goto out;
}
- /*
- * If this is a system chunk allocation then stop right here and do not
- * add the chunk item to the chunk btree. This is to prevent a deadlock
- * because this system chunk allocation can be triggered while COWing
- * some extent buffer of the chunk btree and while holding a lock on a
- * parent extent buffer, in which case attempting to insert the chunk
- * item (or update the device item) would result in a deadlock on that
- * parent extent buffer. In this case defer the chunk btree updates to
- * the second phase of chunk allocation and keep our reservation until
- * the second phase completes.
- *
- * This is a rare case and can only be triggered by the very few cases
- * we have where we need to touch the chunk btree outside chunk allocation
- * and chunk removal. These cases are basically adding a device, removing
- * a device or resizing a device.
- */
- if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
- return 0;
-
ret = btrfs_chunk_alloc_add_chunk_item(trans, bg);
/*
* Normally we are not expected to fail with -ENOSPC here, since we have
* This has happened before and commit eafa4fd0ad0607 ("btrfs: fix exhaustion of
* the system chunk array due to concurrent allocations") provides more details.
*
- * For allocation of system chunks, we defer the updates and insertions into the
- * chunk btree to phase 2. This is to prevent deadlocks on extent buffers because
- * if the chunk allocation is triggered while COWing an extent buffer of the
- * chunk btree, we are holding a lock on the parent of that extent buffer and
- * doing the chunk btree updates and insertions can require locking that parent.
- * This is for the very few and rare cases where we update the chunk btree that
- * are not chunk allocation or chunk removal: adding a device, removing a device
- * or resizing a device.
+ * Allocation of system chunks does not happen through this function. A task that
+ * needs to update the chunk btree (the only btree that uses system chunks), must
+ * preallocate chunk space by calling either check_system_chunk() or
+ * btrfs_reserve_chunk_metadata() - the former is used when allocating a data or
+ * metadata chunk or when removing a chunk, while the later is used before doing
+ * a modification to the chunk btree - use cases for the later are adding,
+ * removing and resizing a device as well as relocation of a system chunk.
+ * See the comment below for more details.
*
* The reservation of system space, done through check_system_chunk(), as well
* as all the updates and insertions into the chunk btree must be done while
if (trans->allocating_chunk)
return -ENOSPC;
/*
- * If we are removing a chunk, don't re-enter or we would deadlock.
- * System space reservation and system chunk allocation is done by the
- * chunk remove operation (btrfs_remove_chunk()).
+ * Allocation of system chunks can not happen through this path, as we
+ * could end up in a deadlock if we are allocating a data or metadata
+ * chunk and there is another task modifying the chunk btree.
+ *
+ * This is because while we are holding the chunk mutex, we will attempt
+ * to add the new chunk item to the chunk btree or update an existing
+ * device item in the chunk btree, while the other task that is modifying
+ * the chunk btree is attempting to COW an extent buffer while holding a
+ * lock on it and on its parent - if the COW operation triggers a system
+ * chunk allocation, then we can deadlock because we are holding the
+ * chunk mutex and we may need to access that extent buffer or its parent
+ * in order to add the chunk item or update a device item.
+ *
+ * Tasks that want to modify the chunk tree should reserve system space
+ * before updating the chunk btree, by calling either
+ * btrfs_reserve_chunk_metadata() or check_system_chunk().
+ * It's possible that after a task reserves the space, it still ends up
+ * here - this happens in the cases described above at do_chunk_alloc().
+ * The task will have to either retry or fail.
*/
- if (trans->removing_chunk)
+ if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
return -ENOSPC;
space_info = btrfs_find_space_info(fs_info, flags);
return num_dev;
}
-/*
- * Reserve space in the system space for allocating or removing a chunk
- */
-void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
+static void reserve_chunk_space(struct btrfs_trans_handle *trans,
+ u64 bytes,
+ u64 type)
{
struct btrfs_fs_info *fs_info = trans->fs_info;
struct btrfs_space_info *info;
u64 left;
- u64 thresh;
int ret = 0;
- u64 num_devs;
/*
* Needed because we can end up allocating a system chunk and for an
left = info->total_bytes - btrfs_space_info_used(info, true);
spin_unlock(&info->lock);
- num_devs = get_profile_num_devs(fs_info, type);
-
- /* num_devs device items to update and 1 chunk item to add or remove */
- thresh = btrfs_calc_metadata_size(fs_info, num_devs) +
- btrfs_calc_insert_metadata_size(fs_info, 1);
-
- if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
+ if (left < bytes && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
- left, thresh, type);
+ left, bytes, type);
btrfs_dump_space_info(fs_info, info, 0, 0);
}
- if (left < thresh) {
+ if (left < bytes) {
u64 flags = btrfs_system_alloc_profile(fs_info);
struct btrfs_block_group *bg;
* needing it, as we might not need to COW all nodes/leafs from
* the paths we visit in the chunk tree (they were already COWed
* or created in the current transaction for example).
