Liu Bo
2012-Aug-23 11:01 UTC
[PATCH 1/2] Btrfs: use flag EXTENT_DEFRAG for snapshot-aware defrag
We''re going to use this flag EXTENT_DEFRAG to indicate which range
belongs to
defragment so that we can implement snapshow-aware defrag:
We set the EXTENT_DEFRAG flag when dirtying the extents that need
defragmented, so later on writeback thread can differentiate between
normal writeback and writeback started by defragmentation.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
---
fs/btrfs/extent_io.c | 11 +++++++++--
fs/btrfs/extent_io.h | 29 ++++++++++++++---------------
fs/btrfs/file.c | 4 ++--
fs/btrfs/free-space-cache.c | 7 ++++---
fs/btrfs/inode.c | 20 ++++++++++++--------
fs/btrfs/ioctl.c | 8 ++++----
6 files changed, 45 insertions(+), 34 deletions(-)
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index bb25e89..f03ceff 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -1143,6 +1143,14 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64
start, u64 end,
NULL, cached_state, mask);
}
+int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
+ struct extent_state **cached_state, gfp_t mask)
+{
+ return set_extent_bit(tree, start, end,
+ EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG,
+ NULL, cached_state, mask);
+}
+
int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end,
gfp_t mask)
{
@@ -3700,8 +3708,7 @@ int try_release_extent_mapping(struct extent_map_tree
*map,
}
if (!test_range_bit(tree, em->start,
extent_map_end(em) - 1,
- EXTENT_LOCKED | EXTENT_WRITEBACK,
- 0, NULL)) {
+ EXTENT_LOCKED, 0, NULL)) {
remove_extent_mapping(map, em);
/* once for the rb tree */
free_extent_map(em);
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 25900af..d9dee94 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -5,21 +5,18 @@
/* bits for the extent state */
#define EXTENT_DIRTY 1
-#define EXTENT_WRITEBACK (1 << 1)
-#define EXTENT_UPTODATE (1 << 2)
-#define EXTENT_LOCKED (1 << 3)
-#define EXTENT_NEW (1 << 4)
-#define EXTENT_DELALLOC (1 << 5)
-#define EXTENT_DEFRAG (1 << 6)
-#define EXTENT_DEFRAG_DONE (1 << 7)
-#define EXTENT_BUFFER_FILLED (1 << 8)
-#define EXTENT_BOUNDARY (1 << 9)
-#define EXTENT_NODATASUM (1 << 10)
-#define EXTENT_DO_ACCOUNTING (1 << 11)
-#define EXTENT_FIRST_DELALLOC (1 << 12)
-#define EXTENT_NEED_WAIT (1 << 13)
-#define EXTENT_DAMAGED (1 << 14)
-#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
+#define EXTENT_UPTODATE (1 << 1)
+#define EXTENT_LOCKED (1 << 2)
+#define EXTENT_NEW (1 << 3)
+#define EXTENT_DELALLOC (1 << 4)
+#define EXTENT_DEFRAG (1 << 5)
+#define EXTENT_BOUNDARY (1 << 6)
+#define EXTENT_NODATASUM (1 << 7)
+#define EXTENT_DO_ACCOUNTING (1 << 8)
+#define EXTENT_FIRST_DELALLOC (1 << 9)
+#define EXTENT_NEED_WAIT (1 << 10)
+#define EXTENT_DAMAGED (1 << 11)
+#define EXTENT_IOBITS (EXTENT_LOCKED)
#define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
/*
@@ -235,6 +232,8 @@ int convert_extent_bit(struct extent_io_tree *tree, u64
start, u64 end,
int bits, int clear_bits, gfp_t mask);
int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end,
struct extent_state **cached_state, gfp_t mask);
+int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end,
+ struct extent_state **cached_state, gfp_t mask);
int find_first_extent_bit(struct extent_io_tree *tree, u64 start,
u64 *start_ret, u64 *end_ret, int bits);
struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree,
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 9aa01ec..c172868 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -1173,8 +1173,8 @@ again:
clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
- EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
- GFP_NOFS);
+ EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
+ 0, 0, &cached_state, GFP_NOFS);
unlock_extent_cached(&BTRFS_I(inode)->io_tree,
start_pos, last_pos - 1, &cached_state,
GFP_NOFS);
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 6b10acf..53ff2da 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -1023,7 +1023,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root,
struct inode *inode,
ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
if (ret < 0) {
clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1,
- EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL,
+ EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DEFRAG,
+ 0, 0, NULL,
GFP_NOFS);
goto out;
}
@@ -1037,8 +1038,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root,
struct inode *inode,
found_key.offset != offset) {
clear_extent_bit(&BTRFS_I(inode)->io_tree, 0,
inode->i_size - 1,
- EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0,
- NULL, GFP_NOFS);
+ EXTENT_DIRTY | EXTENT_DELALLOC |
+ EXTENT_DEFRAG, 0, 0, NULL, GFP_NOFS);
btrfs_release_path(path);
goto out;
}
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index ea6a4ee..85bc35f 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3529,7 +3529,8 @@ again:
}
clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
- EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
+ EXTENT_DIRTY | EXTENT_DELALLOC |
+ EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
0, 0, &cached_state, GFP_NOFS);
ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
@@ -5996,7 +5997,8 @@ unlock:
if (lockstart < lockend) {
if (create && len < lockend - lockstart) {
clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
- lockstart + len - 1, unlock_bits, 1, 0,
+ lockstart + len - 1,
+ unlock_bits | EXTENT_DEFRAG, 1, 0,
&cached_state, GFP_NOFS);
/*
* Beside unlock, we also need to cleanup reserved space
@@ -6004,8 +6006,8 @@ unlock:
*/
clear_extent_bit(&BTRFS_I(inode)->io_tree,
lockstart + len, lockend,
- unlock_bits | EXTENT_DO_ACCOUNTING,
- 1, 0, NULL, GFP_NOFS);
+ unlock_bits | EXTENT_DO_ACCOUNTING |
+ EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
} else {
clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
lockend, unlock_bits, 1, 0,
@@ -6570,8 +6572,8 @@ static void btrfs_invalidatepage(struct page *page,
unsigned long offset)
*/
clear_extent_bit(tree, page_start, page_end,
EXTENT_DIRTY | EXTENT_DELALLOC |
- EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
- &cached_state, GFP_NOFS);
+ EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
+ EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
/*
* whoever cleared the private bit is responsible
* for the finish_ordered_io
@@ -6587,7 +6589,8 @@ static void btrfs_invalidatepage(struct page *page,
unsigned long offset)
}
clear_extent_bit(tree, page_start, page_end,
EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
- EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
+ EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
+ &cached_state, GFP_NOFS);
__btrfs_releasepage(page, GFP_NOFS);
ClearPageChecked(page);
@@ -6683,7 +6686,8 @@ again:
* prepare_pages in the normal write path.
