struct extent_buffer **must_clean);
static int find_next_key(struct btrfs_path *path, int level,
struct btrfs_key *key);
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
+ int dump_block_groups);
static noinline int
block_group_cache_done(struct btrfs_block_group_cache *cache)
int stripe_len;
int i, nr, ret;
+ if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
+ stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
+ cache->bytes_super += stripe_len;
+ ret = add_excluded_extent(root, cache->key.objectid,
+ stripe_len);
+ BUG_ON(ret);
+ }
+
for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
bytenr = btrfs_sb_offset(i);
ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
BUG_ON(ret);
while (nr--) {
+ cache->bytes_super += stripe_len;
ret = add_excluded_extent(root, logical[nr],
stripe_len);
BUG_ON(ret);
if (ret)
break;
- if (extent_start == start) {
+ if (extent_start <= start) {
start = extent_end + 1;
} else if (extent_start > start && extent_start < end) {
size = extent_start - start;
return -ENOMEM;
exclude_super_stripes(extent_root, block_group);
+ spin_lock(&block_group->space_info->lock);
+ block_group->space_info->bytes_super += block_group->bytes_super;
+ spin_unlock(&block_group->space_info->lock);
last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
return ret;
}
-#ifdef BIO_RW_DISCARD
static void btrfs_issue_discard(struct block_device *bdev,
u64 start, u64 len)
{
- blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
+ blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
+ DISCARD_FL_BARRIER);
}
-#endif
static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
u64 num_bytes)
{
-#ifdef BIO_RW_DISCARD
int ret;
u64 map_length = num_bytes;
struct btrfs_multi_bio *multi = NULL;
+ if (!btrfs_test_opt(root, DISCARD))
+ return 0;
+
/* Tell the block device(s) that the sectors can be discarded */
ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
bytenr, &map_length, &multi, 0);
}
return ret;
-#else
- return 0;
-#endif
}
int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
alloc_target);
}
+static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
+{
+ u64 num_bytes;
+ int level;
+
+ level = BTRFS_MAX_LEVEL - 2;
+ /*
+ * NOTE: these calculations are absolutely the worst possible case.
+ * This assumes that _every_ item we insert will require a new leaf, and
+ * that the tree has grown to its maximum level size.
+ */
+
+ /*
+ * for every item we insert we could insert both an extent item and a
+ * extent ref item. Then for ever item we insert, we will need to cow
+ * both the original leaf, plus the leaf to the left and right of it.
+ *
+ * Unless we are talking about the extent root, then we just want the
+ * number of items * 2, since we just need the extent item plus its ref.
+ */
+ if (root == root->fs_info->extent_root)
+ num_bytes = num_items * 2;
+ else
+ num_bytes = (num_items + (2 * num_items)) * 3;
+
+ /*
+ * num_bytes is total number of leaves we could need times the leaf
+ * size, and then for every leaf we could end up cow'ing 2 nodes per
+ * level, down to the leaf level.
+ */
+ num_bytes = (num_bytes * root->leafsize) +
+ (num_bytes * (level * 2)) * root->nodesize;
+
+ return num_bytes;
+}
+
/*
- * for now this just makes sure we have at least 5% of our metadata space free
- * for use.
+ * Unreserve metadata space for delalloc. If we have less reserved credits than
+ * we have extents, this function does nothing.
*/
-int btrfs_check_metadata_free_space(struct btrfs_root *root)
+int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
+ struct inode *inode, int num_items)
{
struct btrfs_fs_info *info = root->fs_info;
struct btrfs_space_info *meta_sinfo;
- u64 alloc_target, thresh;
- int committed = 0, ret;
+ u64 num_bytes;
+ u64 alloc_target;
+ bool bug = false;
/* get the space info for where the metadata will live */
alloc_target = btrfs_get_alloc_profile(root, 0);
meta_sinfo = __find_space_info(info, alloc_target);
-again:
+ num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+ num_items);
+
spin_lock(&meta_sinfo->lock);
- if (!meta_sinfo->full)
- thresh = meta_sinfo->total_bytes * 80;
- else
- thresh = meta_sinfo->total_bytes * 95;
+ spin_lock(&BTRFS_I(inode)->accounting_lock);
+ if (BTRFS_I(inode)->reserved_extents <=
+ BTRFS_I(inode)->outstanding_extents) {
+ spin_unlock(&BTRFS_I(inode)->accounting_lock);
+ spin_unlock(&meta_sinfo->lock);
+ return 0;
+ }
+ spin_unlock(&BTRFS_I(inode)->accounting_lock);
+
+ BTRFS_I(inode)->reserved_extents--;
+ BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
+
+ if (meta_sinfo->bytes_delalloc < num_bytes) {
+ bug = true;
+ meta_sinfo->bytes_delalloc = 0;
+ } else {
+ meta_sinfo->bytes_delalloc -= num_bytes;
+ }
+ spin_unlock(&meta_sinfo->lock);
+
+ BUG_ON(bug);
+
+ return 0;
+}
+
+static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
+{
+ u64 thresh;
+
+ thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use;
+ thresh = meta_sinfo->total_bytes - thresh;
+ thresh *= 80;
do_div(thresh, 100);
+ if (thresh <= meta_sinfo->bytes_delalloc)
+ meta_sinfo->force_delalloc = 1;
+ else
+ meta_sinfo->force_delalloc = 0;
+}
- if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
- meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
- struct btrfs_trans_handle *trans;
- if (!meta_sinfo->full) {
- meta_sinfo->force_alloc = 1;
- spin_unlock(&meta_sinfo->lock);
+struct async_flush {
+ struct btrfs_root *root;
+ struct btrfs_space_info *info;
+ struct btrfs_work work;
+};
- trans = btrfs_start_transaction(root, 1);
- if (!trans)
- return -ENOMEM;
+static noinline void flush_delalloc_async(struct btrfs_work *work)
+{
+ struct async_flush *async;
+ struct btrfs_root *root;
+ struct btrfs_space_info *info;
- ret = do_chunk_alloc(trans, root->fs_info->extent_root,
- 2 * 1024 * 1024, alloc_target, 0);
- btrfs_end_transaction(trans, root);
+ async = container_of(work, struct async_flush, work);
+ root = async->root;
+ info = async->info;
+
+ btrfs_start_delalloc_inodes(root, 0);
+ wake_up(&info->flush_wait);
+ btrfs_wait_ordered_extents(root, 0, 0);
+
+ spin_lock(&info->lock);
+ info->flushing = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->flush_wait);
+
+ kfree(async);
+}
+
+static void wait_on_flush(struct btrfs_space_info *info)
+{
+ DEFINE_WAIT(wait);
+ u64 used;
+
+ while (1) {
+ prepare_to_wait(&info->flush_wait, &wait,
+ TASK_UNINTERRUPTIBLE);
+ spin_lock(&info->lock);
+ if (!info->flushing) {
+ spin_unlock(&info->lock);
+ break;
+ }
+
+ used = info->bytes_used + info->bytes_reserved +
+ info->bytes_pinned + info->bytes_readonly +
+ info->bytes_super + info->bytes_root +
+ info->bytes_may_use + info->bytes_delalloc;
+ if (used < info->total_bytes) {
+ spin_unlock(&info->lock);
+ break;
+ }
+ spin_unlock(&info->lock);
+ schedule();
+ }
+ finish_wait(&info->flush_wait, &wait);
+}
+
+static void flush_delalloc(struct btrfs_root *root,
+ struct btrfs_space_info *info)
+{
+ struct async_flush *async;
+ bool wait = false;
+
+ spin_lock(&info->lock);
+
+ if (!info->flushing) {
+ info->flushing = 1;
+ init_waitqueue_head(&info->flush_wait);
+ } else {
+ wait = true;
+ }
+
+ spin_unlock(&info->lock);
+
+ if (wait) {
+ wait_on_flush(info);
+ return;
+ }
+
+ async = kzalloc(sizeof(*async), GFP_NOFS);
+ if (!async)
+ goto flush;
+
+ async->root = root;
+ async->info = info;
+ async->work.func = flush_delalloc_async;
+
+ btrfs_queue_worker(&root->fs_info->enospc_workers,
+ &async->work);
+ wait_on_flush(info);
+ return;
+
+flush:
+ btrfs_start_delalloc_inodes(root, 0);
+ btrfs_wait_ordered_extents(root, 0, 0);
+
+ spin_lock(&info->lock);
+ info->flushing = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->flush_wait);
+}
+
+static int maybe_allocate_chunk(struct btrfs_root *root,
+ struct btrfs_space_info *info)
+{
+ struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
+ struct btrfs_trans_handle *trans;
+ bool wait = false;
+ int ret = 0;
+ u64 min_metadata;
+ u64 free_space;
+
+ free_space = btrfs_super_total_bytes(disk_super);
+ /*
+ * we allow the metadata to grow to a max of either 10gb or 5% of the
+ * space in the volume.