- *
- * Also, if our caller is allocating a system chunk, do not
- * attempt to insert the chunk item in the chunk btree, as we
- * could deadlock on an extent buffer since our caller may be
- * COWing an extent buffer from the chunk btree.
*/
bg = btrfs_create_chunk(trans, flags);
if (IS_ERR(bg)) {
ret = PTR_ERR(bg);
- } else if (!(type & BTRFS_BLOCK_GROUP_SYSTEM)) {
+ } else {
/*
* If we fail to add the chunk item here, we end up
* trying again at phase 2 of chunk allocation, at
* btrfs_create_pending_block_groups(). So ignore
- * any error here.
+ * any error here. An ENOSPC here could happen, due to
+ * the cases described at do_chunk_alloc() - the system
+ * block group we just created was just turned into RO
+ * mode by a scrub for example, or a running discard
+ * temporarily removed its free space entries, etc.
*/
btrfs_chunk_alloc_add_chunk_item(trans, bg);
}
if (!ret) {
ret = btrfs_block_rsv_add(fs_info->chunk_root,
&fs_info->chunk_block_rsv,
- thresh, BTRFS_RESERVE_NO_FLUSH);
+ bytes, BTRFS_RESERVE_NO_FLUSH);
if (!ret)
- trans->chunk_bytes_reserved += thresh;
+ trans->chunk_bytes_reserved += bytes;
}
}
+/*
+ * Reserve space in the system space for allocating or removing a chunk.
+ * The caller must be holding fs_info->chunk_mutex.
+ */
+void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
+{
+ struct btrfs_fs_info *fs_info = trans->fs_info;
+ const u64 num_devs = get_profile_num_devs(fs_info, type);
+ u64 bytes;
+
+ /* num_devs device items to update and 1 chunk item to add or remove. */
+ bytes = btrfs_calc_metadata_size(fs_info, num_devs) +
+ btrfs_calc_insert_metadata_size(fs_info, 1);
+
+ reserve_chunk_space(trans, bytes, type);
+}
+
+/*
+ * Reserve space in the system space, if needed, for doing a modification to the
+ * chunk btree.
+ *
+ * @trans: A transaction handle.
+ * @is_item_insertion: Indicate if the modification is for inserting a new item
+ * in the chunk btree or if it's for the deletion or update
+ * of an existing item.
+ *
+ * This is used in a context where we need to update the chunk btree outside
+ * block group allocation and removal, to avoid a deadlock with a concurrent
+ * task that is allocating a metadata or data block group and therefore needs to
+ * update the chunk btree while holding the chunk mutex. After the update to the
+ * chunk btree is done, btrfs_trans_release_chunk_metadata() should be called.
+ *
+ */
+void btrfs_reserve_chunk_metadata(struct btrfs_trans_handle *trans,
+ bool is_item_insertion)
+{
+ struct btrfs_fs_info *fs_info = trans->fs_info;
+ u64 bytes;
+
+ if (is_item_insertion)
+ bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
+ else
+ bytes = btrfs_calc_metadata_size(fs_info, 1);
+
+ mutex_lock(&fs_info->chunk_mutex);
+ reserve_chunk_space(trans, bytes, BTRFS_BLOCK_GROUP_SYSTEM);
+ mutex_unlock(&fs_info->chunk_mutex);
+}
+
void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
{
struct btrfs_block_group *block_group;
key.type = BTRFS_DEV_ITEM_KEY;
key.offset = device->devid;
+ btrfs_reserve_chunk_metadata(trans, true);
ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
&key, sizeof(*dev_item));
+ btrfs_trans_release_chunk_metadata(trans);
if (ret)
goto out;
key.type = BTRFS_DEV_ITEM_KEY;
key.offset = device->devid;
+ btrfs_reserve_chunk_metadata(trans, false);
ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
+ btrfs_trans_release_chunk_metadata(trans);
if (ret) {
if (ret > 0)
ret = -ENOENT;
key.type = BTRFS_DEV_ITEM_KEY;
while (1) {
+ btrfs_reserve_chunk_metadata(trans, false);
ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
+ btrfs_trans_release_chunk_metadata(trans);
if (ret < 0)
goto error;
struct btrfs_super_block *super_copy = fs_info->super_copy;
u64 old_total;
u64 diff;
+ int ret;
if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
return -EACCES;
&trans->transaction->dev_update_list);
mutex_unlock(&fs_info->chunk_mutex);
- return btrfs_update_device(trans, device);
+ btrfs_reserve_chunk_metadata(trans, false);
+ ret = btrfs_update_device(trans, device);
+ btrfs_trans_release_chunk_metadata(trans);
+
+ return ret;
}
static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
round_down(old_total - diff, fs_info->sectorsize));
mutex_unlock(&fs_info->chunk_mutex);
+ btrfs_reserve_chunk_metadata(trans, false);
/* Now btrfs_update_device() will change the on-disk size. */
ret = btrfs_update_device(trans, device);
+ btrfs_trans_release_chunk_metadata(trans);
if (ret < 0) {
btrfs_abort_transaction(trans, ret);
btrfs_end_transaction(trans);
OSDN Git Service
