/* * Relocation overview * * [What does relocation do] * * The objective of relocation is to relocate all extents of the target block * group to other block groups. * This is utilized by resize (shrink only), profile converting, compacting * space, or balance routine to spread chunks over devices. * * Before | After * ------------------------------------------------------------------ * BG A: 10 data extents | BG A: deleted * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) * * [How does relocation work] * * 1. Mark the target block group read-only * New extents won't be allocated from the target block group. * * 2.1 Record each extent in the target block group * To build a proper map of extents to be relocated. * * 2.2 Build data reloc tree and reloc trees * Data reloc tree will contain an inode, recording all newly relocated * data extents. * There will be only one data reloc tree for one data block group. * * Reloc tree will be a special snapshot of its source tree, containing * relocated tree blocks. * Each tree referring to a tree block in target block group will get its * reloc tree built. * * 2.3 Swap source tree with its corresponding reloc tree * Each involved tree only refers to new extents after swap. * * 3. Cleanup reloc trees and data reloc tree. * As old extents in the target block group are still referenced by reloc * trees, we need to clean them up before really freeing the target block * group. * * The main complexity is in steps 2.2 and 2.3. * * The entry point of relocation is relocate_block_group() function.
*/
#define RELOCATION_RESERVED_NODES 256 /* * map address of tree root to tree
*/ struct mapping_node { union { /* Use rb_simple_node for search/insert */ struct { struct rb_node rb_node;
u64 bytenr;
};
/* * present a tree block to process
*/ struct tree_block { union { /* Use rb_simple_node for search/insert */ struct { struct rb_node rb_node;
u64 bytenr;
};
/* Stages of data relocation. */ enum reloc_stage {
MOVE_DATA_EXTENTS,
UPDATE_DATA_PTRS
};
struct reloc_control { /* block group to relocate */ struct btrfs_block_group *block_group; /* extent tree */ struct btrfs_root *extent_root; /* inode for moving data */ struct inode *data_inode;
struct btrfs_block_rsv *block_rsv;
struct btrfs_backref_cache backref_cache;
struct file_extent_cluster cluster; /* tree blocks have been processed */ struct extent_io_tree processed_blocks; /* map start of tree root to corresponding reloc tree */ struct mapping_tree reloc_root_tree; /* list of reloc trees */ struct list_head reloc_roots; /* list of subvolume trees that get relocated */ struct list_head dirty_subvol_roots; /* size of metadata reservation for merging reloc trees */
u64 merging_rsv_size; /* size of relocated tree nodes */
u64 nodes_relocated; /* reserved size for block group relocation*/
u64 reserved_bytes;
/* * walk down backref nodes to find start of next reference path
*/ staticstruct btrfs_backref_node *walk_down_backref( struct btrfs_backref_edge *edges[], int *index)
{ struct btrfs_backref_edge *edge; struct btrfs_backref_node *lower; int idx = *index;
staticbool reloc_root_is_dead(conststruct btrfs_root *root)
{ /* * Pair with set_bit/clear_bit in clean_dirty_subvols and * btrfs_update_reloc_root. We need to see the updated bit before * trying to access reloc_root
*/
smp_rmb(); if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) returntrue; returnfalse;
}
/* * Check if this subvolume tree has valid reloc tree. * * Reloc tree after swap is considered dead, thus not considered as valid. * This is enough for most callers, as they don't distinguish dead reloc root * from no reloc root. But btrfs_should_ignore_reloc_root() below is a * special case.
*/ staticbool have_reloc_root(conststruct btrfs_root *root)
{ if (reloc_root_is_dead(root)) returnfalse; if (!root->reloc_root) returnfalse; returntrue;
}
if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) returnfalse;
/* This root has been merged with its reloc tree, we can ignore it */ if (reloc_root_is_dead(root)) returntrue;
reloc_root = root->reloc_root; if (!reloc_root) returnfalse;
if (btrfs_header_generation(reloc_root->commit_root) ==
root->fs_info->running_transaction->transid) returnfalse; /* * If there is reloc tree and it was created in previous transaction * backref lookup can find the reloc tree, so backref node for the fs * tree root is useless for relocation.
*/ returntrue;
}
/* * find reloc tree by address of tree root
*/ struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
{ struct reloc_control *rc = fs_info->reloc_ctl; struct rb_node *rb_node; struct mapping_node *node; struct btrfs_root *root = NULL;
/* * For useless nodes, do two major clean ups: * * - Cleanup the children edges and nodes * If child node is also orphan (no parent) during cleanup, then the child * node will also be cleaned up. * * - Freeing up leaves (level 0), keeps nodes detached * For nodes, the node is still cached as "detached" * * Return false if @node is not in the @useless_nodes list. * Return true if @node is in the @useless_nodes list.