+ */
+ min_metadata = min((u64)10 * 1024 * 1024 * 1024,
+ div64_u64(free_space * 5, 100));
+ if (info->total_bytes >= min_metadata) {
+ spin_unlock(&info->lock);
+ return 0;
+ }
+
+ if (info->full) {
+ spin_unlock(&info->lock);
+ return 0;
+ }
+
+ if (!info->allocating_chunk) {
+ info->force_alloc = 1;
+ info->allocating_chunk = 1;
+ init_waitqueue_head(&info->allocate_wait);
+ } else {
+ wait = true;
+ }
+
+ spin_unlock(&info->lock);
+
+ if (wait) {
+ wait_event(info->allocate_wait,
+ !info->allocating_chunk);
+ return 1;
+ }
+
+ trans = btrfs_start_transaction(root, 1);
+ if (!trans) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ ret = do_chunk_alloc(trans, root->fs_info->extent_root,
+ 4096 + 2 * 1024 * 1024,
+ info->flags, 0);
+ btrfs_end_transaction(trans, root);
+ if (ret)
+ goto out;
+out:
+ spin_lock(&info->lock);
+ info->allocating_chunk = 0;
+ spin_unlock(&info->lock);
+ wake_up(&info->allocate_wait);
+
+ if (ret)
+ return 0;
+ return 1;
+}
+
+/*
+ * Reserve metadata space for delalloc.
+ */
+int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
+ struct inode *inode, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 used;
+ u64 alloc_target;
+ int flushed = 0;
+ int force_delalloc;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
+ num_items);
+again:
+ spin_lock(&meta_sinfo->lock);
+
+ force_delalloc = meta_sinfo->force_delalloc;
+
+ if (unlikely(!meta_sinfo->bytes_root))
+ meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+ if (!flushed)
+ meta_sinfo->bytes_delalloc += num_bytes;
+
+ used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+ if (used > meta_sinfo->total_bytes) {
+ flushed++;
+
+ if (flushed == 1) {
+ if (maybe_allocate_chunk(root, meta_sinfo))
+ goto again;
+ flushed++;
+ } else {
+ spin_unlock(&meta_sinfo->lock);
+ }
+
+ if (flushed == 2) {
+ filemap_flush(inode->i_mapping);
+ goto again;
+ } else if (flushed == 3) {
+ flush_delalloc(root, meta_sinfo);
goto again;
}
+ spin_lock(&meta_sinfo->lock);
+ meta_sinfo->bytes_delalloc -= num_bytes;
spin_unlock(&meta_sinfo->lock);
+ printk(KERN_ERR "enospc, has %d, reserved %d\n",
+ BTRFS_I(inode)->outstanding_extents,
+ BTRFS_I(inode)->reserved_extents);
+ dump_space_info(meta_sinfo, 0, 0);
+ return -ENOSPC;
+ }
- if (!committed) {
- committed = 1;
- trans = btrfs_join_transaction(root, 1);
- if (!trans)
- return -ENOMEM;
- ret = btrfs_commit_transaction(trans, root);
- if (ret)
- return ret;
+ BTRFS_I(inode)->reserved_extents++;
+ check_force_delalloc(meta_sinfo);
+ spin_unlock(&meta_sinfo->lock);
+
+ if (!flushed && force_delalloc)
+ filemap_flush(inode->i_mapping);
+
+ return 0;
+}
+
+/*
+ * unreserve num_items number of items worth of metadata space. This needs to
+ * be paired with btrfs_reserve_metadata_space.
+ *
+ * NOTE: if you have the option, run this _AFTER_ you do a
+ * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
+ * oprations which will result in more used metadata, so we want to make sure we
+ * can do that without issue.
+ */
+int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 alloc_target;
+ bool bug = false;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root, num_items);
+
+ spin_lock(&meta_sinfo->lock);
+ if (meta_sinfo->bytes_may_use < num_bytes) {
+ bug = true;
+ meta_sinfo->bytes_may_use = 0;
+ } else {
+ meta_sinfo->bytes_may_use -= num_bytes;
+ }
+ spin_unlock(&meta_sinfo->lock);
+
+ BUG_ON(bug);
+
+ return 0;
+}
+
+/*
+ * Reserve some metadata space for use. We'll calculate the worste case number
+ * of bytes that would be needed to modify num_items number of items. If we
+ * have space, fantastic, if not, you get -ENOSPC. Please call
+ * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
+ * items you reserved, since whatever metadata you needed should have already
+ * been allocated.
+ *
+ * This will commit the transaction to make more space if we don't have enough
+ * metadata space. THe only time we don't do this is if we're reserving space
+ * inside of a transaction, then we will just return -ENOSPC and it is the
+ * callers responsibility to handle it properly.
