;
> +				es->start = offset;
> +				goto out;
> +			}
> +			p = &(*p)->rb_left;
> +		} else if (offset >= extent_status_end(es)) {
> +			if (extent_status_end(es) == offset) {
> +				es->len += len;
> +				goto out;
> +			}
> +			p = &(*p)->rb_right;
> +		} else
> +			goto out;
> +	}
> +
> +	es = ext4_es_alloc_extent(offset, len);
> +	if (!es)
> +		return -ENOMEM;
> +	rb_link_node(&es->rb_node, parent, p);
> +	rb_insert_color(&es->rb_node, &tree->root);
> +
> +out:
> +	tree->cache_es = es;
> +	ext4_es_print_tree(inode);
> +
> +	return 0;
> +}
> +
> +/*
> + * ext4_es_remove_space() removes a space from a delayed extent tree.
> + * Caller holds inode->i_data_sem.
> + *
> + * Return 0 on success, error code on failure.
> + */
> +int ext4_es_remove_space(struct inode *inode, ext4_lblk_t offset,
> +			 ext4_lblk_t len)

Again, my subjective opinion is that it could be rather called
ext4_es_remove_extent().

> +{
> +	struct rb_node *node;
> +	struct ext4_es_tree *tree;
> +	struct extent_status *es;
> +	struct extent_status orig_es;
> +	ext4_lblk_t len1, len2, end;
> +
> +	es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
> +		 offset, len, inode->i_ino);

Since you're adding tracepoints later on, are those debug messages
necessary ?

> +
> +	end = offset + len;
> +	BUG_ON(end <= offset);
> +	tree = &EXT4_I(inode)->i_es_tree;
> +	es = __es_tree_search(&tree->root, offset);
> +	if (!es)
> +		goto out;
> +
> +	/* Simply invalidate cache_es. */
> +	tree->cache_es = NULL;
> +
> +	orig_es.start = es->start;
> +	orig_es.len = es->len;
> +	len1 = offset > es->start ? offset - es->start : 0;
> +	len2 = extent_status_end(es) > end ?
> +	       extent_status_end(es) - end : 0;
> +	if (len1 > 0)
> +		es->len = len1;
> +	if (len2 > 0) {
> +		if (len1 > 0) {
> +			int err;
> +			err = ext4_es_add_space(inode, end, len2);
> +			if (err) {
> +				es->start = orig_es.start;
> +				es->len = orig_es.len;
> +				return err;
> +			}
> +		} else {
> +			es->start = end;
> +			es->len = len2;
> +		}
> +		goto out;
> +	}
> +
> +	if (len1 > 0) {
> +		node = rb_next(&es->rb_node);
> +		if (!node)
> +			es = rb_entry(node, struct extent_status, rb_node);
> +		else
> +			es = NULL;
> +	}
> +
> +	while (es && extent_status_end(es) <= end) {
> +		node = rb_next(&es->rb_node);
> +		rb_erase(&es->rb_node, &tree->root);
> +		ext4_es_free_extent(es);
> +		if (!node) {
> +			es = NULL;
> +			break;
> +		}
> +		es = rb_entry(node, struct extent_status, rb_node);
> +	}
> +
> +	if (es && es->start < end) {
> +		len1 = extent_status_end(es) - end;
> +		es->start = end;
> +		es->len = len1;
> +	}
> +
> +out:
> +	ext4_es_print_tree(inode);
> +	return 0;
> +}
> diff --git a/fs/ext4/extents_status.h b/fs/ext4/extents_status.h
> index d87fd71..8fe8084 100644
> --- a/fs/ext4/extents_status.h
> +++ b/fs/ext4/extents_status.h
> @@ -8,6 +8,12 @@
>  #ifndef _EXT4_EXTENTS_STATUS_H
>  #define _EXT4_EXTENTS_STATUS_H
>  
> +#ifdef ES_DEBUG
> +#define es_debug(a...)         printk(a)

Please take a look at the ext4_debug() or ext_debug(), maybe you can
use it, or at least reuse the code from it.