*/ staticbool handle_useless_nodes(struct reloc_control *rc, struct btrfs_backref_node *node)
{ struct btrfs_backref_cache *cache = &rc->backref_cache; struct list_head *useless_node = &cache->useless_node; bool ret = false;
while (!list_empty(useless_node)) { struct btrfs_backref_node *cur;
cur = list_first_entry(useless_node, struct btrfs_backref_node,
list);
list_del_init(&cur->list);
/* Only tree root nodes can be added to @useless_nodes */
ASSERT(list_empty(&cur->upper));
if (cur == node)
ret = true;
/* Cleanup the lower edges */ while (!list_empty(&cur->lower)) { struct btrfs_backref_edge *edge; struct btrfs_backref_node *lower;
/* Child node is also orphan, queue for cleanup */ if (list_empty(&lower->upper))
list_add(&lower->list, useless_node);
} /* Mark this block processed for relocation */
mark_block_processed(rc, cur);
/* * Backref nodes for tree leaves are deleted from the cache. * Backref nodes for upper level tree blocks are left in the * cache to avoid unnecessary backref lookup.
*/ if (cur->level > 0) {
cur->detached = 1;
} else {
rb_erase(&cur->rb_node, &cache->rb_root);
btrfs_backref_free_node(cache, cur);
}
} return ret;
}
/* * Build backref tree for a given tree block. Root of the backref tree * corresponds the tree block, leaves of the backref tree correspond roots of * b-trees that reference the tree block. * * The basic idea of this function is check backrefs of a given block to find * upper level blocks that reference the block, and then check backrefs of * these upper level blocks recursively. The recursion stops when tree root is * reached or backrefs for the block is cached. * * NOTE: if we find that backrefs for a block are cached, we know backrefs for * all upper level blocks that directly/indirectly reference the block are also * cached.
*/ static noinline_for_stack struct btrfs_backref_node *build_backref_tree( struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_key *node_key, int level, u64 bytenr)
{ struct btrfs_backref_iter *iter; struct btrfs_backref_cache *cache = &rc->backref_cache; /* For searching parent of TREE_BLOCK_REF */ struct btrfs_path *path; struct btrfs_backref_node *cur; struct btrfs_backref_node *node = NULL; struct btrfs_backref_edge *edge; int ret;
iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info); if (!iter) return ERR_PTR(-ENOMEM);
path = btrfs_alloc_path(); if (!path) {
ret = -ENOMEM; goto out;
}
node = btrfs_backref_alloc_node(cache, bytenr, level); if (!node) {
ret = -ENOMEM; goto out;
}
cur = node;
/* Breadth-first search to build backref cache */ do {
ret = btrfs_backref_add_tree_node(trans, cache, path, iter,
node_key, cur); if (ret < 0) goto out;
edge = list_first_entry_or_null(&cache->pending_edge, struct btrfs_backref_edge, list[UPPER]); /* * The pending list isn't empty, take the first block to * process
*/ if (edge) {
list_del_init(&edge->list[UPPER]);
cur = edge->node[UPPER];
}
} while (edge);
/* Finish the upper linkage of newly added edges/nodes */
ret = btrfs_backref_finish_upper_links(cache, node); if (ret < 0) goto out;
spin_lock(&rc->reloc_root_tree.lock);
rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root, &node->simple_node);
spin_unlock(&rc->reloc_root_tree.lock); if (rb_node) {
btrfs_err(fs_info, "Duplicate root found for start=%llu while inserting into relocation tree",
node->bytenr); return -EEXIST;
}
/* * We only put the reloc root here if it's on the list. There's a lot * of places where the pattern is to splice the rc->reloc_roots, process * the reloc roots, and then add the reloc root back onto * rc->reloc_roots. If we call __del_reloc_root while it's off of the * list we don't want the reference being dropped, because the guy * messing with the list is in charge of the reference.
