Btrfs: More throttle tuning
* Make walk_down_tree wake up throttled tasks more often
* Make walk_down_tree call cond_resched during long loops
* As the size of the ref cache grows, wait longer in throttle
* Get rid of the reada code in walk_down_tree, the leaves don't get
read anymore, thanks to the ref cache.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index d2d1cc8..da9dda4 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -188,13 +188,6 @@
btrfs_csum_final(crc, result);
if (verify) {
- int from_this_trans = 0;
-
- if (root->fs_info->running_transaction &&
- btrfs_header_generation(buf) ==
- root->fs_info->running_transaction->transid)
- from_this_trans = 1;
-
/* FIXME, this is not good */
if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
u32 val;
@@ -203,11 +196,9 @@
read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
printk("btrfs: %s checksum verify failed on %llu "
- "wanted %X found %X from_this_trans %d "
- "level %d\n",
+ "wanted %X found %X level %d\n",
root->fs_info->sb->s_id,
- buf->start, val, found, from_this_trans,
- btrfs_header_level(buf));
+ buf->start, val, found, btrfs_header_level(buf));
return 1;
}
} else {
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 98a1c0f..1aeb695 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2369,6 +2369,11 @@
leaf_owner, leaf_generation,
key.objectid, key.offset, 0);
mutex_unlock(&root->fs_info->alloc_mutex);
+
+ atomic_inc(&root->fs_info->throttle_gen);
+ wake_up(&root->fs_info->transaction_throttle);
+ cond_resched();
+
BUG_ON(ret);
}
return 0;
@@ -2389,6 +2394,11 @@
ref->owner, ref->generation,
info->objectid, info->offset, 0);
mutex_unlock(&root->fs_info->alloc_mutex);
+
+ atomic_inc(&root->fs_info->throttle_gen);
+ wake_up(&root->fs_info->transaction_throttle);
+ cond_resched();
+
BUG_ON(ret);
info++;
}
@@ -2396,51 +2406,6 @@
return 0;
}
-static void noinline reada_walk_down(struct btrfs_root *root,
- struct extent_buffer *node,
- int slot)
-{
- u64 bytenr;
- u64 last = 0;
- u32 nritems;
- u32 refs;
- u32 blocksize;
- int ret;
- int i;
- int level;
- int skipped = 0;
-
- nritems = btrfs_header_nritems(node);
- level = btrfs_header_level(node);
- if (level)
- return;
-
- for (i = slot; i < nritems && skipped < 32; i++) {
- bytenr = btrfs_node_blockptr(node, i);
- if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
- (last > bytenr && last - bytenr > 32 * 1024))) {
- skipped++;
- continue;
- }
- blocksize = btrfs_level_size(root, level - 1);
- if (i != slot) {
- ret = lookup_extent_ref(NULL, root, bytenr,
- blocksize, &refs);
- BUG_ON(ret);
- if (refs != 1) {
- skipped++;
- continue;
- }
- }
- ret = readahead_tree_block(root, bytenr, blocksize,
- btrfs_node_ptr_generation(node, i));
- last = bytenr + blocksize;
- cond_resched();
- if (ret)
- break;
- }
-}
-
int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
u32 *refs)
{
@@ -2549,6 +2514,7 @@
atomic_inc(&root->fs_info->throttle_gen);
wake_up(&root->fs_info->transaction_throttle);
+ cond_resched();
continue;
}
@@ -2578,8 +2544,6 @@
if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
free_extent_buffer(next);
- if (path->slots[*level] == 0)
- reada_walk_down(root, cur, path->slots[*level]);
next = read_tree_block(root, bytenr, blocksize,
ptr_gen);
cond_resched();
@@ -2601,6 +2565,7 @@
path->nodes[*level-1] = next;
*level = btrfs_header_level(next);
path->slots[*level] = 0;
+ cond_resched();
}
out:
WARN_ON(*level < 0);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 9d84daf..cf73342 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -210,7 +210,9 @@
static void throttle_on_drops(struct btrfs_root *root)
{
struct btrfs_fs_info *info = root->fs_info;
+ int harder_count = 0;
+harder:
if (atomic_read(&info->throttles)) {
DEFINE_WAIT(wait);
int thr;
@@ -226,6 +228,19 @@
schedule();
finish_wait(&info->transaction_throttle, &wait);
} while (thr == atomic_read(&info->throttle_gen));
+ harder_count++;
+
+ if (root->fs_info->total_ref_cache_size > 1 * 1024 * 1024 &&
+ harder_count < 2)
+ goto harder;
+
+ if (root->fs_info->total_ref_cache_size > 5 * 1024 * 1024 &&
+ harder_count < 10)
+ goto harder;
+
+ if (root->fs_info->total_ref_cache_size > 10 * 1024 * 1024 &&
+ harder_count < 20)
+ goto harder;
}
}