> +#else
> +#define es_debug(a...)
> +#endif
> +
>  struct extent_status {
>  	struct rb_node rb_node;
>  	ext4_lblk_t start;	/* first block extent covers */
> @@ -19,4 +25,15 @@ struct ext4_es_tree {
>  	struct extent_status *cache_es;	/* recently accessed extent */
>  };
>  
> +extern int __init ext4_init_es(void);
> +extern void ext4_exit_es(void);
> +extern void ext4_es_init_tree(struct ext4_es_tree *tree);
> +
> +extern int ext4_es_add_space(struct inode *inode, ext4_lblk_t start,
> +				ext4_lblk_t len);
> +extern int ext4_es_remove_space(struct inode *inode, ext4_lblk_t start,
> +				ext4_lblk_t len);
> +extern ext4_lblk_t ext4_es_find_extent(struct inode *inode,
> +				struct extent_status *es);
> +
>  #endif /* _EXT4_EXTENTS_STATUS_H */
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Zheng Liu - July 31, 2012, 1:18 p.m.
Hi Lukas,

Thanks for a comprehensive review. :-)

On Tue, Jul 31, 2012 at 01:55:14PM +0200, Lukáš Czerner wrote:
[cut...]
> > +static inline ext4_lblk_t extent_status_end(struct extent_status *es)
> > +{
> > +	if (es->start + es->len < es->start)
> > +		return (ext4_lblk_t)-1;
> 
> Is this possible to happen ? It seems to me that if so, it means
> that the extent is corrupted and we should do something about this
> rather than ignore the problem by returning -1 (callers does not
> actually check for it), at least warning should be in place.

Yes, ext4_warning() will be added to tell user that we meet a problem.

> 
> > +	return es->start + es->len;
> 
> This seems like it should rather be es->start + es->len - 1, since
> the function name suggests that we should get the last block of the
> extent. Having it this way seems error prone to the off by one
> errors.

Will fix it.

> 
> > +}
> > +
> > +/*
> > + * search through the tree for an delayed_extent with a given offset.  If
> > + * it can't be found, try to find next extent.
> 
> This helper could be used by any other extent tree in the ext4 not
> just for delayed extents right (if we're going to add some in the
> future)? So we should change the comment, because it is generic
> enough.

Until now it seems that this function only is used by io tree itself.
As Ted described before, there is an implementation of big extent tree
in Google, and this function might be used by it.  But firstly we will
implement these two features separately.  Certainly I will change the
comment if necessary.

> 
> > + */
> > +static struct extent_status *__es_tree_search(struct rb_root *root,
> > +						ext4_lblk_t offset)
> > +{
> > +	struct rb_node *node = root->rb_node;
> > +	struct extent_status *es = NULL;
> > +
> > +	while (node) {
> > +		es = rb_entry(node, struct extent_status, rb_node);
> > +		if (offset < es->start)
> > +			node = node->rb_left;
> > +		else if (offset >= extent_status_end(es))
> > +			node = node->rb_right;
> > +		else
> > +			return es;
> > +	}
> > +
> > +	if (es && offset < es->start)
> > +		return es;
> > +
> > +	if (es && offset >= extent_status_end(es)) {
> > +		node = rb_next(&es->rb_node);
> > +		return node ? rb_entry(node, struct extent_status, rb_node) :
> > +			      NULL;
> > +	}
> > +
> > +	return NULL;
> > +}
> > +
> > +/*
> > + * ext4_es_find_extent: find the 1st delayed extent covering @es->start
> > + * if it exists, otherwise, the next extent after @es->start.
> > + *
> > + * @inode: the inode which owns delayed extents
> > + * @es: delayed extent that we found
> > + *
> > + * Returns next block beyond the found extent.
> 
> This is not exactly right isn't it ? From what I can see it returns
> the first block of the next extent after the one we're returning in
> es, or EXT_MAX_BLOCKS if no extent is found.

Will fix it.

> 
> > + * Delayed extent is returned via @es.
> > + */
> > +ext4_lblk_t ext4_es_find_extent(struct inode *inode, struct extent_status *es)
> > +{
> > +	struct ext4_es_tree *tree;
> > +	struct extent_status *es1;
> > +	struct rb_node *node;
> > +	ext4_lblk_t ret = EXT_MAX_BLOCKS;
> 
> Would not it make sense to try the cache first ? Maybe we do not
> need to search the tree at all ?

Yes, you are right.  I will add this stuff in next version.