*/
clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
- EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
+ EXTENT_DIRTY | EXTENT_DELALLOC |
+ EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
0, 0, &cached_state, GFP_NOFS);
ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index a1fbca0..9449b84 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -1022,8 +1022,8 @@ again:
page_start, page_end - 1, 0, &cached_state);
clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start,
page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
- EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
- GFP_NOFS);
+ EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0,
+ &cached_state, GFP_NOFS);
if (i_done != page_cnt) {
spin_lock(&BTRFS_I(inode)->lock);
@@ -1034,8 +1034,8 @@ again:
}
- btrfs_set_extent_delalloc(inode, page_start, page_end - 1,
- &cached_state);
+ set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1,
+ &cached_state, GFP_NOFS);
unlock_extent_cached(&BTRFS_I(inode)->io_tree,
page_start, page_end - 1, &cached_state,
--
1.7.7.6
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
This comes from one of btrfs''s project ideas,
As we defragment files, we break any sharing from other snapshots.
The balancing code will preserve the sharing, and defrag needs to grow this
as well.
Now we''re able to fill the blank with this patch, in which we make full
use of
backref walking stuff.
Here is the basic idea,
o set the writeback ranges started by defragment with flag EXTENT_DEFRAG
o at endio, after we finish updating fs tree, we use backref walking to find
all parents of the ranges and re-link them with the new COWed file layout by
adding corresponding backrefs.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Original-Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
---
fs/btrfs/inode.c | 617 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 617 insertions(+), 0 deletions(-)
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 85bc35f..c612bb7 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -54,6 +54,7 @@
#include "locking.h"
#include "free-space-cache.h"
#include "inode-map.h"
+#include "backref.h"
struct btrfs_iget_args {
u64 ino;
@@ -1847,6 +1848,600 @@ out:
return ret;
}
+struct extent_backref {
+ struct rb_node node;
+ struct old_extent *old;
+ u64 root_id;
+ u64 inum;
+ u64 file_pos;
+ u64 extent_offset;
+ u64 num_bytes;
+ u64 generation;
+};
+
+struct old_extent {
+ struct list_head list;
+ struct new_extent *new;
+
+ u64 extent_offset;
+ u64 bytenr;
+ u64 offset;
+ u64 len;
+ int count;
+};
+
+struct new_extent {
+ struct rb_root root;
+ struct list_head head;
+ struct btrfs_path *path;
+ struct inode *inode;
+ u64 file_pos;
+ u64 len;
+ u64 bytenr;
+ u64 disk_len;
+ u64 compress_type;
+};
+
+struct relink_work {
+ struct work_struct work;
+ struct new_extent *new;
+};
+
+static int backref_comp(struct extent_backref *b1, struct extent_backref *b2)
+{
+ if (b1->root_id < b2->root_id)
+ return -1;
+ else if (b1->root_id > b2->root_id)
+ return 1;
+
+ if (b1->inum < b2->inum)
+ return -1;
+ else if (b1->inum > b2->inum)
+ return 1;
+
+ if (b1->file_pos < b2->file_pos)
+ return -1;
+ else if (b1->file_pos > b2->file_pos)
+ return 1;
+
+ WARN_ON(1);
+ return 0;
+}
+
+static void backref_insert(struct rb_root *root, struct extent_backref
*backref)
+{
+ struct rb_node **p = &root->rb_node;
+ struct rb_node *parent = NULL;
+ struct extent_backref *entry;
+ int ret;
+
+ while (*p) {
+ parent = *p;
+ entry = rb_entry(parent, struct extent_backref, node);
+
+ ret = backref_comp(backref, entry);
+ if (ret < 0)
+ p = &(*p)->rb_left;
+ else
+ p = &(*p)->rb_right;
+ }
+
+ rb_link_node(&backref->node, parent, p);
+ rb_insert_color(&backref->node, root);
+}
+
+/*
+ * Note the backref might has changed, and in this case we just return 0.
+ */
+static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
+ void *ctx)
+{
+ struct btrfs_file_extent_item *extent;
+ struct btrfs_fs_info *fs_info;
+ struct old_extent *old = ctx;
+ struct new_extent *new = old->new;
+ struct btrfs_path *path = new->path;
+ struct btrfs_key key;
+ struct btrfs_root *root;
+ struct extent_backref *backref;
+ struct extent_buffer *leaf;
+ struct inode *inode = new->inode;
+ int slot;
+ int ret;
+ u64 extent_offset;
+ u64 num_bytes;
+
+ if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
+ inum == btrfs_ino(inode))
+ return 0;
+
+ key.objectid = root_id;
+ key.type = BTRFS_ROOT_ITEM_KEY;
+ key.offset = (u64)-1;
+
+ fs_info = BTRFS_I(inode)->root->fs_info;
+ root = btrfs_read_fs_root_no_name(fs_info, &key);
+ if (IS_ERR(root)) {
+ if (PTR_ERR(root) == -ENOENT)
+ return 0;
+ WARN_ON(1);
+ return PTR_ERR(root);
+ }
+
+ key.objectid = inum;
+ key.type = BTRFS_EXTENT_DATA_KEY;
+ if (offset > (u64)-1 << 32)
+ key.offset = 0;
+ else
+ key.offset = offset;
+
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ if (ret < 0) {
+ WARN_ON(1);
+ return ret;
+ }
+
+ while (1) {
+ leaf = path->nodes[0];
+ slot = path->slots[0];
+
+ if (slot >= btrfs_header_nritems(leaf)) {
+ ret = btrfs_next_leaf(root, path);
+ if (ret < 0) {
+ goto out;
+ } else if (ret > 0) {
+ ret = 0;
+ goto out;
+ }
+ continue;
+ }
+
+ btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+
+ if (key.objectid != inum || key.type != BTRFS_EXTENT_DATA_KEY)
+ goto next;
+
+ extent = btrfs_item_ptr(leaf, slot,
+ struct btrfs_file_extent_item);
+
+ if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
+ goto next;
+
+ if (key.offset - btrfs_file_extent_offset(leaf, extent) !+ offset)
+ goto next;
+
+ break;
+next:
+ path->slots[0]++;
+ }
+
+ extent_offset = btrfs_file_extent_offset(leaf, extent);
+ num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
+
+ if (extent_offset >= old->extent_offset + old->offset + old->len
||
+ extent_offset + num_bytes < old->extent_offset + old->offset)
+ goto out;
+
+ backref = kmalloc(sizeof(*backref), GFP_NOFS);
+ if (!backref) {
+ ret = -ENOENT;
+ goto out;
+ }
+
+ backref->root_id = root_id;
+ backref->inum = inum;
+ backref->file_pos = offset + extent_offset;
+ backref->num_bytes = num_bytes;
+ backref->extent_offset = extent_offset;
+ backref->generation = btrfs_file_extent_generation(leaf, extent);
+ backref->old = old;
+ backref_insert(&new->root, backref);
+ old->count++;
+out:
+ btrfs_release_path(path);
+ WARN_ON(ret);
+ return ret;
+}
+
+static noinline bool record_extent_backrefs(struct btrfs_path *path,
+ struct new_extent *new)
+{
+ struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
+ struct old_extent *old, *tmp;
+ int ret;
+ bool del = false;
+
+ new->path = path;
+
+ list_for_each_entry_safe(old, tmp, &new->head, list) {
+ if (del)
+ goto del;
+
+ ret = iterate_inodes_from_logical(old->bytenr, fs_info,
+ path, record_one_backref,
+ old);
+ WARN_ON(ret < 0);
+del:
+ /* no backref to be processed for this extent */
+ if (!old->count) {
+ list_del(&old->list);
+ kfree(old);
+ }
+ }
+
+ if (list_empty(&new->head))
+ return false;
+
+ return true;
+}
+
+/*
+ * Note the backref might has changed, and in this case we just return 0.
