struct btrfs_root *root,
int load_cache_only)
{
+ DEFINE_WAIT(wait);
struct btrfs_fs_info *fs_info = cache->fs_info;
struct btrfs_caching_control *caching_ctl;
int ret = 0;
- smp_mb();
- if (cache->cached != BTRFS_CACHE_NO)
+ caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
+ BUG_ON(!caching_ctl);
+
+ INIT_LIST_HEAD(&caching_ctl->list);
+ mutex_init(&caching_ctl->mutex);
+ init_waitqueue_head(&caching_ctl->wait);
+ caching_ctl->block_group = cache;
+ caching_ctl->progress = cache->key.objectid;
+ atomic_set(&caching_ctl->count, 1);
+ caching_ctl->work.func = caching_thread;
+
+ spin_lock(&cache->lock);
+ /*
+ * This should be a rare occasion, but this could happen I think in the
+ * case where one thread starts to load the space cache info, and then
+ * some other thread starts a transaction commit which tries to do an
+ * allocation while the other thread is still loading the space cache
+ * info. The previous loop should have kept us from choosing this block
+ * group, but if we've moved to the state where we will wait on caching
+ * block groups we need to first check if we're doing a fast load here,
+ * so we can wait for it to finish, otherwise we could end up allocating
+ * from a block group who's cache gets evicted for one reason or
+ * another.
+ */
+ while (cache->cached == BTRFS_CACHE_FAST) {
+ struct btrfs_caching_control *ctl;
+
+ ctl = cache->caching_ctl;
+ atomic_inc(&ctl->count);
+ prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
+ spin_unlock(&cache->lock);
+
+ schedule();
+
+ finish_wait(&ctl->wait, &wait);
+ put_caching_control(ctl);
+ spin_lock(&cache->lock);
+ }
+
+ if (cache->cached != BTRFS_CACHE_NO) {
+ spin_unlock(&cache->lock);
+ kfree(caching_ctl);
return 0;
+ }
+ WARN_ON(cache->caching_ctl);
+ cache->caching_ctl = caching_ctl;
+ cache->cached = BTRFS_CACHE_FAST;
+ spin_unlock(&cache->lock);
/*
* We can't do the read from on-disk cache during a commit since we need
if (trans && (!trans->transaction->in_commit) &&
(root && root != root->fs_info->tree_root) &&
btrfs_test_opt(root, SPACE_CACHE)) {
- spin_lock(&cache->lock);
- if (cache->cached != BTRFS_CACHE_NO) {
- spin_unlock(&cache->lock);
- return 0;
- }
- cache->cached = BTRFS_CACHE_STARTED;
- spin_unlock(&cache->lock);
-
ret = load_free_space_cache(fs_info, cache);
spin_lock(&cache->lock);
if (ret == 1) {
+ cache->caching_ctl = NULL;
cache->cached = BTRFS_CACHE_FINISHED;
cache->last_byte_to_unpin = (u64)-1;
} else {
- cache->cached = BTRFS_CACHE_NO;
+ if (load_cache_only) {
+ cache->caching_ctl = NULL;
+ cache->cached = BTRFS_CACHE_NO;
+ } else {
+ cache->cached = BTRFS_CACHE_STARTED;
+ }
}
spin_unlock(&cache->lock);
+ wake_up(&caching_ctl->wait);
if (ret == 1) {
+ put_caching_control(caching_ctl);
free_excluded_extents(fs_info->extent_root, cache);
return 0;
}
+ } else {
+ /*
+ * We are not going to do the fast caching, set cached to the
+ * appropriate value and wakeup any waiters.