> 
> > +
> > +	es->len = 0;
> > +	tree = &EXT4_I(inode)->i_es_tree;
> > +	es1 = __es_tree_search(&tree->root, es->start);
> > +	if (es1) {
> > +		tree->cache_es = es1;
> > +		es->start = es1->start;
> > +		es->len = es1->len;
> > +		node = rb_next(&es1->rb_node);
> > +		if (node) {
> > +			es1 = rb_entry(node, struct extent_status, rb_node);
> > +			ret = es1->start;
> > +		}
> > +	}
> > +
> > +	return ret;
> > +}
> > +
> > +static struct extent_status *
> > +ext4_es_alloc_extent(ext4_lblk_t start, ext4_lblk_t len)
> > +{
> > +	struct extent_status *es;
> > +	es = kmem_cache_alloc(ext4_es_cachep, GFP_NOFS);
> > +	if (es == NULL)
> > +		return NULL;
> > +	es->start = start;
> > +	es->len = len;
> > +	return es;
> > +}
> > +
> > +static void ext4_es_free_extent(struct extent_status *es)
> > +{
> > +	kmem_cache_free(ext4_es_cachep, es);
> > +}
> > +
> > +static void ext4_es_try_to_merge_left(struct ext4_es_tree *tree,
> > +				      struct extent_status *es)
> > +{
> > +	struct extent_status *es1;
> > +	struct rb_node *node;
> > +
> > +	node = rb_prev(&es->rb_node);
> > +	if (!node)
> > +		return;
> > +
> > +	es1 = rb_entry(node, struct extent_status, rb_node);
> > +	if (extent_status_end(es1) == es->start) {
> > +		es1->len += es->len;
> > +		rb_erase(&es->rb_node, &tree->root);
> > +		if (es == tree->cache_es)
> > +			tree->cache_es = es1;
> > +		ext4_es_free_extent(es);
> > +	}
> > +}
> > +
> > +static void ext4_es_try_to_merge_right(struct ext4_es_tree *tree,
> > +				       struct extent_status *es)
> > +{
> > +	struct extent_status *es1;
> > +	struct rb_node *node;
> > +
> > +	node = rb_next(&es->rb_node);
> > +	if (!node)
> > +		return;
> > +
> > +	es1 = rb_entry(node, struct extent_status, rb_node);
> > +	if (es1->start == extent_status_end(es)) {
> > +		es->len += es1->len;
> > +		rb_erase(node, &tree->root);
> > +		if (es1 == tree->cache_es)
> > +			tree->cache_es = es;
> > +		ext4_es_free_extent(es1);
> > +	}
> > +}
> > +
> > +/*
> > + * ext4_es_add_space: adds a space to a delayed extent tree.
> > + * Caller holds inode->i_data_sem.
> > + *
> > + * ext4_es_add_space is callyed by ext4_dealyed_write_begin and
> > + * ext4_es_remove_space.
> > + *
> > + * Return 0 on success, error code on failure.
> > + */
> > +int ext4_es_add_space(struct inode *inode, ext4_lblk_t offset, ext4_lblk_t len)
> 
> I would rather call the function ext4_es_add_extent(), but that may
> be subjective.

IMHO ext4_es_insert_extent() might be a better name because it can be
consistent with ext4_ext_insert_extent().