+ */
+static noinline int relink_extent_backref(struct btrfs_path *path,
+ struct extent_backref *prev,
+ struct extent_backref *backref)
+{
+ struct btrfs_file_extent_item *extent;
+ struct btrfs_file_extent_item *item;
+ struct btrfs_ordered_extent *ordered;
+ struct btrfs_trans_handle *trans;
+ struct btrfs_fs_info *fs_info;
+ struct btrfs_root *root;
+ struct btrfs_key key;
+ struct extent_buffer *leaf;
+ struct old_extent *old = backref->old;
+ struct new_extent *new = old->new;
+ struct inode *src_inode = new->inode;
+ struct inode *inode;
+ struct extent_state *cached = NULL;
+ int ret = 0;
+ u64 hint_byte;
+ u64 start;
+ u64 len;
+ bool merge = false;
+
+ if (prev && prev->root_id == backref->root_id &&
+ prev->inum == backref->inum &&
+ prev->extent_offset == backref->extent_offset &&
+ prev->file_pos + prev->num_bytes == backref->file_pos)
+ merge = true;
+
+ key.objectid = backref->root_id;
+ key.type = BTRFS_ROOT_ITEM_KEY;
+ key.offset = (u64)-1;
+
+ fs_info = BTRFS_I(src_inode)->root->fs_info;
+ root = btrfs_read_fs_root_no_name(fs_info, &key);
+ if (IS_ERR(root)) {
+ if (PTR_ERR(root) == -ENOENT)
+ return 0;
+ return PTR_ERR(root);
+ }
+
+ key.objectid = backref->inum;
+ key.type = BTRFS_INODE_ITEM_KEY;
+ key.offset = 0;
+
+ inode = btrfs_iget(fs_info->sb, &key, root, NULL);
+ if (IS_ERR_OR_NULL(inode) || is_bad_inode(inode)) {
+ if (inode && !IS_ERR(inode))
+ iput(inode);
+ return 0;
+ }
+
+ lock_extent_bits(&BTRFS_I(inode)->io_tree, backref->file_pos,
+ backref->file_pos + backref->num_bytes, 0, &cached);
+
+ ordered = btrfs_lookup_first_ordered_extent(inode,
+ backref->file_pos +
+ backref->num_bytes);
+ if (ordered) {
+ btrfs_put_ordered_extent(ordered);
+ goto out_unlock;
+ }
+
+ trans = btrfs_start_transaction(root, 3);
+ if (IS_ERR(trans)) {
+ ret = PTR_ERR(trans);
+ goto out_unlock;
+ }
+
+ key.objectid = backref->inum;
+ key.type = BTRFS_EXTENT_DATA_KEY;
+ key.offset = backref->file_pos;
+
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ if (ret < 0) {
+ goto out_free_path;
+ } else if (ret > 0) {
+ ret = 0;
+ goto out_free_path;
+ }
+
+ extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
+ struct btrfs_file_extent_item);
+
+ if (btrfs_file_extent_generation(path->nodes[0], extent) !+
backref->generation)
+ goto out_free_path;
+
+ btrfs_release_path(path);
+
+ start = backref->file_pos;
+ if (backref->extent_offset < old->extent_offset + old->offset)
+ start += old->extent_offset + old->offset -
+ backref->extent_offset;
+
+ len = min(backref->extent_offset + backref->num_bytes,
+ old->extent_offset + old->offset + old->len);
+ len -= max(backref->extent_offset, old->extent_offset + old->offset);
+
+ ret = btrfs_drop_extents(trans, inode, start,
+ start + len, &hint_byte, 1);
+ if (ret)
+ goto out_free_path;
+again:
+ key.objectid = btrfs_ino(inode);
+ key.type = BTRFS_EXTENT_DATA_KEY;
+ key.offset = start;
+
+ if (merge) {
+ struct btrfs_file_extent_item *fi;
+ u64 extent_len;
+ struct btrfs_key found_key;
+
+ ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
+ if (ret < 0)
+ goto out_free_path;
+
+ path->slots[0]--;
+ leaf = path->nodes[0];
+ btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+
+ fi = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_file_extent_item);
+ extent_len = btrfs_file_extent_num_bytes(leaf, fi);
+
+ if (btrfs_file_extent_disk_bytenr(leaf, fi) == new->bytenr &&
+ btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_REG &&
+ !btrfs_file_extent_compression(leaf, fi) &&
+ !btrfs_file_extent_encryption(leaf, fi) &&
+ !btrfs_file_extent_other_encoding(leaf, fi) &&
+ extent_len + found_key.offset == start) {
+ btrfs_set_file_extent_num_bytes(leaf, fi,
+ extent_len + len);
+ btrfs_mark_buffer_dirty(leaf);
+ inode_add_bytes(inode, len);
+
+ ret = 1;
+ goto out_free_path;
+ } else {
+ merge = false;
+ btrfs_release_path(path);
+ goto again;
+ }
+ }
+
+ ret = btrfs_insert_empty_item(trans, root, path, &key,
+ sizeof(*extent));
+ BUG_ON(ret);
+
+ leaf = path->nodes[0];
+ item = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_file_extent_item);
+ btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
+ btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
+ btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
+ btrfs_set_file_extent_num_bytes(leaf, item, len);
+ btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
+ btrfs_set_file_extent_generation(leaf, item, trans->transid);
+ btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
+ btrfs_set_file_extent_compression(leaf, item, new->compress_type);
+ btrfs_set_file_extent_encryption(leaf, item, 0);
+ btrfs_set_file_extent_other_encoding(leaf, item, 0);
+
+ btrfs_mark_buffer_dirty(leaf);
+ inode_add_bytes(inode, len);
+
+ ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
+ new->disk_len, 0,
+ backref->root_id, backref->inum,
+ start, 0);
+ BUG_ON(ret);
+
+ ret = 1;
+out_free_path:
+ btrfs_release_path(path);
+ btrfs_end_transaction(trans, root);
+out_unlock:
+ unlock_extent_cached(&BTRFS_I(inode)->io_tree, backref->file_pos,
+ backref->file_pos + backref->num_bytes,
+ &cached, GFP_NOFS);
+ iput(inode);
+ return ret;
+}
+
+static void relink_file_extents(struct work_struct *work)
+{
+ struct relink_work *rwork;
+ struct btrfs_path *path;
+ struct new_extent *new;
+ struct old_extent *old, *tmp;
+ struct extent_backref *backref;
+ struct extent_backref *prev = NULL;
+ struct inode *inode;
+ struct btrfs_root *root;
+ struct rb_node *node;
+ struct extent_state *cached = NULL;
+ int ret;
+
+ rwork = container_of(work, struct relink_work, work);
+ new = rwork->new;
+ inode = new->inode;
+ root = BTRFS_I(inode)->root;
+
+ path = btrfs_alloc_path();
+ if (!path)
+ return;
+
+ if (!record_extent_backrefs(path, new)) {
+ btrfs_free_path(path);
+ goto out;
+ }
+ btrfs_release_path(path);
+
+ lock_extent_bits(&BTRFS_I(inode)->io_tree, new->file_pos,
+ new->file_pos + new->len, 0, &cached);
+
+ while (1) {
+ node = rb_first(&new->root);
+ if (!node)
+ break;
+ rb_erase(node, &new->root);
+
+ backref = rb_entry(node, struct extent_backref, node);
+
+ ret = relink_extent_backref(path, prev, backref);
+ WARN_ON(ret < 0);
+
+ kfree(prev);
+
+ if (ret == 1)
+ prev = backref;
+ else
+ prev = NULL;
+ };
+
+ kfree(prev);
+
+ unlock_extent_cached(&BTRFS_I(inode)->io_tree, new->file_pos,
+ new->file_pos + new->len, &cached, GFP_NOFS);
+
+ btrfs_free_path(path);
+