+ */
+int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
+{
+ struct btrfs_fs_info *info = root->fs_info;
+ struct btrfs_space_info *meta_sinfo;
+ u64 num_bytes;
+ u64 used;
+ u64 alloc_target;
+ int retries = 0;
+
+ /* get the space info for where the metadata will live */
+ alloc_target = btrfs_get_alloc_profile(root, 0);
+ meta_sinfo = __find_space_info(info, alloc_target);
+
+ num_bytes = calculate_bytes_needed(root, num_items);
+again:
+ spin_lock(&meta_sinfo->lock);
+
+ if (unlikely(!meta_sinfo->bytes_root))
+ meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
+
+ if (!retries)
+ meta_sinfo->bytes_may_use += num_bytes;
+
+ used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
+ meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
+ meta_sinfo->bytes_super + meta_sinfo->bytes_root +
+ meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
+
+ if (used > meta_sinfo->total_bytes) {
+ retries++;
+ if (retries == 1) {
+ if (maybe_allocate_chunk(root, meta_sinfo))
+ goto again;
+ retries++;
+ } else {
+ spin_unlock(&meta_sinfo->lock);
+ }
+
+ if (retries == 2) {
+ flush_delalloc(root, meta_sinfo);
goto again;
}
+ spin_lock(&meta_sinfo->lock);
+ meta_sinfo->bytes_may_use -= num_bytes;
+ spin_unlock(&meta_sinfo->lock);
+
+ dump_space_info(meta_sinfo, 0, 0);
return -ENOSPC;
}
+
+ check_force_delalloc(meta_sinfo);
spin_unlock(&meta_sinfo->lock);
return 0;
bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
data_sinfo = BTRFS_I(inode)->space_info;
+ if (!data_sinfo)
+ goto alloc;
+
again:
/* make sure we have enough space to handle the data first */
spin_lock(&data_sinfo->lock);
if (data_sinfo->total_bytes - data_sinfo->bytes_used -
data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
- data_sinfo->bytes_may_use < bytes) {
+ data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
struct btrfs_trans_handle *trans;
/*
data_sinfo->force_alloc = 1;
spin_unlock(&data_sinfo->lock);
-
+alloc:
alloc_target = btrfs_get_alloc_profile(root, 1);
trans = btrfs_start_transaction(root, 1);
if (!trans)
btrfs_end_transaction(trans, root);
if (ret)
return ret;
+
+ if (!data_sinfo) {
+ btrfs_set_inode_space_info(root, inode);
+ data_sinfo = BTRFS_I(inode)->space_info;
+ }
goto again;
}
spin_unlock(&data_sinfo->lock);
/* commit the current transaction and try again */
- if (!committed) {
+ if (!committed && !root->fs_info->open_ioctl_trans) {
committed = 1;
trans = btrfs_join_transaction(root, 1);
if (!trans)
BTRFS_I(inode)->reserved_bytes += bytes;
spin_unlock(&data_sinfo->lock);
- return btrfs_check_metadata_free_space(root);
+ return 0;
}
/*
BUG_ON(!space_info);
spin_lock(&space_info->lock);
- if (space_info->force_alloc) {
+ if (space_info->force_alloc)
force = 1;
- space_info->force_alloc = 0;
- }
if (space_info->full) {
spin_unlock(&space_info->lock);
goto out;
}
thresh = space_info->total_bytes - space_info->bytes_readonly;
- thresh = div_factor(thresh, 6);
+ thresh = div_factor(thresh, 8);
if (!force &&
(space_info->bytes_used + space_info->bytes_pinned +
space_info->bytes_reserved + alloc_bytes) < thresh) {
* we keep a reasonable number of metadata chunks allocated in the
* FS as well.
*/
- if (flags & BTRFS_BLOCK_GROUP_DATA) {
+ if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
fs_info->data_chunk_allocations++;
if (!(fs_info->data_chunk_allocations %
fs_info->metadata_ratio))
}
ret = btrfs_alloc_chunk(trans, extent_root, flags);
+ spin_lock(&space_info->lock);
if (ret)
space_info->full = 1;
+ space_info->force_alloc = 0;
+ spin_unlock(&space_info->lock);
out:
mutex_unlock(&extent_root->fs_info->chunk_mutex);
return ret;
else
old_val -= num_bytes;
btrfs_set_super_bytes_used(&info->super_copy, old_val);
-
- /* block accounting for root item */
- old_val = btrfs_root_used(&root->root_item);
- if (alloc)
- old_val += num_bytes;
- else
- old_val -= num_bytes;
- btrfs_set_root_used(&root->root_item, old_val);
spin_unlock(&info->delalloc_lock);
while (total) {
if (is_data)
goto pinit;
+ /*
+ * discard is sloooow, and so triggering discards on
+ * individual btree blocks isn't a good plan. Just
+ * pin everything in discard mode.
+ */
+ if (btrfs_test_opt(root, DISCARD))
+ goto pinit;
+
buf = btrfs_find_tree_block(root, bytenr, num_bytes);
if (!buf)
goto pinit;
return ret;
}
+int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ u64 bytenr, u32 blocksize,
+ u64 parent, u64 root_objectid, int level)
+{
+ u64 used;
+ spin_lock(&root->node_lock);
+ used = btrfs_root_used(&root->root_item) - blocksize;
+ btrfs_set_root_used(&root->root_item, used);
+ spin_unlock(&root->node_lock);
+
+ return btrfs_free_extent(trans, root, bytenr, blocksize,
+ parent, root_objectid, level, 0);
+}
+
static u64 stripe_align(struct btrfs_root *root, u64 val)
{
u64 mask = ((u64)root->stripesize - 1);
}
enum btrfs_loop_type {
- LOOP_CACHED_ONLY = 0,
+ LOOP_FIND_IDEAL = 0,
LOOP_CACHING_NOWAIT = 1,
LOOP_CACHING_WAIT = 2,
LOOP_ALLOC_CHUNK = 3,
struct btrfs_block_group_cache *block_group = NULL;
int empty_cluster = 2 * 1024 * 1024;
int allowed_chunk_alloc = 0;
+ int done_chunk_alloc = 0;
struct btrfs_space_info *space_info;
int last_ptr_loop = 0;
int loop = 0;
bool found_uncached_bg = false;
+ bool failed_cluster_refill = false;
+ bool failed_alloc = false;
+ u64 ideal_cache_percent = 0;
+ u64 ideal_cache_offset = 0;
WARN_ON(num_bytes < root->sectorsize);
btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
empty_cluster = 0;
if (search_start == hint_byte) {
+ideal_cache:
block_group = btrfs_lookup_block_group(root->fs_info,
search_start);
/*
* we don't want to use the block group if it doesn't match our
* allocation bits, or if its not cached.
+ *
+ * However if we are re-searching with an ideal block group
+ * picked out then we don't care that the block group is cached.
*/
if (block_group && block_group_bits(block_group, data) &&
- block_group_cache_done(block_group)) {
+ (block_group->cached != BTRFS_CACHE_NO ||
+ search_start == ideal_cache_offset)) {
down_read(&space_info->groups_sem);
if (list_empty(&block_group->list) ||
block_group->ro) {
*/
btrfs_put_block_group(block_group);
up_read(&space_info->groups_sem);
- } else
+ } else {
goto have_block_group;
+ }
} else if (block_group) {
btrfs_put_block_group(block_group);
}
}
-
search:
down_read(&space_info->groups_sem);
list_for_each_entry(block_group, &space_info->block_groups, list) {
have_block_group:
if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
+ u64 free_percent;
+
+ free_percent = btrfs_block_group_used(&block_group->item);
+ free_percent *= 100;
+ free_percent = div64_u64(free_percent,
+ block_group->key.offset);
+ free_percent = 100 - free_percent;
+ if (free_percent > ideal_cache_percent &&
+ likely(!block_group->ro)) {
+ ideal_cache_offset = block_group->key.objectid;
+ ideal_cache_percent = free_percent;
+ }
+
/*
- * we want to start caching kthreads, but not too many
- * right off the bat so we don't overwhelm the system,
- * so only start them if there are less than 2 and we're
- * in the initial allocation phase.
+ * We only want to start kthread caching if we are at
+ * the point where we will wait for caching to make
+ * progress, or if our ideal search is over and we've
+ * found somebody to start caching.
