Btrfs: switch extent_map to a rw lock
There are two main users of the extent_map tree. The
first is regular file inodes, where it is evenly spread
between readers and writers.
The second is the chunk allocation tree, which maps blocks from
logical addresses to phyiscal ones, and it is 99.99% reads.
The mapping tree is a point of lock contention during heavy IO
workloads, so this commit switches things to a rw lock.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index de1e2fd..78451a5 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -507,10 +507,10 @@
*/
set_page_extent_mapped(page);
lock_extent(tree, last_offset, end, GFP_NOFS);
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, last_offset,
PAGE_CACHE_SIZE);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
if (!em || last_offset < em->start ||
(last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
@@ -594,11 +594,11 @@
em_tree = &BTRFS_I(inode)->extent_tree;
/* we need the actual starting offset of this extent in the file */
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree,
page_offset(bio->bi_io_vec->bv_page),
PAGE_CACHE_SIZE);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
compressed_len = em->block_len;
cb = kmalloc(compressed_bio_size(root, compressed_len), GFP_NOFS);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 20cefc6..b6cfdd9 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -121,15 +121,15 @@
struct extent_map *em;
int ret;
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, len);
if (em) {
em->bdev =
BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
goto out;
}
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
em = alloc_extent_map(GFP_NOFS);
if (!em) {
@@ -142,7 +142,7 @@
em->block_start = 0;
em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
if (ret == -EEXIST) {
u64 failed_start = em->start;
@@ -161,7 +161,7 @@
free_extent_map(em);
em = NULL;
}
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret)
em = ERR_PTR(ret);
@@ -1323,9 +1323,9 @@
offset = page_offset(page);
em_tree = &BTRFS_I(inode)->extent_tree;
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
if (!em) {
__unplug_io_fn(bdi, page);
return;
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 72a2b9c..edd86ae 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5396,9 +5396,9 @@
lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
while (1) {
int ret;
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 8d7a152..41cf1b4 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2786,15 +2786,15 @@
u64 len;
while (start <= end) {
len = end - start + 1;
- spin_lock(&map->lock);
+ write_lock(&map->lock);
em = lookup_extent_mapping(map, start, len);
if (!em || IS_ERR(em)) {
- spin_unlock(&map->lock);
+ write_unlock(&map->lock);
break;
}
if (test_bit(EXTENT_FLAG_PINNED, &em->flags) ||
em->start != start) {
- spin_unlock(&map->lock);
+ write_unlock(&map->lock);
free_extent_map(em);
break;
}
@@ -2808,7 +2808,7 @@
free_extent_map(em);
}
start = extent_map_end(em);
- spin_unlock(&map->lock);
+ write_unlock(&map->lock);
/* once for us */
free_extent_map(em);
diff --git a/fs/btrfs/extent_map.c b/fs/btrfs/extent_map.c
index 30c9365..72e9fa3 100644
--- a/fs/btrfs/extent_map.c
+++ b/fs/btrfs/extent_map.c
@@ -36,7 +36,7 @@
void extent_map_tree_init(struct extent_map_tree *tree, gfp_t mask)
{
tree->map.rb_node = NULL;
- spin_lock_init(&tree->lock);
+ rwlock_init(&tree->lock);
}
/**
@@ -222,7 +222,6 @@
ret = -EEXIST;
goto out;
}
- assert_spin_locked(&tree->lock);
rb = tree_insert(&tree->map, em->start, &em->rb_node);
if (rb) {
ret = -EEXIST;
@@ -285,7 +284,6 @@
struct rb_node *next = NULL;
u64 end = range_end(start, len);
- assert_spin_locked(&tree->lock);
rb_node = __tree_search(&tree->map, start, &prev, &next);
if (!rb_node && prev) {
em = rb_entry(prev, struct extent_map, rb_node);
@@ -331,7 +329,6 @@
int ret = 0;
WARN_ON(test_bit(EXTENT_FLAG_PINNED, &em->flags));
- assert_spin_locked(&tree->lock);
rb_erase(&em->rb_node, &tree->map);
em->in_tree = 0;
return ret;
diff --git a/fs/btrfs/extent_map.h b/fs/btrfs/extent_map.h
index fb6eeef..6216dfb 100644
--- a/fs/btrfs/extent_map.h
+++ b/fs/btrfs/extent_map.h
@@ -31,7 +31,7 @@
struct extent_map_tree {
struct rb_root map;
- spinlock_t lock;
+ rwlock_t lock;
};
static inline u64 extent_map_end(struct extent_map *em)
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index a760d97..8a9c76a 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -188,15 +188,15 @@
if (!split2)
split2 = alloc_extent_map(GFP_NOFS);
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, len);
if (!em) {
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
break;
}
flags = em->flags;
if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (em->start <= start &&
(!testend || em->start + em->len >= start + len)) {
free_extent_map(em);
@@ -259,7 +259,7 @@
free_extent_map(split);
split = NULL;
}
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
/* once for us */