+ */
+ spin_lock(&cache->lock);
+ if (load_cache_only) {
+ cache->caching_ctl = NULL;
+ cache->cached = BTRFS_CACHE_NO;
+ } else {
+ cache->cached = BTRFS_CACHE_STARTED;
+ }
+ spin_unlock(&cache->lock);
+ wake_up(&caching_ctl->wait);
}
- if (load_cache_only)
- return 0;
-
- caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
- BUG_ON(!caching_ctl);
-
- INIT_LIST_HEAD(&caching_ctl->list);
- mutex_init(&caching_ctl->mutex);
- init_waitqueue_head(&caching_ctl->wait);
- caching_ctl->block_group = cache;
- caching_ctl->progress = cache->key.objectid;
- /* one for caching kthread, one for caching block group list */
- atomic_set(&caching_ctl->count, 2);
- caching_ctl->work.func = caching_thread;
-
- spin_lock(&cache->lock);
- if (cache->cached != BTRFS_CACHE_NO) {
- spin_unlock(&cache->lock);
- kfree(caching_ctl);
+ if (load_cache_only) {
+ put_caching_control(caching_ctl);
return 0;
}
- cache->caching_ctl = caching_ctl;
- cache->cached = BTRFS_CACHE_STARTED;
- spin_unlock(&cache->lock);
down_write(&fs_info->extent_commit_sem);
+ atomic_inc(&caching_ctl->count);
list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
up_write(&fs_info->extent_commit_sem);
{
int ret;
u64 discarded_bytes = 0;
- struct btrfs_multi_bio *multi = NULL;
+ struct btrfs_bio *bbio = NULL;
/* Tell the block device(s) that the sectors can be discarded */
ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
- bytenr, &num_bytes, &multi, 0);
+ bytenr, &num_bytes, &bbio, 0);
if (!ret) {
- struct btrfs_bio_stripe *stripe = multi->stripes;
+ struct btrfs_bio_stripe *stripe = bbio->stripes;
int i;
- for (i = 0; i < multi->num_stripes; i++, stripe++) {
+ for (i = 0; i < bbio->num_stripes; i++, stripe++) {
if (!stripe->dev->can_discard)
continue;
*/
ret = 0;
}
- kfree(multi);
+ kfree(bbio);
}
if (actual_bytes)
/*
* shrink metadata reservation for delalloc
*/
-static int shrink_delalloc(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, u64 to_reclaim,
+static int shrink_delalloc(struct btrfs_root *root, u64 to_reclaim,
bool wait_ordered)
{
struct btrfs_block_rsv *block_rsv;
struct btrfs_space_info *space_info;
+ struct btrfs_trans_handle *trans;
u64 reserved;
u64 max_reclaim;
u64 reclaimed = 0;
int loops = 0;
unsigned long progress;
+ trans = (struct btrfs_trans_handle *)current->journal_info;
block_rsv = &root->fs_info->delalloc_block_rsv;
space_info = block_rsv->space_info;
smp_mb();
nr_pages = min_t(unsigned long, nr_pages,
root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
- writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
+ writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages,
+ WB_REASON_FS_FREE_SPACE);
spin_lock(&space_info->lock);
if (reserved > space_info->bytes_may_use)
return reclaimed >= to_reclaim;
}
+/**
+ * maybe_commit_transaction - possibly commit the transaction if its ok to
+ * @root - the root we're allocating for
+ * @bytes - the number of bytes we want to reserve
+ * @force - force the commit
+ *
+ * This will check to make sure that committing the transaction will actually
+ * get us somewhere and then commit the transaction if it does. Otherwise it
+ * will return -ENOSPC.
+ */
+static int may_commit_transaction(struct btrfs_root *root,
+ struct btrfs_space_info *space_info,
+ u64 bytes, int force)
+{
+ struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
+ struct btrfs_trans_handle *trans;
+
+ trans = (struct btrfs_trans_handle *)current->journal_info;
+ if (trans)
+ return -EAGAIN;
+
+ if (force)
+ goto commit;
+
+ /* See if there is enough pinned space to make this reservation */
+ spin_lock(&space_info->lock);
+ if (space_info->bytes_pinned >= bytes) {
+ spin_unlock(&space_info->lock);
+ goto commit;
+ }
+ spin_unlock(&space_info->lock);
+
+ /*
+ * See if there is some space in the delayed insertion reservation for
+ * this reservation.
+ */
+ if (space_info != delayed_rsv->space_info)
+ return -ENOSPC;
+
+ spin_lock(&delayed_rsv->lock);
+ if (delayed_rsv->size < bytes) {
+ spin_unlock(&delayed_rsv->lock);
+ return -ENOSPC;
+ }
+ spin_unlock(&delayed_rsv->lock);
+
+commit:
+ trans = btrfs_join_transaction(root);
+ if (IS_ERR(trans))
+ return -ENOSPC;
+
+ return btrfs_commit_transaction(trans, root);
+}
+
/**
* reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
* @root - the root we're allocating for
u64 orig_bytes, int flush)
{
struct btrfs_space_info *space_info = block_rsv->space_info;
- struct btrfs_trans_handle *trans;
u64 used;
u64 num_bytes = orig_bytes;
int retries = 0;
bool flushing = false;
bool wait_ordered = false;
- trans = (struct btrfs_trans_handle *)current->journal_info;
again:
ret = 0;
spin_lock(&space_info->lock);
* deadlock since we are waiting for the flusher to finish, but
* hold the current transaction open.
*/
- if (trans)
+ if (current->journal_info)
return -EAGAIN;
ret = wait_event_interruptible(space_info->wait,
!space_info->flush);
*/
avail = (space_info->total_bytes - space_info->bytes_used) * 8;
do_div(avail, 10);
- if (space_info->bytes_pinned >= avail && flush && !trans &&
- !committed) {
+ if (space_info->bytes_pinned >= avail && flush && !committed) {
space_info->flush = 1;
flushing = true;
spin_unlock(&space_info->lock);
- goto commit;
+ ret = may_commit_transaction(root, space_info,
+ orig_bytes, 1);
+ if (ret)
+ goto out;
+ committed = true;
+ goto again;
}
spin_lock(&root->fs_info->free_chunk_lock);
* We do synchronous shrinking since we don't actually unreserve
* metadata until after the IO is completed.