*/
if (loop > LOOP_CACHING_NOWAIT ||
- atomic_read(&space_info->caching_threads) < 2) {
+ (loop > LOOP_FIND_IDEAL &&
+ atomic_read(&space_info->caching_threads) < 2)) {
ret = cache_block_group(block_group);
BUG_ON(ret);
}
- }
-
- cached = block_group_cache_done(block_group);
- if (unlikely(!cached)) {
found_uncached_bg = true;
- /* if we only want cached bgs, loop */
- if (loop == LOOP_CACHED_ONLY)
+ /*
+ * If loop is set for cached only, try the next block
+ * group.
+ */
+ if (loop == LOOP_FIND_IDEAL)
goto loop;
}
+ cached = block_group_cache_done(block_group);
+ if (unlikely(!cached))
+ found_uncached_bg = true;
+
if (unlikely(block_group->ro))
goto loop;
- if (last_ptr) {
+ /*
+ * Ok we want to try and use the cluster allocator, so lets look
+ * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
+ * have tried the cluster allocator plenty of times at this
+ * point and not have found anything, so we are likely way too
+ * fragmented for the clustering stuff to find anything, so lets
+ * just skip it and let the allocator find whatever block it can
+ * find
+ */
+ if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
/*
* the refill lock keeps out other
* people trying to start a new cluster
spin_unlock(&last_ptr->refill_lock);
goto checks;
}
- } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
+ } else if (!cached && loop > LOOP_CACHING_NOWAIT
+ && !failed_cluster_refill) {
spin_unlock(&last_ptr->refill_lock);
+ failed_cluster_refill = true;
wait_block_group_cache_progress(block_group,
num_bytes + empty_cluster + empty_size);
goto have_block_group;
* cluster. Free the cluster we've been trying
* to use, and go to the next block group
*/
- if (loop < LOOP_NO_EMPTY_SIZE) {
- btrfs_return_cluster_to_free_space(NULL,
- last_ptr);
- spin_unlock(&last_ptr->refill_lock);
- goto loop;
- }
+ btrfs_return_cluster_to_free_space(NULL, last_ptr);
spin_unlock(&last_ptr->refill_lock);
+ goto loop;
}
offset = btrfs_find_space_for_alloc(block_group, search_start,
num_bytes, empty_size);
- if (!offset && (cached || (!cached &&
- loop == LOOP_CACHING_NOWAIT))) {
- goto loop;
- } else if (!offset && (!cached &&
- loop > LOOP_CACHING_NOWAIT)) {
+ /*
+ * If we didn't find a chunk, and we haven't failed on this
+ * block group before, and this block group is in the middle of
+ * caching and we are ok with waiting, then go ahead and wait
+ * for progress to be made, and set failed_alloc to true.
+ *
+ * If failed_alloc is true then we've already waited on this
+ * block group once and should move on to the next block group.
+ */
+ if (!offset && !failed_alloc && !cached &&
+ loop > LOOP_CACHING_NOWAIT) {
wait_block_group_cache_progress(block_group,
- num_bytes + empty_size);
+ num_bytes + empty_size);
+ failed_alloc = true;
goto have_block_group;
+ } else if (!offset) {
+ goto loop;
}
checks:
search_start = stripe_align(root, offset);
/* we are all good, lets return */
break;
loop:
+ failed_cluster_refill = false;
+ failed_alloc = false;
btrfs_put_block_group(block_group);
}
up_read(&space_info->groups_sem);
- /* LOOP_CACHED_ONLY, only search fully cached block groups
- * LOOP_CACHING_NOWAIT, search partially cached block groups, but
- * dont wait foR them to finish caching
+ /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
+ * for them to make caching progress. Also
+ * determine the best possible bg to cache
+ * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
+ * caching kthreads as we move along
* LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
* LOOP_ALLOC_CHUNK, force a chunk allocation and try again
* LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
(found_uncached_bg || empty_size || empty_cluster ||
allowed_chunk_alloc)) {
- if (found_uncached_bg) {
+ if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
found_uncached_bg = false;
- if (loop < LOOP_CACHING_WAIT) {
- loop++;
+ loop++;
+ if (!ideal_cache_percent &&
+ atomic_read(&space_info->caching_threads))
goto search;
- }
+
+ /*
+ * 1 of the following 2 things have happened so far
+ *
+ * 1) We found an ideal block group for caching that
+ * is mostly full and will cache quickly, so we might
+ * as well wait for it.
+ *
+ * 2) We searched for cached only and we didn't find
+ * anything, and we didn't start any caching kthreads
+ * either, so chances are we will loop through and
+ * start a couple caching kthreads, and then come back
+ * around and just wait for them. This will be slower
+ * because we will have 2 caching kthreads reading at
+ * the same time when we could have just started one
+ * and waited for it to get far enough to give us an
+ * allocation, so go ahead and go to the wait caching
+ * loop.
+ */
+ loop = LOOP_CACHING_WAIT;
+ search_start = ideal_cache_offset;
+ ideal_cache_percent = 0;
+ goto ideal_cache;
+ } else if (loop == LOOP_FIND_IDEAL) {
+ /*
+ * Didn't find a uncached bg, wait on anything we find
+ * next.
+ */
+ loop = LOOP_CACHING_WAIT;
+ goto search;
+ }
+
+ if (loop < LOOP_CACHING_WAIT) {
+ loop++;
+ goto search;
}
if (loop == LOOP_ALLOC_CHUNK) {
ret = do_chunk_alloc(trans, root, num_bytes +
2 * 1024 * 1024, data, 1);
allowed_chunk_alloc = 0;
- } else {
+ done_chunk_alloc = 1;
+ } else if (!done_chunk_alloc) {
space_info->force_alloc = 1;
}
return ret;
}
-static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
+static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
+ int dump_block_groups)
{
struct btrfs_block_group_cache *cache;
+ spin_lock(&info->lock);
printk(KERN_INFO "space_info has %llu free, is %sfull\n",
(unsigned long long)(info->total_bytes - info->bytes_used -
- info->bytes_pinned - info->bytes_reserved),
+ info->bytes_pinned - info->bytes_reserved -
+ info->bytes_super),
(info->full) ? "" : "not ");
printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
- " may_use=%llu, used=%llu\n",
+ " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
+ "\n",
(unsigned long long)info->total_bytes,
(unsigned long long)info->bytes_pinned,
(unsigned long long)info->bytes_delalloc,
(unsigned long long)info->bytes_may_use,
- (unsigned long long)info->bytes_used);
+ (unsigned long long)info->bytes_used,
+ (unsigned long long)info->bytes_root,
+ (unsigned long long)info->bytes_super,
+ (unsigned long long)info->bytes_reserved);
+ spin_unlock(&info->lock);
+
+ if (!dump_block_groups)
+ return;
down_read(&info->groups_sem);
list_for_each_entry(cache, &info->block_groups, list) {
{
int ret;
u64 search_start = 0;
- struct btrfs_fs_info *info = root->fs_info;
data = btrfs_get_alloc_profile(root, data);
again:
* the only place that sets empty_size is btrfs_realloc_node, which
* is not called recursively on allocations
*/
- if (empty_size || root->ref_cows) {
- if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
- ret = do_chunk_alloc(trans, root->fs_info->extent_root,
- 2 * 1024 * 1024,
- BTRFS_BLOCK_GROUP_METADATA |
- (info->metadata_alloc_profile &
- info->avail_metadata_alloc_bits), 0);
- }
+ if (empty_size || root->ref_cows)
ret = do_chunk_alloc(trans, root->fs_info->extent_root,
num_bytes + 2 * 1024 * 1024, data, 0);
- }
WARN_ON(num_bytes < root->sectorsize);
ret = find_free_extent(trans, root, num_bytes, empty_size,
printk(KERN_ERR "btrfs allocation failed flags %llu, "
"wanted %llu\n", (unsigned long long)data,
(unsigned long long)num_bytes);
- dump_space_info(sinfo, num_bytes);
+ dump_space_info(sinfo, num_bytes, 1);
}
return ret;
extent_op);
BUG_ON(ret);
}
+
+ if (root_objectid == root->root_key.objectid) {
+ u64 used;
+ spin_lock(&root->node_lock);
+ used = btrfs_root_used(&root->root_item) + num_bytes;
+ btrfs_set_root_used(&root->root_item, used);
+ spin_unlock(&root->node_lock);
+ }
return ret;
}
btrfs_set_buffer_uptodate(buf);
if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
- set_extent_dirty(&root->dirty_log_pages, buf->start,
- buf->start + buf->len - 1, GFP_NOFS);
+ /*
+ * we allow two log transactions at a time, use different
+ * EXENT bit to differentiate dirty pages.