+ list_for_each_entry_safe(old, tmp, &new->head, list) {
+ list_del(&old->list);
+ kfree(old);
+ }
+out:
+ atomic_dec(&root->fs_info->defrag_running);
+ wake_up(&root->fs_info->transaction_wait);
+
+ kfree(new);
+ kfree(rwork);
+}
+
+static struct new_extent *
+record_old_file_extents(struct inode *inode,
+ struct btrfs_ordered_extent *ordered)
+{
+ struct btrfs_root *root = BTRFS_I(inode)->root;
+ struct btrfs_path *path;
+ struct btrfs_key key;
+ struct old_extent *old, *tmp;
+ struct new_extent *new;
+ int ret;
+
+ new = kmalloc(sizeof(*new), GFP_NOFS);
+ if (!new)
+ return NULL;
+
+ new->inode = inode;
+ new->file_pos = ordered->file_offset;
+ new->len = ordered->len;
+ new->bytenr = ordered->start;
+ new->disk_len = ordered->disk_len;
+ new->compress_type = ordered->compress_type;
+ new->root = RB_ROOT;
+ INIT_LIST_HEAD(&new->head);
+
+ path = btrfs_alloc_path();
+ if (!path)
+ goto out_kfree;
+
+ key.objectid = btrfs_ino(inode);
+ key.type = BTRFS_EXTENT_DATA_KEY;
+ key.offset = new->file_pos;
+
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ if (ret < 0)
+ goto out_free_path;
+ if (ret > 0 && path->slots[0] > 0)
+ path->slots[0]--;
+
+ /* find out all the old extents for the file range */
+ while (1) {
+ struct btrfs_file_extent_item *extent;
+ struct extent_buffer *l;
+ int slot;
+ u64 num_bytes;
+ u64 offset;
+ u64 end;
+
+ l = path->nodes[0];
+ slot = path->slots[0];
+
+ if (slot >= btrfs_header_nritems(l)) {
+ ret = btrfs_next_leaf(root, path);
+ if (ret < 0)
+ goto out_free_list;
+ else if (ret > 0)
+ break;
+ continue;
+ }
+
+ btrfs_item_key_to_cpu(l, &key, slot);
+
+ if (key.objectid != btrfs_ino(inode))
+ break;
+ if (key.type != BTRFS_EXTENT_DATA_KEY)
+ break;
+ if (key.offset >= new->file_pos + new->len)
+ break;
+
+ extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
+
+ num_bytes = btrfs_file_extent_num_bytes(l, extent);
+ if (key.offset + num_bytes < new->file_pos)
+ goto next;
+
+ old = kmalloc(sizeof(*old), GFP_NOFS);
+ if (!old)
+ goto out_free_list;
+
+ offset = max(new->file_pos, key.offset);
+ end = min(new->file_pos + new->len, key.offset + num_bytes);
+
+ old->bytenr = btrfs_file_extent_disk_bytenr(l, extent);
+ old->extent_offset = btrfs_file_extent_offset(l, extent);
+ old->offset = offset - key.offset;
+ old->len = end - offset;
+ old->new = new;
+ old->count = 0;
+ list_add_tail(&old->list, &new->head);
+next:
+ path->slots[0]++;
+ }
+
+ btrfs_free_path(path);
+ atomic_inc(&root->fs_info->defrag_running);
+
+ return new;
+
+out_free_list:
+ list_for_each_entry_safe(old, tmp, &new->head, list) {
+ list_del(&old->list);
+ kfree(old);
+ }
+out_free_path:
+ btrfs_free_path(path);
+out_kfree:
+ kfree(new);
+ return NULL;
+}
+
/*
* helper function for btrfs_finish_ordered_io, this
* just reads in some of the csum leaves to prime them into ram
@@ -1864,6 +2459,7 @@ static int btrfs_finish_ordered_io(struct
btrfs_ordered_extent *ordered_extent)
struct btrfs_trans_handle *trans = NULL;
struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
struct extent_state *cached_state = NULL;
+ struct relink_work *work = NULL;
int compress_type = 0;
int ret;
bool nolock;
@@ -1897,6 +2493,23 @@ static int btrfs_finish_ordered_io(struct
btrfs_ordered_extent *ordered_extent)
ordered_extent->file_offset + ordered_extent->len - 1,
0, &cached_state);
+ ret = test_range_bit(io_tree, ordered_extent->file_offset,
+ ordered_extent->file_offset + ordered_extent->len - 1,
+ EXTENT_DEFRAG, 1, cached_state);
+ if (ret && (btrfs_root_last_snapshot(&root->root_item) >+
BTRFS_I(inode)->generation)) {
+ /* the inode is shared */
+ work = kmalloc(sizeof(*work), GFP_NOFS);
+ if (work) {
+ work->new = record_old_file_extents(inode,
+ ordered_extent);
+ if (!work->new) {
+ kfree(work);
+ work = NULL;
+ }
+ }
+ }
+
if (nolock)
trans = btrfs_join_transaction_nolock(root);
else
@@ -1973,6 +2586,10 @@ out:
*/
btrfs_remove_ordered_extent(inode, ordered_extent);
+ /* for snapshot-aware defrag */
+ if (work)
+ relink_file_extents(&work->work);
+
/* once for us */
btrfs_put_ordered_extent(ordered_extent);
/* once for the tree */
--
1.7.7.6
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Liu Bo
2012-Aug-23 11:08 UTC
Re: [PATCH 1/2] Btrfs: use flag EXTENT_DEFRAG for snapshot-aware defrag
On 08/23/2012 07:01 PM, Liu Bo wrote:> We''re going to use this flag EXTENT_DEFRAG to indicate which range > belongs to > defragment so that we can implement snapshow-aware defrag: > > We set the EXTENT_DEFRAG flag when dirtying the extents that need > defragmented, so later on writeback thread can differentiate between > normal writeback and writeback started by defragmentation. > > Signed-off-by: Liu Bo <bo.li.liu@oracle.com>Sorry, I missed this: Original-Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>> --- > fs/btrfs/extent_io.c | 11 +++++++++-- > fs/btrfs/extent_io.h | 29 ++++++++++++++--------------- > fs/btrfs/file.c | 4 ++-- > fs/btrfs/free-space-cache.c | 7 ++++--- > fs/btrfs/inode.c | 20 ++++++++++++-------- > fs/btrfs/ioctl.c | 8 ++++---- > 6 files changed, 45 insertions(+), 34 deletions(-) > > diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c > index bb25e89..f03ceff 100644 > --- a/fs/btrfs/extent_io.c > +++ b/fs/btrfs/extent_io.c > @@ -1143,6 +1143,14 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, > NULL, cached_state, mask); > } > > +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, > + struct extent_state **cached_state, gfp_t mask) > +{ > + return set_extent_bit(tree, start, end, > + EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, > + NULL, cached_state, mask); > +} > + > int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, > gfp_t mask) > { > @@ -3700,8 +3708,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, > } > if (!test_range_bit(tree, em->start, > extent_map_end(em) - 1, > - EXTENT_LOCKED | EXTENT_WRITEBACK, > - 0, NULL)) { > + EXTENT_LOCKED, 0, NULL)) { > remove_extent_mapping(map, em); > /* once for the rb tree */ > free_extent_map(em); > diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h > index 25900af..d9dee94 100644 > --- a/fs/btrfs/extent_io.h > +++ b/fs/btrfs/extent_io.h > @@ -5,21 +5,18 @@ > > /* bits for the extent state */ > #define EXTENT_DIRTY 1 > -#define EXTENT_WRITEBACK (1 << 1) > -#define EXTENT_UPTODATE (1 << 2) > -#define EXTENT_LOCKED (1 << 3) > -#define EXTENT_NEW (1 << 4) > -#define EXTENT_DELALLOC (1 << 