> 
> > +{
> > +	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
> > +	struct rb_node **p = &tree->root.rb_node;
> > +	struct rb_node *parent = NULL;
> > +	struct extent_status *es;
> > +	ext4_lblk_t end = offset + len;
> > +
> > +	BUG_ON(end <= offset);
> > +
> > +	es = tree->cache_es;
> > +	es_debug("add [%u/%u) to extent status tree of inode %lu\n",
> > +		 offset, len, inode->i_ino);
> > +
> > +	if (es && extent_status_end(es) == offset) {
> > +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> > +		es->len += len;
> > +		ext4_es_try_to_merge_right(tree, es);
> > +		goto out;
> > +	} else if (es && es->start == end) {
> > +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> > +		es->start = offset;
> > +		es->len += len;
> > +		ext4_es_try_to_merge_left(tree, es);
> > +		goto out;
> > +	} else if (es && es->start <= offset &&
> > +		   extent_status_end(es) >= end) {
> > +		es_debug("cached by [%u/%u)\n", es->start, es->len);
> > +		goto out;
> > +	}
> > +
> > +	while (*p) {
> > +		parent = *p;
> > +		es = rb_entry(parent, struct extent_status, rb_node);
> > +
> > +		if (offset < es->start) {
> > +			if (end == es->start) {
> > +				es->len += len;
> > +				es->start = offset;
> > +				goto out;
> > +			}
> > +			p = &(*p)->rb_left;
> > +		} else if (offset >= extent_status_end(es)) {
> > +			if (extent_status_end(es) == offset) {
> > +				es->len += len;
> > +				goto out;
> > +			}
> > +			p = &(*p)->rb_right;
> > +		} else
> > +			goto out;
> > +	}
> > +
> > +	es = ext4_es_alloc_extent(offset, len);
> > +	if (!es)
> > +		return -ENOMEM;
> > +	rb_link_node(&es->rb_node, parent, p);
> > +	rb_insert_color(&es->rb_node, &tree->root);
> > +
> > +out:
> > +	tree->cache_es = es;
> > +	ext4_es_print_tree(inode);
> > +
> > +	return 0;
> > +}
> > +
> > +/*
> > + * ext4_es_remove_space() removes a space from a delayed extent tree.
> > + * Caller holds inode->i_data_sem.
> > + *
> > + * Return 0 on success, error code on failure.
> > + */
> > +int ext4_es_remove_space(struct inode *inode, ext4_lblk_t offset,
> > +			 ext4_lblk_t len)
> 
> Again, my subjective opinion is that it could be rather called
> ext4_es_remove_extent().

Will fix it.

> 
> > +{
> > +	struct rb_node *node;
> > +	struct ext4_es_tree *tree;
> > +	struct extent_status *es;
> > +	struct extent_status orig_es;
> > +	ext4_lblk_t len1, len2, end;
> > +
> > +	es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
> > +		 offset, len, inode->i_ino);
> 
> Since you're adding tracepoints later on, are those debug messages
> necessary ?

Agree, I will remove this debug message.

> 
> > +
> > +	end = offset + len;
> > +	BUG_ON(end <= offset);
> > +	tree = &EXT4_I(inode)->i_es_tree;
> > +	es = __es_tree_search(&tree->root, offset);
> > +	if (!es)
> > +		goto out;
> > +
> > +	/* Simply invalidate cache_es. */
> > +	tree->cache_es = NULL;
> > +
> > +	orig_es.start = es->start;
> > +	orig_es.len = es->len;
> > +	len1 = offset > es->start ? offset - es->start : 0;
> > +	len2 = extent_status_end(es) > end ?
> > +	       extent_status_end(es) - end : 0;
> > +	if (len1 > 0)
> > +		es->len = len1;
> > +	if (len2 > 0) {
> > +		if (len1 > 0) {
> > +			int err;
> > +			err = ext4_es_add_space(inode, end, len2);
> > +			if (err) {
> > +				es->start = orig_es.start;
> > +				es->len = orig_es.len;
> > +				return err;
> > +			}
> > +		} else {
> > +			es->start = end;
> > +			es->len = len2;
> > +		}
> > +		goto out;
> > +	}
> > +
> > +	if (len1 > 0) {
> > +		node = rb_next(&es->rb_node);
> > +		if (!node)
> > +			es = rb_entry(node, struct extent_status, rb_node);
> > +		else
> > +			es = NULL;
> > +	}
> > +
> > +	while (es && extent_status_end(es) <= end) {
> > +		node = rb_next(&es->rb_node);
> > +		rb_erase(&es->rb_node, &tree->root);
> > +		ext4_es_free_extent(es);
> > +		if (!node) {
> > +			es = NULL;
> > +			break;
> > +		}
> > +		es = rb_entry(node, struct extent_status, rb_node);
> > +	}
> > +
> > +	if (es && es->start < end) {
> > +		len1 = extent_status_end(es) - end;
> > +		es->start = end;
> > +		es->len = len1;
> > +	}
> > +
> > +out:
> > +	ext4_es_print_tree(inode);
> > +	return 0;
> > +}
> > diff --git a/fs/ext4/extents_status.h b/fs/ext4/extents_status.h
> > index d87fd71..8fe8084 100644
> > --- a/fs/ext4/extents_status.h
> > +++ b/fs/ext4/extents_status.h
> > @@ -8,6 +8,12 @@
> >  #ifndef _EXT4_EXTENTS_STATUS_H
> >  #define _EXT4_EXTENTS_STATUS_H
> >  
> > +#ifdef ES_DEBUG
> > +#define es_debug(a...)         printk(a)
> 
> Please take a look at the ext4_debug() or ext_debug(), maybe you can
> use it, or at least reuse the code from it.