+ */
+ if (root->log_transid % 2 == 0)
+ set_extent_dirty(&root->dirty_log_pages, buf->start,
+ buf->start + buf->len - 1, GFP_NOFS);
+ else
+ set_extent_new(&root->dirty_log_pages, buf->start,
+ buf->start + buf->len - 1, GFP_NOFS);
} else {
set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
buf->start + buf->len - 1, GFP_NOFS);
return buf;
}
-#if 0
-int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, struct extent_buffer *leaf)
-{
- u64 disk_bytenr;
- u64 num_bytes;
- struct btrfs_key key;
- struct btrfs_file_extent_item *fi;
- u32 nritems;
- int i;
- int ret;
-
- BUG_ON(!btrfs_is_leaf(leaf));
- nritems = btrfs_header_nritems(leaf);
-
- for (i = 0; i < nritems; i++) {
- cond_resched();
- btrfs_item_key_to_cpu(leaf, &key, i);
-
- /* only extents have references, skip everything else */
- if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
- continue;
-
- fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
-
- /* inline extents live in the btree, they don't have refs */
- if (btrfs_file_extent_type(leaf, fi) ==
- BTRFS_FILE_EXTENT_INLINE)
- continue;
-
- disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
-
- /* holes don't have refs */
- if (disk_bytenr == 0)
- continue;
-
- num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
- ret = btrfs_free_extent(trans, root, disk_bytenr, num_bytes,
- leaf->start, 0, key.objectid, 0);
- BUG_ON(ret);
- }
- return 0;
-}
-
-static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
- struct btrfs_root *root,
- struct btrfs_leaf_ref *ref)
-{
- int i;
- int ret;
- struct btrfs_extent_info *info;
- struct refsort *sorted;
-
- if (ref->nritems == 0)
- return 0;
-
- sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
- for (i = 0; i < ref->nritems; i++) {
- sorted[i].bytenr = ref->extents[i].bytenr;
- sorted[i].slot = i;
- }
- sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
-
- /*
- * the items in the ref were sorted when the ref was inserted
- * into the ref cache, so this is already in order
- */
- for (i = 0; i < ref->nritems; i++) {
- info = ref->extents + sorted[i].slot;
- ret = btrfs_free_extent(trans, root, info->bytenr,
- info->num_bytes, ref->bytenr,
- ref->owner, ref->generation,
- info->objectid, 0);
-
- atomic_inc(&root->fs_info->throttle_gen);
- wake_up(&root->fs_info->transaction_throttle);
- cond_resched();
-
- BUG_ON(ret);
- info++;
- }
-
- kfree(sorted);
- return 0;
-}
-
-
-static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, u64 start,
- u64 len, u32 *refs)
-{
- int ret;
-
- ret = btrfs_lookup_extent_refs(trans, root, start, len, refs);
- BUG_ON(ret);
-
-#if 0 /* some debugging code in case we see problems here */
- /* if the refs count is one, it won't get increased again. But
- * if the ref count is > 1, someone may be decreasing it at
- * the same time we are.
- */
- if (*refs != 1) {
- struct extent_buffer *eb = NULL;
- eb = btrfs_find_create_tree_block(root, start, len);
- if (eb)
- btrfs_tree_lock(eb);
-
- mutex_lock(&root->fs_info->alloc_mutex);
- ret = lookup_extent_ref(NULL, root, start, len, refs);
- BUG_ON(ret);
- mutex_unlock(&root->fs_info->alloc_mutex);
-
- if (eb) {
- btrfs_tree_unlock(eb);
- free_extent_buffer(eb);
- }
- if (*refs == 1) {
- printk(KERN_ERR "btrfs block %llu went down to one "
- "during drop_snap\n", (unsigned long long)start);
- }
-
- }
-#endif
-
- cond_resched();
- return ret;
-}
-
-
-/*
- * this is used while deleting old snapshots, and it drops the refs
- * on a whole subtree starting from a level 1 node.
- *
- * The idea is to sort all the leaf pointers, and then drop the
- * ref on all the leaves in order. Most of the time the leaves
- * will have ref cache entries, so no leaf IOs will be required to
- * find the extents they have references on.
- *
- * For each leaf, any references it has are also dropped in order
- *
- * This ends up dropping the references in something close to optimal
- * order for reading and modifying the extent allocation tree.
- */
-static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
- struct btrfs_root *root,
- struct btrfs_path *path)
-{
- u64 bytenr;
- u64 root_owner;
- u64 root_gen;
- struct extent_buffer *eb = path->nodes[1];
- struct extent_buffer *leaf;
- struct btrfs_leaf_ref *ref;
- struct refsort *sorted = NULL;
- int nritems = btrfs_header_nritems(eb);
- int ret;
- int i;
- int refi = 0;
- int slot = path->slots[1];
- u32 blocksize = btrfs_level_size(root, 0);
- u32 refs;
-
- if (nritems == 0)
- goto out;
-
- root_owner = btrfs_header_owner(eb);
- root_gen = btrfs_header_generation(eb);
- sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
-
- /*
- * step one, sort all the leaf pointers so we don't scribble
- * randomly into the extent allocation tree
- */
- for (i = slot; i < nritems; i++) {
- sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
- sorted[refi].slot = i;
- refi++;
- }
-
- /*
- * nritems won't be zero, but if we're picking up drop_snapshot
- * after a crash, slot might be > 0, so double check things
- * just in case.
- */
- if (refi == 0)
- goto out;
-
- sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
-
- /*
- * the first loop frees everything the leaves point to
- */
- for (i = 0; i < refi; i++) {
- u64 ptr_gen;
-
- bytenr = sorted[i].bytenr;
-
- /*
- * check the reference count on this leaf. If it is > 1
- * we just decrement it below and don't update any
- * of the refs the leaf points to.