5) > -#define EXTENT_DEFRAG (1 << 6) > -#define EXTENT_DEFRAG_DONE (1 << 7) > -#define EXTENT_BUFFER_FILLED (1 << 8) > -#define EXTENT_BOUNDARY (1 << 9) > -#define EXTENT_NODATASUM (1 << 10) > -#define EXTENT_DO_ACCOUNTING (1 << 11) > -#define EXTENT_FIRST_DELALLOC (1 << 12) > -#define EXTENT_NEED_WAIT (1 << 13) > -#define EXTENT_DAMAGED (1 << 14) > -#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) > +#define EXTENT_UPTODATE (1 << 1) > +#define EXTENT_LOCKED (1 << 2) > +#define EXTENT_NEW (1 << 3) > +#define EXTENT_DELALLOC (1 << 4) > +#define EXTENT_DEFRAG (1 << 5) > +#define EXTENT_BOUNDARY (1 << 6) > +#define EXTENT_NODATASUM (1 << 7) > +#define EXTENT_DO_ACCOUNTING (1 << 8) > +#define EXTENT_FIRST_DELALLOC (1 << 9) > +#define EXTENT_NEED_WAIT (1 << 10) > +#define EXTENT_DAMAGED (1 << 11) > +#define EXTENT_IOBITS (EXTENT_LOCKED) > #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC) > > /* > @@ -235,6 +232,8 @@ int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, > int bits, int clear_bits, gfp_t mask); > int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, > struct extent_state **cached_state, gfp_t mask); > +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, > + struct extent_state **cached_state, gfp_t mask); > int find_first_extent_bit(struct extent_io_tree *tree, u64 start, > u64 *start_ret, u64 *end_ret, int bits); > struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, > diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c > index 9aa01ec..c172868 100644 > --- a/fs/btrfs/file.c > +++ b/fs/btrfs/file.c > @@ -1173,8 +1173,8 @@ again: > > clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, > last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, > - GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > + 0, 0, &cached_state, GFP_NOFS); > unlock_extent_cached(&BTRFS_I(inode)->io_tree, > start_pos, last_pos - 1, &cached_state, > GFP_NOFS); > diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c > index 6b10acf..53ff2da 100644 > --- a/fs/btrfs/free-space-cache.c > +++ b/fs/btrfs/free-space-cache.c > @@ -1023,7 +1023,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, > ret = btrfs_search_slot(trans, root, &key, path, 0, 1); > if (ret < 0) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, > - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, > + EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DEFRAG, > + 0, 0, NULL, > GFP_NOFS); > goto out; > } > @@ -1037,8 +1038,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, > found_key.offset != offset) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, > inode->i_size - 1, > - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, > - NULL, GFP_NOFS); > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DEFRAG, 0, 0, NULL, GFP_NOFS); > btrfs_release_path(path); > goto out; > } > diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c > index ea6a4ee..85bc35f 100644 > --- a/fs/btrfs/inode.c > +++ b/fs/btrfs/inode.c > @@ -3529,7 +3529,8 @@ again: > } > > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, > - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > 0, 0, &cached_state, GFP_NOFS); > > ret = btrfs_set_extent_delalloc(inode, page_start, page_end, > @@ -5996,7 +5997,8 @@ unlock: > if (lockstart < lockend) { > if (create && len < lockend - lockstart) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, > - lockstart + len - 1, unlock_bits, 1, 0, > + lockstart + len - 1, > + unlock_bits | EXTENT_DEFRAG, 1, 0, > &cached_state, GFP_NOFS); > /* > * Beside unlock, we also need to cleanup reserved space > @@ -6004,8 +6006,8 @@ unlock: > */ > clear_extent_bit(&BTRFS_I(inode)->io_tree, > lockstart + len, lockend, > - unlock_bits | EXTENT_DO_ACCOUNTING, > - 1, 0, NULL, GFP_NOFS); > + unlock_bits | EXTENT_DO_ACCOUNTING | > + EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS); > } else { > clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, > lockend, unlock_bits, 1, 0, > @@ -6570,8 +6572,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) > */ > clear_extent_bit(tree, page_start, page_end, > EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, > - &cached_state, GFP_NOFS); > + EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | > + EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); > /* > * whoever cleared the private bit is responsible > * for the finish_ordered_io > @@ -6587,7 +6589,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) > } > clear_extent_bit(tree, page_start, page_end, > EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, > + &cached_state, GFP_NOFS); > __btrfs_releasepage(page, GFP_NOFS); > > ClearPageChecked(page); > @@ -6683,7 +6686,8 @@ again: > * prepare_pages in the normal write path. > */ > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, > - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > 0, 0, &cached_state, GFP_NOFS); > > ret = btrfs_set_extent_delalloc(inode, page_start, page_end, > diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c > index a1fbca0..9449b84 100644 > --- a/fs/btrfs/ioctl.c > +++ b/fs/btrfs/ioctl.c > @@ -1022,8 +1022,8 @@ again: > page_start, page_end - 1, 0, &cached_state); > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, > page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, > - GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, > + &cached_state, GFP_NOFS); > > if (i_done != page_cnt) { > spin_lock(&BTRFS_I(inode)->lock); > @@ -1034,8 +1034,8 @@ again: > } > > > - btrfs_set_extent_delalloc(inode, page_start, page_end - 1, > - &cached_state); > + set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, > + &cached_state, GFP_NOFS); > > unlock_extent_cached(&BTRFS_I(inode)->io_tree, > page_start, page_end - 1, &cached_state, >-- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
David Sterba
2012-Aug-24 17:34 UTC
Re: [PATCH 1/2] Btrfs: use flag EXTENT_DEFRAG for snapshot-aware defrag