*/
spin_lock(&fs_info->trans_lock); if (!list_empty(&root->root_list)) {
put_ref = true;
list_del_init(&root->root_list);
}
spin_unlock(&fs_info->trans_lock); if (put_ref)
btrfs_put_root(root);
kfree(node);
}
/* * helper to update the 'address of tree root -> reloc tree' * mapping
*/ staticint __update_reloc_root(struct btrfs_root *root)
{ struct btrfs_fs_info *fs_info = root->fs_info; struct rb_node *rb_node; struct mapping_node *node = NULL; struct reloc_control *rc = fs_info->reloc_ctl;
if (btrfs_root_id(root) == objectid) {
u64 commit_root_gen;
/* * Relocation will wait for cleaner thread, and any half-dropped * subvolume will be fully cleaned up at mount time. * So here we shouldn't hit a subvolume with non-zero drop_progress. * * If this isn't the case, error out since it can make us attempt to * drop references for extents that were already dropped before.
*/ if (unlikely(btrfs_disk_key_objectid(&root->root_item.drop_progress))) { struct btrfs_key cpu_key;
/* called by btrfs_init_reloc_root */
ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
BTRFS_TREE_RELOC_OBJECTID); if (ret) goto fail;
/* * Set the last_snapshot field to the generation of the commit * root - like this ctree.c:btrfs_block_can_be_shared() behaves * correctly (returns true) when the relocation root is created * either inside the critical section of a transaction commit * (through transaction.c:qgroup_account_snapshot()) and when * it's created before the transaction commit is started.
*/
commit_root_gen = btrfs_header_generation(root->commit_root);
btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
} else { /* * called by btrfs_reloc_post_snapshot_hook. * the source tree is a reloc tree, all tree blocks * modified after it was created have RELOC flag * set in their headers. so it's OK to not update * the 'last_snapshot'.
*/
ret = btrfs_copy_root(trans, root, root->node, &eb,
BTRFS_TREE_RELOC_OBJECTID); if (ret) goto fail;
}
/* * We have changed references at this point, we must abort the * transaction if anything fails.
*/
must_abort = true;
ret = btrfs_insert_root(trans, fs_info->tree_root,
&root_key, root_item); if (ret) goto fail;
kfree(root_item);
reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); if (IS_ERR(reloc_root)) {
ret = PTR_ERR(reloc_root); goto abort;
}
set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
btrfs_set_root_last_trans(reloc_root, trans->transid); return reloc_root;
fail:
kfree(root_item);
abort: if (must_abort)
btrfs_abort_transaction(trans, ret); return ERR_PTR(ret);
}
/* * create reloc tree for a given fs tree. reloc tree is just a * snapshot of the fs tree with special root objectid. * * The reloc_root comes out of here with two references, one for * root->reloc_root, and another for being on the rc->reloc_roots list.
*/ int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, struct btrfs_root *root)
{ struct btrfs_fs_info *fs_info = root->fs_info; struct btrfs_root *reloc_root; struct reloc_control *rc = fs_info->reloc_ctl; struct btrfs_block_rsv *rsv; int clear_rsv = 0; int ret;
if (!rc) return 0;
/* * The subvolume has reloc tree but the swap is finished, no need to * create/update the dead reloc tree
*/ if (reloc_root_is_dead(root)) return 0;
/* * This is subtle but important. We do not do * record_root_in_transaction for reloc roots, instead we record their * corresponding fs root, and then here we update the last trans for the * reloc root. This means that we have to do this for the entire life * of the reloc root, regardless of which stage of the relocation we are * in.
*/ if (root->reloc_root) {
reloc_root = root->reloc_root;
btrfs_set_root_last_trans(reloc_root, trans->transid); return 0;
}
/* * We are merging reloc roots, we do not need new reloc trees. Also * reloc trees never need their own reloc tree.
*/ if (!rc->create_reloc_tree || btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) return 0;
if (!trans->reloc_reserved) {
rsv = trans->block_rsv;
trans->block_rsv = rc->block_rsv;
clear_rsv = 1;
}
reloc_root = create_reloc_root(trans, root, btrfs_root_id(root)); if (clear_rsv)
trans->block_rsv = rsv; if (IS_ERR(reloc_root)) return PTR_ERR(reloc_root);
ret = __add_reloc_root(reloc_root);
ASSERT(ret != -EEXIST); if (ret) { /* Pairs with create_reloc_root */
btrfs_put_root(reloc_root); return ret;
}
root->reloc_root = btrfs_grab_root(reloc_root); return 0;
}
/* * update root item of reloc tree
*/ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, struct btrfs_root *root)
{ struct btrfs_fs_info *fs_info = root->fs_info; struct btrfs_root *reloc_root; struct btrfs_root_item *root_item; int ret;
/* * We are probably ok here, but __del_reloc_root() will drop its ref of * the root. We have the ref for root->reloc_root, but just in case * hold it while we update the reloc root.