- */
- ret = drop_snap_lookup_refcount(trans, root, bytenr,
- blocksize, &refs);
- BUG_ON(ret);
- if (refs != 1)
- continue;
-
- ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
-
- /*
- * the leaf only had one reference, which means the
- * only thing pointing to this leaf is the snapshot
- * we're deleting. It isn't possible for the reference
- * count to increase again later
- *
- * The reference cache is checked for the leaf,
- * and if found we'll be able to drop any refs held by
- * the leaf without needing to read it in.
- */
- ref = btrfs_lookup_leaf_ref(root, bytenr);
- if (ref && ref->generation != ptr_gen) {
- btrfs_free_leaf_ref(root, ref);
- ref = NULL;
- }
- if (ref) {
- ret = cache_drop_leaf_ref(trans, root, ref);
- BUG_ON(ret);
- btrfs_remove_leaf_ref(root, ref);
- btrfs_free_leaf_ref(root, ref);
- } else {
- /*
- * the leaf wasn't in the reference cache, so
- * we have to read it.
- */
- leaf = read_tree_block(root, bytenr, blocksize,
- ptr_gen);
- ret = btrfs_drop_leaf_ref(trans, root, leaf);
- BUG_ON(ret);
- free_extent_buffer(leaf);
- }
- atomic_inc(&root->fs_info->throttle_gen);
- wake_up(&root->fs_info->transaction_throttle);
- cond_resched();
- }
-
- /*
- * run through the loop again to free the refs on the leaves.
- * This is faster than doing it in the loop above because
- * the leaves are likely to be clustered together. We end up
- * working in nice chunks on the extent allocation tree.
- */
- for (i = 0; i < refi; i++) {
- bytenr = sorted[i].bytenr;
- ret = btrfs_free_extent(trans, root, bytenr,
- blocksize, eb->start,
- root_owner, root_gen, 0, 1);
- BUG_ON(ret);
-
- atomic_inc(&root->fs_info->throttle_gen);
- wake_up(&root->fs_info->transaction_throttle);
- cond_resched();
- }
-out:
- kfree(sorted);
-
- /*
- * update the path to show we've processed the entire level 1
- * node. This will get saved into the root's drop_snapshot_progress
- * field so these drops are not repeated again if this transaction
- * commits.
- */
- path->slots[1] = nritems;
- return 0;
-}
-
-/*
- * helper function for drop_snapshot, this walks down the tree dropping ref
- * counts as it goes.
- */
-static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
- struct btrfs_root *root,
- struct btrfs_path *path, int *level)
-{
- u64 root_owner;
- u64 root_gen;
- u64 bytenr;
- u64 ptr_gen;
- struct extent_buffer *next;
- struct extent_buffer *cur;
- struct extent_buffer *parent;
- u32 blocksize;
- int ret;
- u32 refs;
-
- WARN_ON(*level < 0);
- WARN_ON(*level >= BTRFS_MAX_LEVEL);
- ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
- path->nodes[*level]->len, &refs);
- BUG_ON(ret);
- if (refs > 1)
- goto out;
-
- /*
- * walk down to the last node level and free all the leaves
- */
- while (*level >= 0) {
- WARN_ON(*level < 0);
- WARN_ON(*level >= BTRFS_MAX_LEVEL);
- cur = path->nodes[*level];
-
- if (btrfs_header_level(cur) != *level)
- WARN_ON(1);
-
- if (path->slots[*level] >=
- btrfs_header_nritems(cur))
- break;
-
- /* the new code goes down to level 1 and does all the
- * leaves pointed to that node in bulk. So, this check
- * for level 0 will always be false.
- *
- * But, the disk format allows the drop_snapshot_progress
- * field in the root to leave things in a state where
- * a leaf will need cleaning up here. If someone crashes
- * with the old code and then boots with the new code,
- * we might find a leaf here.
- */
- if (*level == 0) {
- ret = btrfs_drop_leaf_ref(trans, root, cur);
- BUG_ON(ret);
- break;
- }
-
- /*
- * once we get to level one, process the whole node
- * at once, including everything below it.
- */
- if (*level == 1) {
- ret = drop_level_one_refs(trans, root, path);
- BUG_ON(ret);
- break;
- }
-
- bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
- ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
- blocksize = btrfs_level_size(root, *level - 1);
-
- ret = drop_snap_lookup_refcount(trans, root, bytenr,
- blocksize, &refs);
- BUG_ON(ret);
-
- /*
- * if there is more than one reference, we don't need
- * to read that node to drop any references it has. We
- * just drop the ref we hold on that node and move on to the
- * next slot in this level.
- */
- if (refs != 1) {
- parent = path->nodes[*level];
- root_owner = btrfs_header_owner(parent);
- root_gen = btrfs_header_generation(parent);
- path->slots[*level]++;
-
- ret = btrfs_free_extent(trans, root, bytenr,
- blocksize, parent->start,
- root_owner, root_gen,
- *level - 1, 1);
- BUG_ON(ret);
-
- atomic_inc(&root->fs_info->throttle_gen);
- wake_up(&root->fs_info->transaction_throttle);
- cond_resched();
-
- continue;
- }
-
- /*
- * we need to keep freeing things in the next level down.
- * read the block and loop around to process it
- */
- next = read_tree_block(root, bytenr, blocksize, ptr_gen);
- WARN_ON(*level <= 0);
- if (path->nodes[*level-1])
- free_extent_buffer(path->nodes[*level-1]);
- path->nodes[*level-1] = next;
- *level = btrfs_header_level(next);
- path->slots[*level] = 0;
- cond_resched();
- }
-out:
- WARN_ON(*level < 0);
- WARN_ON(*level >= BTRFS_MAX_LEVEL);
-
- if (path->nodes[*level] == root->node) {
- parent = path->nodes[*level];
- bytenr = path->nodes[*level]->start;
- } else {
- parent = path->nodes[*level + 1];
- bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
- }
-
- blocksize = btrfs_level_size(root, *level);
- root_owner = btrfs_header_owner(parent);
- root_gen = btrfs_header_generation(parent);
-
- /*
- * cleanup and free the reference on the last node
- * we processed
- */
- ret = btrfs_free_extent(trans, root, bytenr, blocksize,
- parent->start, root_owner, root_gen,
- *level, 1);
- free_extent_buffer(path->nodes[*level]);
- path->nodes[*level] = NULL;
-
- *level += 1;
- BUG_ON(ret);
-
- cond_resched();
- return 0;
-}
-#endif
-
struct walk_control {
u64 refs[BTRFS_MAX_LEVEL];
u64 flags[BTRFS_MAX_LEVEL];
u64 bytenr;
u64 generation;
u64 refs;
+ u64 flags;
u64 last = 0;
u32 nritems;
u32 blocksize;
generation <= root->root_key.offset)
continue;
+ /* We don't lock the tree block, it's OK to be racy here */
+ ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+ &refs, &flags);
+ BUG_ON(ret);
+ BUG_ON(refs == 0);
+
if (wc->stage == DROP_REFERENCE) {
- ret = btrfs_lookup_extent_info(trans, root,
- bytenr, blocksize,
- &refs, NULL);
- BUG_ON(ret);
- BUG_ON(refs == 0);
if (refs == 1)
goto reada;
+ if (wc->level == 1 &&
+ (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ continue;
if (!wc->update_ref ||
generation <= root->root_key.offset)
continue;
&wc->update_progress);
if (ret < 0)
continue;
+ } else {
+ if (wc->level == 1 &&
+ (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ continue;
}
reada:
ret = readahead_tree_block(root, bytenr, blocksize,
static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path,
- struct walk_control *wc)
+ struct walk_control *wc, int lookup_info)
{
int level = wc->level;
struct extent_buffer *eb = path->nodes[level];
* when reference count of tree block is 1, it won't increase
* again. once full backref flag is set, we never clear it.