Yes, I will reuse the code from ext4_debug() as much as possible.

Regards,
Zheng

> 
> > +#else
> > +#define es_debug(a...)
> > +#endif
> > +
> >  struct extent_status {
> >  	struct rb_node rb_node;
> >  	ext4_lblk_t start;	/* first block extent covers */
> > @@ -19,4 +25,15 @@ struct ext4_es_tree {
> >  	struct extent_status *cache_es;	/* recently accessed extent */
> >  };
> >  
> > +extern int __init ext4_init_es(void);
> > +extern void ext4_exit_es(void);
> > +extern void ext4_es_init_tree(struct ext4_es_tree *tree);
> > +
> > +extern int ext4_es_add_space(struct inode *inode, ext4_lblk_t start,
> > +				ext4_lblk_t len);
> > +extern int ext4_es_remove_space(struct inode *inode, ext4_lblk_t start,
> > +				ext4_lblk_t len);
> > +extern ext4_lblk_t ext4_es_find_extent(struct inode *inode,
> > +				struct extent_status *es);
> > +
> >  #endif /* _EXT4_EXTENTS_STATUS_H */
> > 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/ext4/Makefile b/fs/ext4/Makefile
index 56fd8f86..41f22be 100644
--- a/fs/ext4/Makefile
+++ b/fs/ext4/Makefile
@@ -7,7 +7,7 @@  obj-$(CONFIG_EXT4_FS) += ext4.o
 ext4-y	:= balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
 		ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
 		ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
-		mmp.o indirect.o
+		mmp.o indirect.o extents_status.o
 
