* [PATCH 1/4] update the allocator
@ 2008-09-25 8:50 Zheng Yan
0 siblings, 0 replies; only message in thread
From: Zheng Yan @ 2008-09-25 8:50 UTC (permalink / raw)
To: linux-btrfs, chris.mason
Hello,
This patch updates the allocator for the new space balancing
code. It adds reserved extents accounting, changes a few ENOSPC
check, and makes btrfs_find_block_group not return NULL for
volumes that have been ebalanced.
Regards
Yan Zheng
---
diff -r 4b327f4fcee0 ctree.h
--- a/ctree.h Tue Sep 23 13:14:14 2008 -0400
+++ b/ctree.h Thu Sep 25 16:00:18 2008 +0800
@@ -498,6 +498,7 @@
u64 total_bytes;
u64 bytes_used;
u64 bytes_pinned;
+ u64 bytes_reserved;
int full;
int force_alloc;
struct list_head list;
@@ -519,6 +520,7 @@
struct btrfs_block_group_item item;
spinlock_t lock;
u64 pinned;
+ u64 reserved;
u64 flags;
int cached;
int ro;
diff -r 4b327f4fcee0 extent-tree.c
--- a/extent-tree.c Tue Sep 23 13:14:14 2008 -0400
+++ b/extent-tree.c Thu Sep 25 16:00:18 2008 +0800
@@ -325,12 +325,9 @@
struct btrfs_block_group_cache *cache = *cache_ret;
struct btrfs_free_space *info = NULL;
u64 last;
- u64 total_fs_bytes;
u64 search_start = *start_ret;
WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
- total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
-
if (!cache)
goto out;
@@ -354,7 +351,7 @@
last = cache->key.objectid + cache->key.offset;
cache = btrfs_lookup_first_block_group(root->fs_info, last);
- if (!cache || cache->key.objectid >= total_fs_bytes)
+ if (!cache)
goto out;
*cache_ret = cache;
@@ -385,7 +382,6 @@
return found;
}
return NULL;
-
}
static struct btrfs_block_group_cache *
@@ -396,7 +392,6 @@
struct btrfs_block_group_cache *cache;
struct btrfs_block_group_cache *found_group = NULL;
struct btrfs_fs_info *info = root->fs_info;
- struct btrfs_space_info *sinfo;
u64 used;
u64 last = 0;
u64 free_check;
@@ -413,7 +408,7 @@
if (shint && block_group_bits(shint, data) && !shint->ro) {
spin_lock(&shint->lock);
used = btrfs_block_group_used(&shint->item);
- if (used + shint->pinned <
+ if (used + shint->pinned + shint->reserved <
div_factor(shint->key.offset, factor)) {
spin_unlock(&shint->lock);
return shint;
@@ -424,7 +419,7 @@
if (hint && !hint->ro && block_group_bits(hint, data)) {
spin_lock(&hint->lock);
used = btrfs_block_group_used(&hint->item);
- if (used + hint->pinned <
+ if (used + hint->pinned + hint->reserved <
div_factor(hint->key.offset, factor)) {
spin_unlock(&hint->lock);
return hint;
@@ -437,27 +432,9 @@
else
last = search_start;
}
- sinfo = __find_space_info(root->fs_info, data);
- if (!sinfo)
- goto found;
-again:
- while(1) {
- struct list_head *l;
-
- cache = NULL;
-
- spin_lock(&sinfo->lock);
- list_for_each(l, &sinfo->block_groups) {
- struct btrfs_block_group_cache *entry;
- entry = list_entry(l, struct btrfs_block_group_cache,
- list);
- if ((entry->key.objectid >= last) &&
- (!cache || (entry->key.objectid <
- cache->key.objectid)))
- cache = entry;
- }
- spin_unlock(&sinfo->lock);
-
+again:
+ while (1) {
+ cache = btrfs_lookup_first_block_group(root->fs_info, last);
if (!cache)
break;
@@ -467,7 +444,8 @@
if (!cache->ro && block_group_bits(cache, data)) {
free_check = div_factor(cache->key.offset, factor);
- if (used + cache->pinned < free_check) {
+ if (used + cache->pinned + cache->reserved <
+ free_check) {
found_group = cache;
spin_unlock(&cache->lock);
goto found;
@@ -1414,6 +1392,7 @@
if (!cache)
break;
+ cache->dirty = 0;
last += cache->key.offset;
err = write_one_cache_group(trans, root,
@@ -1427,8 +1406,6 @@
werr = err;
continue;
}
-
- cache->dirty = 0;
}
btrfs_free_path(path);
mutex_unlock(&root->fs_info->alloc_mutex);
@@ -1460,6 +1437,7 @@
found->total_bytes = total_bytes;
found->bytes_used = bytes_used;
found->bytes_pinned = 0;
+ found->bytes_reserved = 0;
found->full = 0;
found->force_alloc = 0;
*space_info = found;
@@ -1539,8 +1517,8 @@
thresh = div_factor(space_info->total_bytes, 6);
if (!force &&
- (space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
- thresh)
+ (space_info->bytes_used + space_info->bytes_pinned +
+ space_info->bytes_reserved + alloc_bytes) < thresh)
goto out;
mutex_lock(&extent_root->fs_info->chunk_mutex);
@@ -1621,7 +1599,6 @@
return cache->key.objectid;
}
-
int btrfs_update_pinned_extents(struct btrfs_root *root,
u64 bytenr, u64 num, int pin)
{
@@ -1639,30 +1616,51 @@
}
while (num > 0) {
cache = btrfs_lookup_block_group(fs_info, bytenr);
- if (!cache) {
- u64 first = first_logical_byte(root, bytenr);
- WARN_ON(first < bytenr);
- len = min(first - bytenr, num);
- } else {
- len = min(num, cache->key.offset -