*/
- if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
- (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
+ if (lookup_info &&
+ ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
+ (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
BUG_ON(!path->locks[level]);
ret = btrfs_lookup_extent_info(trans, root,
eb->start, eb->len,
static noinline int do_walk_down(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path,
- struct walk_control *wc)
+ struct walk_control *wc, int *lookup_info)
{
u64 bytenr;
u64 generation;
* for the subtree
*/
if (wc->stage == UPDATE_BACKREF &&
- generation <= root->root_key.offset)
+ generation <= root->root_key.offset) {
+ *lookup_info = 1;
return 1;
+ }
bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
blocksize = btrfs_level_size(root, level - 1);
btrfs_tree_lock(next);
btrfs_set_lock_blocking(next);
- if (wc->stage == DROP_REFERENCE) {
- ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
- &wc->refs[level - 1],
- &wc->flags[level - 1]);
- BUG_ON(ret);
- BUG_ON(wc->refs[level - 1] == 0);
+ ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
+ &wc->refs[level - 1],
+ &wc->flags[level - 1]);
+ BUG_ON(ret);
+ BUG_ON(wc->refs[level - 1] == 0);
+ *lookup_info = 0;
+ if (wc->stage == DROP_REFERENCE) {
if (wc->refs[level - 1] > 1) {
+ if (level == 1 &&
+ (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ goto skip;
+
if (!wc->update_ref ||
generation <= root->root_key.offset)
goto skip;
wc->stage = UPDATE_BACKREF;
wc->shared_level = level - 1;
}
+ } else {
+ if (level == 1 &&
+ (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
+ goto skip;
}
if (!btrfs_buffer_uptodate(next, generation)) {
btrfs_tree_unlock(next);
free_extent_buffer(next);
next = NULL;
+ *lookup_info = 1;
}
if (!next) {
skip:
wc->refs[level - 1] = 0;
wc->flags[level - 1] = 0;
+ if (wc->stage == DROP_REFERENCE) {
+ if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
+ parent = path->nodes[level]->start;
+ } else {
+ BUG_ON(root->root_key.objectid !=
+ btrfs_header_owner(path->nodes[level]));
+ parent = 0;
+ }
- if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
- parent = path->nodes[level]->start;
- } else {
- BUG_ON(root->root_key.objectid !=
- btrfs_header_owner(path->nodes[level]));
- parent = 0;
+ ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
+ root->root_key.objectid, level - 1, 0);
+ BUG_ON(ret);
}
-
- ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
- root->root_key.objectid, level - 1, 0);
- BUG_ON(ret);
-
btrfs_tree_unlock(next);
free_extent_buffer(next);
+ *lookup_info = 1;
return 1;
}
struct walk_control *wc)
{
int level = wc->level;
+ int lookup_info = 1;
int ret;
while (level >= 0) {
btrfs_header_nritems(path->nodes[level]))
break;
- ret = walk_down_proc(trans, root, path, wc);
+ ret = walk_down_proc(trans, root, path, wc, lookup_info);
if (ret > 0)
break;
if (level == 0)
break;
- ret = do_walk_down(trans, root, path, wc);
+ ret = do_walk_down(trans, root, path, wc, &lookup_info);
if (ret > 0) {
path->slots[level]++;
continue;
return 0;
}
-#if 0
-static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
- struct btrfs_root *root,
- u64 objectid, u64 size)
-{
- struct btrfs_path *path;
- struct btrfs_inode_item *item;
- struct extent_buffer *leaf;
- int ret;
-
- path = btrfs_alloc_path();
- if (!path)
- return -ENOMEM;
-
- path->leave_spinning = 1;
- ret = btrfs_insert_empty_inode(trans, root, path, objectid);
- if (ret)
- goto out;
-
- leaf = path->nodes[0];
- item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
- memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
- btrfs_set_inode_generation(leaf, item, 1);
- btrfs_set_inode_size(leaf, item, size);
- btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
- btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
- btrfs_mark_buffer_dirty(leaf);
- btrfs_release_path(root, path);
-out:
- btrfs_free_path(path);
- return ret;
-}
-
-static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
- struct btrfs_block_group_cache *group)
+/*
+ * checks to see if its even possible to relocate this block group.
+ *
+ * @return - -1 if it's not a good idea to relocate this block group, 0 if its
+ * ok to go ahead and try.
+ */
+int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
{
- struct inode *inode = NULL;
- struct btrfs_trans_handle *trans;
- struct btrfs_root *root;
- struct btrfs_key root_key;
- u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
- int err = 0;
+ struct btrfs_block_group_cache *block_group;
+ struct btrfs_space_info *space_info;
+ struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
+ struct btrfs_device *device;
+ int full = 0;
+ int ret = 0;
- root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
- root_key.type = BTRFS_ROOT_ITEM_KEY;
- root_key.offset = (u64)-1;
- root = btrfs_read_fs_root_no_name(fs_info, &root_key);
- if (IS_ERR(root))
- return ERR_CAST(root);
+ block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
- trans = btrfs_start_transaction(root, 1);
- BUG_ON(!trans);
+ /* odd, couldn't find the block group, leave it alone */
+ if (!block_group)
+ return -1;
- err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
- if (err)
+ /* no bytes used, we're good */
+ if (!btrfs_block_group_used(&block_group->item))
goto out;
- err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
- BUG_ON(err);
-
- err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
- group->key.offset, 0, group->key.offset,
- 0, 0, 0);
- BUG_ON(err);
-
- inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
- if (inode->i_state & I_NEW) {
- BTRFS_I(inode)->root = root;
- BTRFS_I(inode)->location.objectid = objectid;
- BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
- BTRFS_I(inode)->location.offset = 0;
- btrfs_read_locked_inode(inode);
- unlock_new_inode(inode);
- BUG_ON(is_bad_inode(inode));
- } else {
- BUG_ON(1);
- }
- BTRFS_I(inode)->index_cnt = group->key.objectid;
-
- err = btrfs_orphan_add(trans, inode);
-out:
- btrfs_end_transaction(trans, root);
- if (err) {
- if (inode)
- iput(inode);
- inode = ERR_PTR(err);
- }
- return inode;
-}
-
-int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
-{
-
- struct btrfs_ordered_sum *sums;
- struct btrfs_sector_sum *sector_sum;
- struct btrfs_ordered_extent *ordered;
- struct btrfs_root *root = BTRFS_I(inode)->root;
- struct list_head list;
- size_t offset;
- int ret;
- u64 disk_bytenr;
-
- INIT_LIST_HEAD(&list);
-
- ordered = btrfs_lookup_ordered_extent(inode, file_pos);
- BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
-
- disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
- ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
- disk_bytenr + len - 1, &list);
-
- while (!list_empty(&list)) {
- sums = list_entry(list.next, struct btrfs_ordered_sum, list);
- list_del_init(&sums->list);
+ space_info = block_group->space_info;
+ spin_lock(&space_info->lock);
- sector_sum = sums->sums;
- sums->bytenr = ordered->start;
+ full = space_info->full;
- offset = 0;
- while (offset < sums->len) {
- sector_sum->bytenr += ordered->start - disk_bytenr;
- sector_sum++;
- offset += root->sectorsize;
- }
-
- btrfs_add_ordered_sum(inode, ordered, sums);
+ /*
+ * if this is the last block group we have in this space, we can't
+ * relocate it unless we're able to allocate a new chunk below.