*/
- ret = shrink_delalloc(trans, root, num_bytes, wait_ordered);
+ ret = shrink_delalloc(root, num_bytes, wait_ordered);
if (ret < 0)
goto out;
goto again;
}
- ret = -EAGAIN;
- if (trans)
- goto out;
-
-commit:
ret = -ENOSPC;
if (committed)
goto out;
- trans = btrfs_join_transaction(root);
- if (IS_ERR(trans))
- goto out;
- ret = btrfs_commit_transaction(trans, root);
+ ret = may_commit_transaction(root, space_info, orig_bytes, 0);
if (!ret) {
- trans = NULL;
committed = true;
goto again;
}
kfree(rsv);
}
-int btrfs_block_rsv_add(struct btrfs_root *root,
- struct btrfs_block_rsv *block_rsv,
- u64 num_bytes)
+static inline int __block_rsv_add(struct btrfs_root *root,
+ struct btrfs_block_rsv *block_rsv,
+ u64 num_bytes, int flush)
{
int ret;
if (num_bytes == 0)
return 0;
- ret = reserve_metadata_bytes(root, block_rsv, num_bytes, 1);
+ ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
if (!ret) {
block_rsv_add_bytes(block_rsv, num_bytes, 1);
return 0;
return ret;
}
+int btrfs_block_rsv_add(struct btrfs_root *root,
+ struct btrfs_block_rsv *block_rsv,
+ u64 num_bytes)
+{
+ return __block_rsv_add(root, block_rsv, num_bytes, 1);
+}
+
+int btrfs_block_rsv_add_noflush(struct btrfs_root *root,
+ struct btrfs_block_rsv *block_rsv,
+ u64 num_bytes)
+{
+ return __block_rsv_add(root, block_rsv, num_bytes, 0);
+}
+
int btrfs_block_rsv_check(struct btrfs_root *root,
struct btrfs_block_rsv *block_rsv, int min_factor)
{
*/
static unsigned drop_outstanding_extent(struct inode *inode)
{
+ unsigned drop_inode_space = 0;
unsigned dropped_extents = 0;
BUG_ON(!BTRFS_I(inode)->outstanding_extents);
BTRFS_I(inode)->outstanding_extents--;
+ if (BTRFS_I(inode)->outstanding_extents == 0 &&
+ BTRFS_I(inode)->delalloc_meta_reserved) {
+ drop_inode_space = 1;
+ BTRFS_I(inode)->delalloc_meta_reserved = 0;
+ }
+
/*
* If we have more or the same amount of outsanding extents than we have
* reserved then we need to leave the reserved extents count alone.
*/
if (BTRFS_I(inode)->outstanding_extents >=
BTRFS_I(inode)->reserved_extents)
- return 0;
+ return drop_inode_space;
dropped_extents = BTRFS_I(inode)->reserved_extents -
BTRFS_I(inode)->outstanding_extents;
BTRFS_I(inode)->reserved_extents -= dropped_extents;
- return dropped_extents;
+ return dropped_extents + drop_inode_space;
}
/**
nr_extents = BTRFS_I(inode)->outstanding_extents -
BTRFS_I(inode)->reserved_extents;
BTRFS_I(inode)->reserved_extents += nr_extents;
+ }
- to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
+ /*
+ * Add an item to reserve for updating the inode when we complete the
+ * delalloc io.
+ */
+ if (!BTRFS_I(inode)->delalloc_meta_reserved) {
+ nr_extents++;
+ BTRFS_I(inode)->delalloc_meta_reserved = 1;
}
+
+ to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
spin_unlock(&BTRFS_I(inode)->lock);
}
have_block_group:
- if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
+ cached = block_group_cache_done(block_group);
+ if (unlikely(!cached)) {
u64 free_percent;
+ found_uncached_bg = true;
ret = cache_block_group(block_group, trans,
orig_root, 1);
if (block_group->cached == BTRFS_CACHE_FINISHED)
- goto have_block_group;
+ goto alloc;
free_percent = btrfs_block_group_used(&block_group->item);
free_percent *= 100;
orig_root, 0);
BUG_ON(ret);
}
- found_uncached_bg = true;
/*
* If loop is set for cached only, try the next block
goto loop;
}
- cached = block_group_cache_done(block_group);
- if (unlikely(!cached))
- found_uncached_bg = true;
-
+alloc:
if (unlikely(block_group->ro))
goto loop;
projects / ~andy / linux / blobdiff