free_extent_map(em);
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 04b53b5..f1df117 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -612,9 +612,9 @@
set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
while (1) {
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
@@ -748,9 +748,9 @@
set_bit(EXTENT_FLAG_PINNED, &em->flags);
while (1) {
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
@@ -1081,9 +1081,9 @@
em->bdev = root->fs_info->fs_devices->latest_bdev;
set_bit(EXTENT_FLAG_PINNED, &em->flags);
while (1) {
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
@@ -1670,13 +1670,13 @@
failrec->last_mirror = 0;
failrec->bio_flags = 0;
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, failrec->len);
if (em->start > start || em->start + em->len < start) {
free_extent_map(em);
em = NULL;
}
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
if (!em || IS_ERR(em)) {
kfree(failrec);
@@ -4069,11 +4069,11 @@
int compressed;
again:
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, start, len);
if (em)
em->bdev = root->fs_info->fs_devices->latest_bdev;
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
if (em) {
if (em->start > start || em->start + em->len <= start)
@@ -4264,7 +4264,7 @@
}
err = 0;
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
/* it is possible that someone inserted the extent into the tree
* while we had the lock dropped. It is also possible that
@@ -4304,7 +4304,7 @@
err = 0;
}
}
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
out:
if (path)
btrfs_free_path(path);
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index c04f7f2..4adab90 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -2646,9 +2646,9 @@
lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
while (1) {
int ret;
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
if (ret != -EEXIST) {
free_extent_map(em);
break;
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index a7e5377..d2358c0 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -1749,9 +1749,9 @@
* step two, delete the device extents and the
* chunk tree entries
*/
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, chunk_offset, 1);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
BUG_ON(em->start > chunk_offset ||
em->start + em->len < chunk_offset);
@@ -1780,9 +1780,9 @@
ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
BUG_ON(ret);
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
remove_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
kfree(map);
em->bdev = NULL;
@@ -2294,9 +2294,9 @@
em->block_len = em->len;
em_tree = &extent_root->fs_info->mapping_tree.map_tree;
- spin_lock(&em_tree->lock);
+ write_lock(&em_tree->lock);
ret = add_extent_mapping(em_tree, em);
- spin_unlock(&em_tree->lock);
+ write_unlock(&em_tree->lock);
BUG_ON(ret);
free_extent_map(em);
@@ -2491,9 +2491,9 @@
int readonly = 0;
int i;
- spin_lock(&map_tree->map_tree.lock);
+ read_lock(&map_tree->map_tree.lock);
em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
- spin_unlock(&map_tree->map_tree.lock);
+ read_unlock(&map_tree->map_tree.lock);
if (!em)
return 1;
@@ -2518,11 +2518,11 @@
struct extent_map *em;
while (1) {
- spin_lock(&tree->map_tree.lock);
+ write_lock(&tree->map_tree.lock);
em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
if (em)
remove_extent_mapping(&tree->map_tree, em);
- spin_unlock(&tree->map_tree.lock);
+ write_unlock(&tree->map_tree.lock);
if (!em)
break;
kfree(em->bdev);
@@ -2540,9 +2540,9 @@
struct extent_map_tree *em_tree = &map_tree->map_tree;
int ret;
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, logical, len);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
BUG_ON(!em);
BUG_ON(em->start > logical || em->start + em->len < logical);
@@ -2604,9 +2604,9 @@
atomic_set(&multi->error, 0);
}
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, logical, *length);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
if (!em && unplug_page)
return 0;
@@ -2763,9 +2763,9 @@
u64 stripe_nr;
int i, j, nr = 0;
- spin_lock(&em_tree->lock);
+ read_lock(&em_tree->lock);
em = lookup_extent_mapping(em_tree, chunk_start, 1);
- spin_unlock(&em_tree->lock);
+ read_unlock(&em_tree->lock);
BUG_ON(!em || em->start != chunk_start);
map = (struct map_lookup *)em->bdev;
@@ -3053,9 +3053,9 @@
logical = key->offset;
length = btrfs_chunk_length(leaf, chunk);
- spin_lock(&map_tree->map_tree.lock);
+ read_lock(&map_tree->map_tree.lock);
em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
- spin_unlock(&map_tree->map_tree.lock);
+ read_unlock(&map_tree->map_tree.lock);
/* already mapped? */
if (em && em->start <= logical && em->start + em->len > logical) {
@@ -3114,9 +3114,9 @@
map->stripes[i].dev->in_fs_metadata = 1;
}
- spin_lock(&map_tree->map_tree.lock);
+ write_lock(&map_tree->map_tree.lock);
ret = add_extent_mapping(&map_tree->map_tree, em);
- spin_unlock(&map_tree->map_tree.lock);
+ write_unlock(&map_tree->map_tree.lock);
BUG_ON(ret);
free_extent_map(em);