 ext4-$(CONFIG_EXT4_FS_XATTR)		+= xattr.o xattr_user.o xattr_trusted.o
 ext4-$(CONFIG_EXT4_FS_POSIX_ACL)	+= acl.o
diff --git a/fs/ext4/extents_status.c b/fs/ext4/extents_status.c
new file mode 100644
index 0000000..bd4e589
--- /dev/null
+++ b/fs/ext4/extents_status.c
@@ -0,0 +1,418 @@ 
+/*
+ *  fs/ext4/extents_status.c
+ *
+ * Written by Yongqiang Yang <xiaoqiangnk@gmail.com>
+ *
+ * Ext4 extents status tree core functions.
+ */
+#include <linux/rbtree.h>
+#include "ext4.h"
+#include "extents_status.h"
+#include "ext4_extents.h"
+
+/*
+ * extents status tree implementation for ext4.
+ *
+ *
+ * ==========================================================================
+ * Extents status encompass delayed extents and extent locks
+ *
+ * 1. Why delayed extent implementation ?
+ *
+ * Without delayed extent, ext4 identifies a delayed extent by looking up
+ * page cache, this has several deficiencies - complicated, buggy, and
+ * inefficient code.
+ *
+ * FIEMAP, SEEK_HOLE/DATA, bigalloc, punch hole and writeout all need to know if
+ * a block or a range of blocks are belonged to a delayed extent.
+ *
+ * Let us have a look at how they do without delayed extents implementation.
+ *   --	FIEMAP
+ *	FIEMAP looks up page cache to identify delayed allocations from holes.
+ *
+ *   --	SEEK_HOLE/DATA
+ *	SEEK_HOLE/DATA has the same problem as FIEMAP.
+ *
+ *   --	bigalloc
+ *	bigalloc looks up page cache to figure out if a block is already
+ *	under delayed allocation or not to determine whether quota reserving
+ *	is needed for the cluster.
+ *
+ *   -- punch hole
+ *	punch hole looks up page cache to identify a delayed extent.
+ *
+ *   --	writeout
+ *	Writeout looks up whole page cache to see if a buffer is mapped, If
+ *	there are not very many delayed buffers, then it is time comsuming.
+ *
+ * With delayed extents implementation, FIEMAP, SEEK_HOLE/DATA, bigalloc and
+ * writeout can figure out if a block or a range of blocks is under delayed
+ * allocation(belonged to a delayed extent) or not by searching the delayed
+ * extent tree.
+ *
+ *
+ * ==========================================================================
+ * 2. ext4 delayed extents impelmentation
+ *
+ *   --	delayed extent
+ *	A delayed extent is a range of blocks which are contiguous logically and
+ *	under delayed allocation.  Unlike extent in ext4, delayed extent in ext4
+ *	is a in-memory struct, there is no corresponding on-disk data.  There is
+ *	no limit on length of delayed extent, so a delayed extent can contain as
+ *	many blocks as they are contiguous logically.
+ *
+ *   --	delayed extent tree
+ *	Every inode has a delayed extent tree and all under delayed allocation
+ *	blocks are added to the tree as dealyed extents.  Delayed extents in
+ *	the tree are ordered by logical block no.
+ *
+ *   --	operations on a delayed extent tree
+ *	There are three operations on a delayed extent tree: find next delayed
+ *	extent, adding a space(a range of blocks) and removing a space.
+ *
+ *   --	race on a delayed extent tree
+ *	Delayed extent tree is protected inode->i_data_sem like extent tree.
+ *
+ *
+ * ==========================================================================
+ * 3. performance analysis
+ *   --	overhead
+ *	1. Apart from operations on a delayed extent tree, we need to
+ *	down_write(inode->i_data_sem) in delayed write path to maintain delayed
+ *	extent tree, this can have impact on parallel read-write and write-write
+ *
+ *	2. There is a cache extent for write access, so if writes are not very
+ *	random, adding space operaions are in O(1) time.
+ *
+ *   --	gain
+ *	3. Code is much simpler, more readable, more maintainable and
+ *      more efficient.
+ */
+
+static struct kmem_cache *ext4_es_cachep;
+
+int __init ext4_init_es(void)
+{
+	ext4_es_cachep = KMEM_CACHE(extent_status, SLAB_RECLAIM_ACCOUNT);
+	if (ext4_es_cachep == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+void ext4_exit_es(void)
+{
+	if (ext4_es_cachep)
+		kmem_cache_destroy(ext4_es_cachep);
+}
+
+void ext4_es_init_tree(struct ext4_es_tree *tree)
+{
+	tree->root = RB_ROOT;
+	tree->cache_es = NULL;
+}
+
+#ifdef SE_DEBUG