- (bytenr - cache->key.objectid));
- }
+ BUG_ON(!cache);
+ len = min(num, cache->key.offset -
+ (bytenr - cache->key.objectid));
if (pin) {
- if (cache) {
- spin_lock(&cache->lock);
- cache->pinned += len;
- cache->space_info->bytes_pinned += len;
- spin_unlock(&cache->lock);
- }
+ spin_lock(&cache->lock);
+ cache->pinned += len;
+ cache->space_info->bytes_pinned += len;
+ spin_unlock(&cache->lock);
fs_info->total_pinned += len;
} else {
- if (cache) {
- spin_lock(&cache->lock);
- cache->pinned -= len;
- cache->space_info->bytes_pinned -= len;
- spin_unlock(&cache->lock);
- }
+ spin_lock(&cache->lock);
+ cache->pinned -= len;
+ cache->space_info->bytes_pinned -= len;
+ spin_unlock(&cache->lock);
fs_info->total_pinned -= len;
+ }
+ bytenr += len;
+ num -= len;
+ }
+ return 0;
+}
+
+static int update_reserved_extents(struct btrfs_root *root,
+ u64 bytenr, u64 num, int reserve)
+{
+ u64 len;
+ struct btrfs_block_group_cache *cache;
+ struct btrfs_fs_info *fs_info = root->fs_info;
+
+ WARN_ON(!mutex_is_locked(&root->fs_info->alloc_mutex));
+ while (num > 0) {
+ cache = btrfs_lookup_block_group(fs_info, bytenr);
+ BUG_ON(!cache);
+ len = min(num, cache->key.offset -
+ (bytenr - cache->key.objectid));
+ if (reserve) {
+ spin_lock(&cache->lock);
+ cache->reserved += len;
+ cache->space_info->bytes_reserved += len;
+ spin_unlock(&cache->lock);
+ } else {
+ spin_lock(&cache->lock);
+ cache->reserved -= len;
+ cache->space_info->bytes_reserved -= len;
+ spin_unlock(&cache->lock);
}
bytenr += len;
num -= len;
@@ -2126,6 +2124,7 @@
cache = btrfs_lookup_block_group(root->fs_info, bytenr);
BUG_ON(!cache);
btrfs_add_free_space(cache, bytenr, num_bytes);
+ update_reserved_extents(root, bytenr, num_bytes, 0);
return 0;
}
pin = 1;
@@ -2225,14 +2224,11 @@
search_start = max(search_start, first_logical_byte(root, 0));
orig_search_start = search_start;
- if (search_end == (u64)-1)
- search_end = btrfs_super_total_bytes(&info->super_copy);
-
search_start = max(search_start, hint_byte);
total_needed += empty_size;
new_group:
- block_group = btrfs_lookup_block_group(info, search_start);
+ block_group = btrfs_lookup_first_block_group(info, search_start);
/*
* Ok this looks a little tricky, buts its really simple. First if we
@@ -2257,12 +2253,8 @@
ret = do_chunk_alloc(trans, root,
num_bytes + 2 * 1024 * 1024,
data, 1);
- if (ret < 0) {
- struct btrfs_space_info *info;
-
- info = __find_space_info(root->fs_info, data);
+ if (ret < 0)
goto error;
- }
BUG_ON(ret);
chunk_alloc_done = 1;
search_start = orig_search_start;
@@ -2378,22 +2370,24 @@
struct list_head *l;
printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
- info->total_bytes - info->bytes_used - info->bytes_pinned,
- (info->full) ? "" : "not ");
+ info->total_bytes - info->bytes_used - info->bytes_pinned -
+ info->bytes_reserved, (info->full) ? "" : "not ");
spin_lock(&info->lock);
list_for_each(l, &info->block_groups) {
cache = list_entry(l, struct btrfs_block_group_cache, list);
spin_lock(&cache->lock);
printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
- "%Lu pinned\n",
+ "%Lu pinned %Lu reserved\n",
cache->key.objectid, cache->key.offset,
- btrfs_block_group_used(&cache->item), cache->pinned);
+ btrfs_block_group_used(&cache->item),
+ cache->pinned, cache->reserved);
btrfs_dump_free_space(cache, bytes);
spin_unlock(&cache->lock);
}
spin_unlock(&info->lock);
}
+
static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
u64 num_bytes, u64 min_alloc_size,
@@ -2500,6 +2494,7 @@
ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
empty_size, hint_byte, search_end, ins,
data);
+ update_reserved_extents(root, ins->objectid, ins->offset, 1);
maybe_unlock_mutex(root);
return ret;
}
@@ -2625,6 +2620,7 @@
ret = __btrfs_alloc_reserved_extent(trans, root, parent,
root_objectid, ref_generation,
owner, owner_offset, ins);
+ update_reserved_extents(root, ins->objectid, ins->offset, 0);
maybe_unlock_mutex(root);
return ret;
}
@@ -2685,6 +2681,8 @@
owner_objectid, owner_offset, ins);
BUG_ON(ret);
+ } else {
+ update_reserved_extents(root, ins->objectid, ins->offset, 1);
}
maybe_unlock_mutex(root);
return ret;
@@ -3970,10 +3968,6 @@
ret = btrfs_add_block_group_cache(root->fs_info, cache);
BUG_ON(ret);
-
- if (key.objectid >=
- btrfs_super_total_bytes(&info->super_copy))
- break;
}
ret = 0;
error:
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2008-09-25 8:50 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-09-25 8:50 [PATCH 1/4] update the allocator Zheng Yan
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.