+ *
+ * Otherwise, we need to make sure we have room in the space to handle
+ * all of the extents from this block group. If we can, we're good
+ */
+ if ((space_info->total_bytes != block_group->key.offset) &&
+ (space_info->bytes_used + space_info->bytes_reserved +
+ space_info->bytes_pinned + space_info->bytes_readonly +
+ btrfs_block_group_used(&block_group->item) <
+ space_info->total_bytes)) {
+ spin_unlock(&space_info->lock);
+ goto out;
}
- btrfs_put_ordered_extent(ordered);
- return 0;
-}
-
-int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
-{
- struct btrfs_trans_handle *trans;
- struct btrfs_path *path;
- struct btrfs_fs_info *info = root->fs_info;
- struct extent_buffer *leaf;
- struct inode *reloc_inode;
- struct btrfs_block_group_cache *block_group;
- struct btrfs_key key;
- u64 skipped;
- u64 cur_byte;
- u64 total_found;
- u32 nritems;
- int ret;
- int progress;
- int pass = 0;
-
- root = root->fs_info->extent_root;
-
- block_group = btrfs_lookup_block_group(info, group_start);
- BUG_ON(!block_group);
-
- printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
- (unsigned long long)block_group->key.objectid,
- (unsigned long long)block_group->flags);
-
- path = btrfs_alloc_path();
- BUG_ON(!path);
-
- reloc_inode = create_reloc_inode(info, block_group);
- BUG_ON(IS_ERR(reloc_inode));
-
- __alloc_chunk_for_shrink(root, block_group, 1);
- set_block_group_readonly(block_group);
-
- btrfs_start_delalloc_inodes(info->tree_root);
- btrfs_wait_ordered_extents(info->tree_root, 0);
-again:
- skipped = 0;
- total_found = 0;
- progress = 0;
- key.objectid = block_group->key.objectid;
- key.offset = 0;
- key.type = 0;
- cur_byte = key.objectid;
-
- trans = btrfs_start_transaction(info->tree_root, 1);
- btrfs_commit_transaction(trans, info->tree_root);
+ spin_unlock(&space_info->lock);
- mutex_lock(&root->fs_info->cleaner_mutex);
- btrfs_clean_old_snapshots(info->tree_root);
- btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
- mutex_unlock(&root->fs_info->cleaner_mutex);
+ /*
+ * ok we don't have enough space, but maybe we have free space on our
+ * devices to allocate new chunks for relocation, so loop through our
+ * alloc devices and guess if we have enough space. However, if we
+ * were marked as full, then we know there aren't enough chunks, and we
+ * can just return.
+ */
+ ret = -1;
+ if (full)
+ goto out;
- trans = btrfs_start_transaction(info->tree_root, 1);
- btrfs_commit_transaction(trans, info->tree_root);
+ mutex_lock(&root->fs_info->chunk_mutex);
+ list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
+ u64 min_free = btrfs_block_group_used(&block_group->item);
+ u64 dev_offset, max_avail;
- while (1) {
- ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
- if (ret < 0)
- goto out;
-next:
- leaf = path->nodes[0];
- nritems = btrfs_header_nritems(leaf);
- if (path->slots[0] >= nritems) {
- ret = btrfs_next_leaf(root, path);
- if (ret < 0)
- goto out;
- if (ret == 1) {
- ret = 0;
+ /*
+ * check to make sure we can actually find a chunk with enough
+ * space to fit our block group in.
+ */
+ if (device->total_bytes > device->bytes_used + min_free) {
+ ret = find_free_dev_extent(NULL, device, min_free,
+ &dev_offset, &max_avail);
+ if (!ret)
break;
- }
- leaf = path->nodes[0];
- nritems = btrfs_header_nritems(leaf);
- }
-
- btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
-
- if (key.objectid >= block_group->key.objectid +
- block_group->key.offset)
- break;
-
- if (progress && need_resched()) {
- btrfs_release_path(root, path);
- cond_resched();
- progress = 0;
- continue;
- }
- progress = 1;
-
- if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
- key.objectid + key.offset <= cur_byte) {
- path->slots[0]++;
- goto next;
+ ret = -1;
}
-
- total_found++;
- cur_byte = key.objectid + key.offset;
- btrfs_release_path(root, path);
-
- __alloc_chunk_for_shrink(root, block_group, 0);
- ret = relocate_one_extent(root, path, &key, block_group,
- reloc_inode, pass);
- BUG_ON(ret < 0);
- if (ret > 0)
- skipped++;
-
- key.objectid = cur_byte;
- key.type = 0;
- key.offset = 0;
}
-
- btrfs_release_path(root, path);
-
- if (pass == 0) {
- btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
- invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
- }
-
- if (total_found > 0) {
- printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
- (unsigned long long)total_found, pass);
- pass++;
- if (total_found == skipped && pass > 2) {
- iput(reloc_inode);
- reloc_inode = create_reloc_inode(info, block_group);
- pass = 0;
- }
- goto again;
- }
-
- /* delete reloc_inode */
- iput(reloc_inode);
-
- /* unpin extents in this range */
- trans = btrfs_start_transaction(info->tree_root, 1);
- btrfs_commit_transaction(trans, info->tree_root);
-
- spin_lock(&block_group->lock);
- WARN_ON(block_group->pinned > 0);
- WARN_ON(block_group->reserved > 0);
- WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
- spin_unlock(&block_group->lock);
- btrfs_put_block_group(block_group);
- ret = 0;
+ mutex_unlock(&root->fs_info->chunk_mutex);
out:
- btrfs_free_path(path);
+ btrfs_put_block_group(block_group);
return ret;
}
-#endif
static int find_first_block_group(struct btrfs_root *root,
struct btrfs_path *path, struct btrfs_key *key)
* time, particularly in the full case.
*/
if (found_key.offset == btrfs_block_group_used(&cache->item)) {
+ exclude_super_stripes(root, cache);
cache->last_byte_to_unpin = (u64)-1;
cache->cached = BTRFS_CACHE_FINISHED;
+ free_excluded_extents(root, cache);
} else if (btrfs_block_group_used(&cache->item) == 0) {
exclude_super_stripes(root, cache);
cache->last_byte_to_unpin = (u64)-1;
&space_info);
BUG_ON(ret);
cache->space_info = space_info;
+ spin_lock(&cache->space_info->lock);
+ cache->space_info->bytes_super += cache->bytes_super;
+ spin_unlock(&cache->space_info->lock);
+
down_write(&space_info->groups_sem);
list_add_tail(&cache->list, &space_info->block_groups);
up_write(&space_info->groups_sem);
ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
&cache->space_info);
BUG_ON(ret);
+
+ spin_lock(&cache->space_info->lock);
+ cache->space_info->bytes_super += cache->bytes_super;
+ spin_unlock(&cache->space_info->lock);
+
down_write(&cache->space_info->groups_sem);
list_add_tail(&cache->list, &cache->space_info->block_groups);
up_write(&cache->space_info->groups_sem);