On Thu, Aug 23, 2012 at 07:01:33PM +0800, Liu Bo wrote:> We''re going to use this flag EXTENT_DEFRAG to indicate which range > belongs to > defragment so that we can implement snapshow-aware defrag: > > We set the EXTENT_DEFRAG flag when dirtying the extents that need > defragmented, so later on writeback thread can differentiate between > normal writeback and writeback started by defragmentation. > > Signed-off-by: Liu Bo <bo.li.liu@oracle.com> > --- > fs/btrfs/extent_io.c | 11 +++++++++-- > fs/btrfs/extent_io.h | 29 ++++++++++++++--------------- > fs/btrfs/file.c | 4 ++-- > fs/btrfs/free-space-cache.c | 7 ++++--- > fs/btrfs/inode.c | 20 ++++++++++++-------- > fs/btrfs/ioctl.c | 8 ++++---- > 6 files changed, 45 insertions(+), 34 deletions(-) > > diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c > index bb25e89..f03ceff 100644 > --- a/fs/btrfs/extent_io.c > +++ b/fs/btrfs/extent_io.c > @@ -1143,6 +1143,14 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, > NULL, cached_state, mask); > } > > +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, > + struct extent_state **cached_state, gfp_t mask) > +{ > + return set_extent_bit(tree, start, end, > + EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, > + NULL, cached_state, mask); > +} > + > int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, > gfp_t mask) > { > @@ -3700,8 +3708,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, > } > if (!test_range_bit(tree, em->start, > extent_map_end(em) - 1, > - EXTENT_LOCKED | EXTENT_WRITEBACK, > - 0, NULL)) { > + EXTENT_LOCKED, 0, NULL)) { > remove_extent_mapping(map, em); > /* once for the rb tree */ > free_extent_map(em); > diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h > index 25900af..d9dee94 100644 > --- a/fs/btrfs/extent_io.h > +++ b/fs/btrfs/extent_io.h > @@ -5,21 +5,18 @@ > > /* bits for the extent state */ > #define EXTENT_DIRTY 1 > -#define EXTENT_WRITEBACK (1 << 1) > -#define EXTENT_UPTODATE (1 << 2) > -#define EXTENT_LOCKED (1 << 3) > -#define EXTENT_NEW (1 << 4) > -#define EXTENT_DELALLOC (1 << 5) > -#define EXTENT_DEFRAG (1 << 6) > -#define EXTENT_DEFRAG_DONE (1 << 7) > -#define EXTENT_BUFFER_FILLED (1 << 8) > -#define EXTENT_BOUNDARY (1 << 9) > -#define EXTENT_NODATASUM (1 << 10) > -#define EXTENT_DO_ACCOUNTING (1 << 11) > -#define EXTENT_FIRST_DELALLOC (1 << 12) > -#define EXTENT_NEED_WAIT (1 << 13) > -#define EXTENT_DAMAGED (1 << 14) > -#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) > +#define EXTENT_UPTODATE (1 << 1) > +#define EXTENT_LOCKED (1 << 2) > +#define EXTENT_NEW (1 << 3) > +#define EXTENT_DELALLOC (1 << 4) > +#define EXTENT_DEFRAG (1 << 5) > +#define EXTENT_BOUNDARY (1 << 6) > +#define EXTENT_NODATASUM (1 << 7) > +#define EXTENT_DO_ACCOUNTING (1 << 8) > +#define EXTENT_FIRST_DELALLOC (1 << 9) > +#define EXTENT_NEED_WAIT (1 << 10) > +#define EXTENT_DAMAGED (1 << 11) > +#define EXTENT_IOBITS (EXTENT_LOCKED)Uh, please don''t do that, the values are used only for in-memory accounting and we do not care about their real values. You create a hole in the number sequence, removing EXTENT_WRITEBACK, EXTENT_DEFRAG_DONE and EXTENT_BUFFER_FILLED, fine, so put there a comment instead. We can reuse the value anytime later. Also I''d like to hear why it''s safe to remove the WRITEBACK flag. It''s embedded inside the IOBITS flag that is checked in many places, so if you remove it the condition would be more relaxed and some code can be unexpectedly executed. On the other hand, the WRITEBACK flag is not set anywhere, either directly or via IOBITS, so it''s probably a leftover (like the other flags you remove). So, I''d rather see a separate patch that just removes the unused bits, and then patch that adds the snapshot-aware defrag logic, this is more friendly to testing.> #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC) > > /* > @@ -235,6 +232,8 @@ int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, > int bits, int clear_bits, gfp_t mask); > int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, > struct extent_state **cached_state, gfp_t mask); > +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, > + struct extent_state **cached_state, gfp_t mask); > int find_first_extent_bit(struct extent_io_tree *tree, u64 start, > u64 *start_ret, u64 *end_ret, int bits); > struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, > diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c > index 9aa01ec..c172868 100644 > --- a/fs/btrfs/file.c > +++ b/fs/btrfs/file.c > @@ -1173,8 +1173,8 @@ again: > > clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, > last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, > - GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > + 0, 0, &cached_state, GFP_NOFS); > unlock_extent_cached(&BTRFS_I(inode)->io_tree, > start_pos, last_pos - 1, &cached_state, > GFP_NOFS); > diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c > index 6b10acf..53ff2da 100644 > --- a/fs/btrfs/free-space-cache.c > +++ b/fs/btrfs/free-space-cache.c > @@ -1023,7 +1023,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, > ret = btrfs_search_slot(trans, root, &key, path, 0, 1); > if (ret < 0) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, > - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, > + EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DEFRAG, > + 0, 0, NULL,This does not make much sense to me, why do you need to clear DEFRAG for the free space inode?> GFP_NOFS); > goto out; > } > @@ -1037,8 +1038,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, > found_key.offset != offset) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, > inode->i_size - 1, > - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, > - NULL, GFP_NOFS); > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DEFRAG, 0, 0, NULL, GFP_NOFS); > btrfs_release_path(path); > goto out; > } > diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c > index ea6a4ee..85bc35f 100644 > --- a/fs/btrfs/inode.c > +++ b/fs/btrfs/inode.c > @@ -3529,7 +3529,8 @@ again: > } > > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, > - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > 0, 0, &cached_state, GFP_NOFS); > > ret = btrfs_set_extent_delalloc(inode, page_start, page_end, > @@ -5996,7 +5997,8 @@ unlock: > if (lockstart < lockend) { > if (create && len < lockend - lockstart) { > clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, > - lockstart + len - 1, unlock_bits, 1, 0, > + lockstart + len - 1, > + unlock_bits | EXTENT_DEFRAG, 1, 0,this> &cached_state, GFP_NOFS); > /* > * Beside unlock, we also need to cleanup reserved space > @@ -6004,8 +6006,8 @@ unlock: > */ > clear_extent_bit(&BTRFS_I(inode)->io_tree, > lockstart + len, lockend, > - unlock_bits | EXTENT_DO_ACCOUNTING, > - 1, 0, NULL, GFP_NOFS); > + unlock_bits | EXTENT_DO_ACCOUNTING | > + EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);and this come from the patch that fixes the DIO problem (Btrfs: fix a dio write regression), right? Please mention that, makes reviewer''s life easier :)> } else { > clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, > lockend, unlock_bits, 1, 0, > @@ -6570,8 +6572,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) > */ > clear_extent_bit(tree, page_start, page_end, > EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, > - &cached_state, GFP_NOFS); > + EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | > + EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); > /* > * whoever cleared the private bit is responsible > * for the finish_ordered_io > @@ -6587,7 +6589,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) > } > clear_extent_bit(tree, page_start, page_end, > EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, > + &cached_state, GFP_NOFS); > __btrfs_releasepage(page, GFP_NOFS); > > ClearPageChecked(page); > @@ -6683,7 +6686,8 @@ again: > * prepare_pages in the normal write path. > */ > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, > - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, > + EXTENT_DIRTY | EXTENT_DELALLOC | > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, > 0, 0, &cached_state, GFP_NOFS); > > ret = btrfs_set_extent_delalloc(inode, page_start, page_end, > diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c > index a1fbca0..9449b84 100644 > --- a/fs/btrfs/ioctl.c > +++ b/fs/btrfs/ioctl.c > @@ -1022,8 +1022,8 @@ again: > page_start, page_end - 1, 0, &cached_state); > clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, > page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | > - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, > - GFP_NOFS); > + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, > + &cached_state, GFP_NOFS); > > if (i_done != page_cnt) { > spin_lock(&BTRFS_I(inode)->lock); > @@ -1034,8 +1034,8 @@ again: > } > > > - btrfs_set_extent_delalloc(inode, page_start, page_end - 1, > - &cached_state); > + set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, > + &cached_state, GFP_NOFS); > > unlock_extent_cached(&BTRFS_I(inode)->io_tree, > page_start, page_end - 1, &cached_state, > ---otherwise looks ok. david -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Liu Bo
2012-Aug-25 10:36 UTC
Re: [PATCH 1/2] Btrfs: use flag EXTENT_DEFRAG for snapshot-aware defrag
On 08/25/2012 01:34 AM, David Sterba wrote:> On Thu, Aug 23, 2012 at 07:01:33PM +0800, Liu Bo wrote: >> We''re going to use this flag EXTENT_DEFRAG to indicate which range >> belongs to >> defragment so that we can implement snapshow-aware defrag: >> >> We set the EXTENT_DEFRAG flag when dirtying the extents that need >> defragmented, so later on writeback thread can differentiate between >> normal writeback and writeback started by defragmentation. >> >> Signed-off-by: Liu Bo <bo.li.liu@oracle.com> >> --- >> fs/btrfs/extent_io.c | 11 +++++++++-- >> fs/btrfs/extent_io.h | 29 ++++++++++++++--------------- >> fs/btrfs/file.c | 4 ++-- >> fs/btrfs/free-space-cache.c | 7 ++++--- >> fs/btrfs/inode.c | 20 ++++++++++++-------- >> fs/btrfs/ioctl.c | 8 ++++---- >> 6 files changed, 45 insertions(+), 34 deletions(-) >> >> diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c >> index bb25e89..f03ceff 100644 >> --- a/fs/btrfs/extent_io.c >> +++ b/fs/btrfs/extent_io.c >> @@ -1143,6 +1143,14 @@ int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, >> NULL, cached_state, mask); >> } >> >> +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, >> + struct extent_state **cached_state, gfp_t mask) >> +{ >> + return set_extent_bit(tree, start, end, >> + EXTENT_DELALLOC | EXTENT_UPTODATE | EXTENT_DEFRAG, >> + NULL, cached_state, mask); >> +} >> + >> int clear_extent_dirty(struct extent_io_tree *tree, u64 start, u64 end, >> gfp_t mask) >> { >> @@ -3700,8 +3708,7 @@ int try_release_extent_mapping(struct extent_map_tree *map, >> } >> if (!test_range_bit(tree, em->start, >> extent_map_end(em) - 1, >> - EXTENT_LOCKED | EXTENT_WRITEBACK, >> - 0, NULL)) { >> + EXTENT_LOCKED, 0, NULL)) { >> remove_extent_mapping(map, em); >> /* once for the rb tree */ >> free_extent_map(em); >> diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h >> index 25900af..d9dee94 100644 >> --- a/fs/btrfs/extent_io.h >> +++ b/fs/btrfs/extent_io.h >> @@ -5,21 +5,18 @@ >> >> /* bits for the extent state */ >> #define EXTENT_DIRTY 1 >> -#define EXTENT_WRITEBACK (1 << 1) >> -#define EXTENT_UPTODATE (1 << 2) >> -#define EXTENT_LOCKED (1 << 3) >> -#define EXTENT_NEW (1 << 4) >> -#define EXTENT_DELALLOC (1 << 5) >> -#define EXTENT_DEFRAG (1 << 6) >> -#define EXTENT_DEFRAG_DONE (1 << 7) >> -#define EXTENT_BUFFER_FILLED (1 << 8) >> -#define EXTENT_BOUNDARY (1 << 9) >> -#define EXTENT_NODATASUM (1 << 10) >> -#define EXTENT_DO_ACCOUNTING (1 << 11) >> -#define EXTENT_FIRST_DELALLOC (1 << 12) >> -#define EXTENT_NEED_WAIT (1 << 13) >> -#define EXTENT_DAMAGED (1 << 14) >> -#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) >> +#define EXTENT_UPTODATE (1 << 1) >> +#define EXTENT_LOCKED (1 << 2) >> +#define EXTENT_NEW (1 << 3) >> +#define EXTENT_DELALLOC (1 << 4) >> +#define EXTENT_DEFRAG (1 << 5) >> +#define EXTENT_BOUNDARY (1 << 6) >> +#define EXTENT_NODATASUM (1 << 7) >> +#define EXTENT_DO_ACCOUNTING (1 << 8) >> +#define EXTENT_FIRST_DELALLOC (1 << 9) >> +#define EXTENT_NEED_WAIT (1 << 10) >> +#define EXTENT_DAMAGED (1 << 11) >> +#define EXTENT_IOBITS (EXTENT_LOCKED) > > Uh, please don''t do that, the values are used only for in-memory > accounting and we do not care about their real values. You create a hole > in the number sequence, removing EXTENT_WRITEBACK, EXTENT_DEFRAG_DONE > and EXTENT_BUFFER_FILLED, fine, so put there a comment instead. We can > reuse the value anytime later. > > Also I''d like to hear why it''s safe to remove the WRITEBACK flag. It''s > embedded inside the IOBITS flag that is checked in many places, so if you > remove it the condition would be more relaxed and some code can be > unexpectedly executed. > > On the other hand, the WRITEBACK flag is not set anywhere, either > directly or via IOBITS, so it''s probably a leftover (like the other > flags you remove). >Yeah, it might be some historical reasons why WRITEBACK flag is kept till now.