+static void ext4_es_print_tree(struct inode *inode)
+{
+	struct ext4_es_tree *tree;
+	struct rb_node *node;
+
+	printk(KERN_DEBUG "status extents for inode %lu:", inode->i_ino);
+	tree = &EXT4_I(inode)->i_es_tree;
+	node = rb_first(&tree->root);
+	while (node) {
+		struct extent_status *es;
+		es = rb_entry(node, struct extent_status, rb_node);
+		printk(KERN_DEBUG " [%u/%u)", es->start, es->len);
+		node = rb_next(node);
+	}
+	printk(KERN_DEBUG "\n");
+}
+#else
+#define ext4_es_print_tree(inode)
+#endif
+
+static inline ext4_lblk_t extent_status_end(struct extent_status *es)
+{
+	if (es->start + es->len < es->start)
+		return (ext4_lblk_t)-1;
+	return es->start + es->len;
+}
+
+/*
+ * search through the tree for an delayed_extent with a given offset.  If
+ * it can't be found, try to find next extent.
+ */
+static struct extent_status *__es_tree_search(struct rb_root *root,
+						ext4_lblk_t offset)
+{
+	struct rb_node *node = root->rb_node;
+	struct extent_status *es = NULL;
+
+	while (node) {
+		es = rb_entry(node, struct extent_status, rb_node);
+		if (offset < es->start)
+			node = node->rb_left;
+		else if (offset >= extent_status_end(es))
+			node = node->rb_right;
+		else
+			return es;
+	}
+
+	if (es && offset < es->start)
+		return es;
+
+	if (es && offset >= extent_status_end(es)) {
+		node = rb_next(&es->rb_node);
+		return node ? rb_entry(node, struct extent_status, rb_node) :
+			      NULL;
+	}
+
+	return NULL;
+}
+
+/*
+ * ext4_es_find_extent: find the 1st delayed extent covering @es->start
+ * if it exists, otherwise, the next extent after @es->start.
+ *
+ * @inode: the inode which owns delayed extents
+ * @es: delayed extent that we found
+ *
+ * Returns next block beyond the found extent.
+ * Delayed extent is returned via @es.
+ */
+ext4_lblk_t ext4_es_find_extent(struct inode *inode, struct extent_status *es)
+{
+	struct ext4_es_tree *tree;
+	struct extent_status *es1;
+	struct rb_node *node;
+	ext4_lblk_t ret = EXT_MAX_BLOCKS;
+
+	es->len = 0;
+	tree = &EXT4_I(inode)->i_es_tree;
+	es1 = __es_tree_search(&tree->root, es->start);
+	if (es1) {
+		tree->cache_es = es1;
+		es->start = es1->start;
+		es->len = es1->len;
+		node = rb_next(&es1->rb_node);
+		if (node) {
+			es1 = rb_entry(node, struct extent_status, rb_node);
+			ret = es1->start;
+		}
+	}
+
+	return ret;
+}
+
+static struct extent_status *
+ext4_es_alloc_extent(ext4_lblk_t start, ext4_lblk_t len)
+{
+	struct extent_status *es;
+	es = kmem_cache_alloc(ext4_es_cachep, GFP_NOFS);
+	if (es == NULL)
+		return NULL;
+	es->start = start;
+	es->len = len;
+	return es;
+}
+
+static void ext4_es_free_extent(struct extent_status *es)
+{
+	kmem_cache_free(ext4_es_cachep, es);
+}
+
+static void ext4_es_try_to_merge_left(struct ext4_es_tree *tree,
+				      struct extent_status *es)
+{
+	struct extent_status *es1;
+	struct rb_node *node;
+
+	node = rb_prev(&es->rb_node);
+	if (!node)
+		return;
+
+	es1 = rb_entry(node, struct extent_status, rb_node);
+	if (extent_status_end(es1) == es->start) {
+		es1->len += es->len;
+		rb_erase(&es->rb_node, &tree->root);
+		if (es == tree->cache_es)
+			tree->cache_es = es1;
+		ext4_es_free_extent(es);
+	}
+}
+
+static void ext4_es_try_to_merge_right(struct ext4_es_tree *tree,
+				       struct extent_status *es)
+{
+	struct extent_status *es1;
+	struct rb_node *node;
+
+	node = rb_next(&es->rb_node);
+	if (!node)
+		return;
+
+	es1 = rb_entry(node, struct extent_status, rb_node);
+	if (es1->start == extent_status_end(es)) {
+		es->len += es1->len;
+		rb_erase(node, &tree->root);
+		if (es1 == tree->cache_es)
+			tree->cache_es = es;
+		ext4_es_free_extent(es1);
+	}
+}
+
+/*
+ * ext4_es_add_space: adds a space to a delayed extent tree.
+ * Caller holds inode->i_data_sem.
+ *
+ * ext4_es_add_space is callyed by ext4_dealyed_write_begin and
+ * ext4_es_remove_space.
+ *
+ * Return 0 on success, error code on failure.
+ */
+int ext4_es_add_space(struct inode *inode, ext4_lblk_t offset, ext4_lblk_t len)
+{
+	struct ext4_es_tree *tree = &EXT4_I(inode)->i_es_tree;
+	struct rb_node **p = &tree->root.rb_node;
+	struct rb_node *parent = NULL;
+	struct extent_status *es;