> So, I''d rather see a separate patch that just removes the unused bits, > and then patch that adds the snapshot-aware defrag logic, this is more > friendly to testing. >hmm, make sense, I''ll make a separated patch.> >> #define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC) >> >> /* >> @@ -235,6 +232,8 @@ int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, >> int bits, int clear_bits, gfp_t mask); >> int set_extent_delalloc(struct extent_io_tree *tree, u64 start, u64 end, >> struct extent_state **cached_state, gfp_t mask); >> +int set_extent_defrag(struct extent_io_tree *tree, u64 start, u64 end, >> + struct extent_state **cached_state, gfp_t mask); >> int find_first_extent_bit(struct extent_io_tree *tree, u64 start, >> u64 *start_ret, u64 *end_ret, int bits); >> struct extent_state *find_first_extent_bit_state(struct extent_io_tree *tree, >> diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c >> index 9aa01ec..c172868 100644 >> --- a/fs/btrfs/file.c >> +++ b/fs/btrfs/file.c >> @@ -1173,8 +1173,8 @@ again: >> >> clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, >> last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC | >> - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, >> - GFP_NOFS); >> + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, >> + 0, 0, &cached_state, GFP_NOFS); >> unlock_extent_cached(&BTRFS_I(inode)->io_tree, >> start_pos, last_pos - 1, &cached_state, >> GFP_NOFS); >> diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c >> index 6b10acf..53ff2da 100644 >> --- a/fs/btrfs/free-space-cache.c >> +++ b/fs/btrfs/free-space-cache.c >> @@ -1023,7 +1023,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, >> ret = btrfs_search_slot(trans, root, &key, path, 0, 1); >> if (ret < 0) { >> clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, inode->i_size - 1, >> - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, NULL, >> + EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DEFRAG, >> + 0, 0, NULL, > > This does not make much sense to me, why do you need to clear DEFRAG for > the free space inode? >...Actually I didn''t notice this, I just did grep for clear_extent_bit and add that flag...>> GFP_NOFS); >> goto out; >> } >> @@ -1037,8 +1038,8 @@ int __btrfs_write_out_cache(struct btrfs_root *root, struct inode *inode, >> found_key.offset != offset) { >> clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, >> inode->i_size - 1, >> - EXTENT_DIRTY | EXTENT_DELALLOC, 0, 0, >> - NULL, GFP_NOFS); >> + EXTENT_DIRTY | EXTENT_DELALLOC | >> + EXTENT_DEFRAG, 0, 0, NULL, GFP_NOFS); >> btrfs_release_path(path); >> goto out; >> } >> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c >> index ea6a4ee..85bc35f 100644 >> --- a/fs/btrfs/inode.c >> +++ b/fs/btrfs/inode.c >> @@ -3529,7 +3529,8 @@ again: >> } >> >> clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, >> - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, >> + EXTENT_DIRTY | EXTENT_DELALLOC | >> + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, >> 0, 0, &cached_state, GFP_NOFS); >> >> ret = btrfs_set_extent_delalloc(inode, page_start, page_end, >> @@ -5996,7 +5997,8 @@ unlock: >> if (lockstart < lockend) { >> if (create && len < lockend - lockstart) { >> clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, >> - lockstart + len - 1, unlock_bits, 1, 0, >> + lockstart + len - 1, >> + unlock_bits | EXTENT_DEFRAG, 1, 0, > > this > >> &cached_state, GFP_NOFS); >> /* >> * Beside unlock, we also need to cleanup reserved space >> @@ -6004,8 +6006,8 @@ unlock: >> */ >> clear_extent_bit(&BTRFS_I(inode)->io_tree, >> lockstart + len, lockend, >> - unlock_bits | EXTENT_DO_ACCOUNTING, >> - 1, 0, NULL, GFP_NOFS); >> + unlock_bits | EXTENT_DO_ACCOUNTING | >> + EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS); > > and this come from the patch that fixes the DIO problem (Btrfs: fix a > dio write regression), right? Please mention that, makes reviewer''s life > easier :) >I''m so sorry, Dave, I should write a NOTE in front of the patchset.>> } else { >> clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, >> lockend, unlock_bits, 1, 0, >> @@ -6570,8 +6572,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) >> */ >> clear_extent_bit(tree, page_start, page_end, >> EXTENT_DIRTY | EXTENT_DELALLOC | >> - EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0, >> - &cached_state, GFP_NOFS); >> + EXTENT_LOCKED | EXTENT_DO_ACCOUNTING | >> + EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS); >> /* >> * whoever cleared the private bit is responsible >> * for the finish_ordered_io >> @@ -6587,7 +6589,8 @@ static void btrfs_invalidatepage(struct page *page, unsigned long offset) >> } >> clear_extent_bit(tree, page_start, page_end, >> EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC | >> - EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS); >> + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1, >> + &cached_state, GFP_NOFS); >> __btrfs_releasepage(page, GFP_NOFS); >> >> ClearPageChecked(page); >> @@ -6683,7 +6686,8 @@ again: >> * prepare_pages in the normal write path. >> */ >> clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end, >> - EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING, >> + EXTENT_DIRTY | EXTENT_DELALLOC | >> + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, >> 0, 0, &cached_state, GFP_NOFS); >> >> ret = btrfs_set_extent_delalloc(inode, page_start, page_end, >> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c >> index a1fbca0..9449b84 100644 >> --- a/fs/btrfs/ioctl.c >> +++ b/fs/btrfs/ioctl.c >> @@ -1022,8 +1022,8 @@ again: >> page_start, page_end - 1, 0, &cached_state); >> clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, >> page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | >> - EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, >> - GFP_NOFS); >> + EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, >> + &cached_state, GFP_NOFS); >> >> if (i_done != page_cnt) { >> spin_lock(&BTRFS_I(inode)->lock); >> @@ -1034,8 +1034,8 @@ again: >> } >> >> >> - btrfs_set_extent_delalloc(inode, page_start, page_end - 1, >> - &cached_state); >> + set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, >> + &cached_state, GFP_NOFS); >> >> unlock_extent_cached(&BTRFS_I(inode)->io_tree, >> page_start, page_end - 1, &cached_state, >> --- > > otherwise looks ok. >Thanks for reviewing this! :) thanks, liubo> david >-- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html