+	ext4_lblk_t end = offset + len;
+
+	BUG_ON(end <= offset);
+
+	es = tree->cache_es;
+	es_debug("add [%u/%u) to extent status tree of inode %lu\n",
+		 offset, len, inode->i_ino);
+
+	if (es && extent_status_end(es) == offset) {
+		es_debug("cached by [%u/%u)\n", es->start, es->len);
+		es->len += len;
+		ext4_es_try_to_merge_right(tree, es);
+		goto out;
+	} else if (es && es->start == end) {
+		es_debug("cached by [%u/%u)\n", es->start, es->len);
+		es->start = offset;
+		es->len += len;
+		ext4_es_try_to_merge_left(tree, es);
+		goto out;
+	} else if (es && es->start <= offset &&
+		   extent_status_end(es) >= end) {
+		es_debug("cached by [%u/%u)\n", es->start, es->len);
+		goto out;
+	}
+
+	while (*p) {
+		parent = *p;
+		es = rb_entry(parent, struct extent_status, rb_node);
+
+		if (offset < es->start) {
+			if (end == es->start) {
+				es->len += len;
+				es->start = offset;
+				goto out;
+			}
+			p = &(*p)->rb_left;
+		} else if (offset >= extent_status_end(es)) {
+			if (extent_status_end(es) == offset) {
+				es->len += len;
+				goto out;
+			}
+			p = &(*p)->rb_right;
+		} else
+			goto out;
+	}
+
+	es = ext4_es_alloc_extent(offset, len);
+	if (!es)
+		return -ENOMEM;
+	rb_link_node(&es->rb_node, parent, p);
+	rb_insert_color(&es->rb_node, &tree->root);
+
+out:
+	tree->cache_es = es;
+	ext4_es_print_tree(inode);
+
+	return 0;
+}
+
+/*
+ * ext4_es_remove_space() removes a space from a delayed extent tree.
+ * Caller holds inode->i_data_sem.
+ *
+ * Return 0 on success, error code on failure.
+ */
+int ext4_es_remove_space(struct inode *inode, ext4_lblk_t offset,
+			 ext4_lblk_t len)
+{
+	struct rb_node *node;
+	struct ext4_es_tree *tree;
+	struct extent_status *es;
+	struct extent_status orig_es;
+	ext4_lblk_t len1, len2, end;
+
+	es_debug("remove [%u/%u) from extent status tree of inode %lu\n",
+		 offset, len, inode->i_ino);
+
+	end = offset + len;
+	BUG_ON(end <= offset);
+	tree = &EXT4_I(inode)->i_es_tree;
+	es = __es_tree_search(&tree->root, offset);
+	if (!es)
+		goto out;
+
+	/* Simply invalidate cache_es. */
+	tree->cache_es = NULL;
+
+	orig_es.start = es->start;
+	orig_es.len = es->len;
+	len1 = offset > es->start ? offset - es->start : 0;
+	len2 = extent_status_end(es) > end ?
+	       extent_status_end(es) - end : 0;
+	if (len1 > 0)
+		es->len = len1;
+	if (len2 > 0) {
+		if (len1 > 0) {
+			int err;
+			err = ext4_es_add_space(inode, end, len2);
+			if (err) {
+				es->start = orig_es.start;
+				es->len = orig_es.len;
+				return err;
+			}
+		} else {
+			es->start = end;
+			es->len = len2;
+		}
+		goto out;
+	}
+
+	if (len1 > 0) {
+		node = rb_next(&es->rb_node);
+		if (!node)
+			es = rb_entry(node, struct extent_status, rb_node);
+		else
+			es = NULL;
+	}
+
+	while (es && extent_status_end(es) <= end) {
+		node = rb_next(&es->rb_node);
+		rb_erase(&es->rb_node, &tree->root);
+		ext4_es_free_extent(es);
+		if (!node) {
+			es = NULL;
+			break;
+		}
+		es = rb_entry(node, struct extent_status, rb_node);
+	}
+
+	if (es && es->start < end) {
+		len1 = extent_status_end(es) - end;
+		es->start = end;
+		es->len = len1;
+	}
+
+out:
+	ext4_es_print_tree(inode);
+	return 0;
+}
diff --git a/fs/ext4/extents_status.h b/fs/ext4/extents_status.h
index d87fd71..8fe8084 100644
--- a/fs/ext4/extents_status.h
+++ b/fs/ext4/extents_status.h
@@ -8,6 +8,12 @@ 
 #ifndef _EXT4_EXTENTS_STATUS_H
 #define _EXT4_EXTENTS_STATUS_H
 
+#ifdef ES_DEBUG
+#define es_debug(a...)         printk(a)
+#else
+#define es_debug(a...)
+#endif
+
 struct extent_status {
 	struct rb_node rb_node;
 	ext4_lblk_t start;	/* first block extent covers */
@@ -19,4 +25,15 @@  struct ext4_es_tree {
 	struct extent_status *cache_es;	/* recently accessed extent */
 };
 
+extern int __init ext4_init_es(void);
+extern void ext4_exit_es(void);
+extern void ext4_es_init_tree(struct ext4_es_tree *tree);
+
+extern int ext4_es_add_space(struct inode *inode, ext4_lblk_t start,
+				ext4_lblk_t len);
+extern int ext4_es_remove_space(struct inode *inode, ext4_lblk_t start,
+				ext4_lblk_t len);
+extern ext4_lblk_t ext4_es_find_extent(struct inode *inode,
+				struct extent_status *es);
+
 #endif /* _EXT4_EXTENTS_STATUS_H */