*/
btrfs_grab_root(reloc_root);
/* root->reloc_root will stay until current relocation finished */ if (fs_info->reloc_ctl && fs_info->reloc_ctl->merge_reloc_tree &&
btrfs_root_refs(root_item) == 0) {
set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); /* * Mark the tree as dead before we change reloc_root so * have_reloc_root will not touch it from now on.
*/
smp_wmb();
__del_reloc_root(reloc_root);
}
/* * update file extent items in the tree leaf to point to * the new locations.
*/ static noinline_for_stack int replace_file_extents(struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_root *root, struct extent_buffer *leaf)
{ struct btrfs_fs_info *fs_info = root->fs_info; struct btrfs_key key; struct btrfs_file_extent_item *fi; struct btrfs_inode *inode = NULL;
u64 parent;
u64 bytenr;
u64 new_bytenr = 0;
u64 num_bytes;
u64 end;
u32 nritems;
u32 i; int ret = 0; int first = 1;
if (rc->stage != UPDATE_DATA_PTRS) return 0;
/* reloc trees always use full backref */ if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID)
parent = leaf->start; else
parent = 0;
nritems = btrfs_header_nritems(leaf); for (i = 0; i < nritems; i++) { struct btrfs_ref ref = { 0 };
cond_resched();
btrfs_item_key_to_cpu(leaf, &key, i); if (key.type != BTRFS_EXTENT_DATA_KEY) continue;
fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); if (btrfs_file_extent_type(leaf, fi) ==
BTRFS_FILE_EXTENT_INLINE) continue;
bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); if (bytenr == 0) continue; if (!in_range(bytenr, rc->block_group->start,
rc->block_group->length)) continue;
/* * if we are modifying block in fs tree, wait for read_folio * to complete and drop the extent cache
*/ if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) { if (first) {
inode = btrfs_find_first_inode(root, key.objectid);
first = 0;
} elseif (inode && btrfs_ino(inode) < key.objectid) {
btrfs_add_delayed_iput(inode);
inode = btrfs_find_first_inode(root, key.objectid);
} if (inode && btrfs_ino(inode) == key.objectid) { struct extent_state *cached_state = NULL;
end = key.offset +
btrfs_file_extent_num_bytes(leaf, fi);
WARN_ON(!IS_ALIGNED(key.offset,
fs_info->sectorsize));
WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
end--; /* Take mmap lock to serialize with reflinks. */ if (!down_read_trylock(&inode->i_mmap_lock)) continue;
ret = btrfs_try_lock_extent(&inode->io_tree, key.offset,
end, &cached_state); if (!ret) {
up_read(&inode->i_mmap_lock); continue;
}
ret = get_new_location(rc->data_inode, &new_bytenr,
bytenr, num_bytes); if (ret) { /* * Don't have to abort since we've not changed anything * in the file extent yet.
*/ break;
}
/* * try to replace tree blocks in fs tree with the new blocks * in reloc tree. tree blocks haven't been modified since the * reloc tree was create can be replaced. * * if a block was replaced, level of the block + 1 is returned. * if no block got replaced, 0 is returned. if there are other * errors, a negative error number is returned.
*/ static noinline_for_stack int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_root *dest, struct btrfs_root *src, struct btrfs_path *path, struct btrfs_key *next_key, int lowest_level, int max_level)
{ struct btrfs_fs_info *fs_info = dest->fs_info; struct extent_buffer *eb; struct extent_buffer *parent; struct btrfs_ref ref = { 0 }; struct btrfs_key key;
u64 old_bytenr;
u64 new_bytenr;
u64 old_ptr_gen;
u64 new_ptr_gen;
u64 last_snapshot;
u32 blocksize; int cow = 0; int level; int ret; int slot;
path->lowest_level = level;
set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
path->lowest_level = 0; if (ret) { if (ret > 0)
ret = -ENOENT; break;
}
/* * Info qgroup to trace both subtrees. * * We must trace both trees. * 1) Tree reloc subtree * If not traced, we will leak data numbers * 2) Fs subtree * If not traced, we will double count old data * * We don't scan the subtree right now, but only record * the swapped tree blocks. * The real subtree rescan is delayed until we have new * CoW on the subtree root node before transaction commit.
*/
ret = btrfs_qgroup_add_swapped_blocks(dest,
rc->block_group, parent, slot,
path->nodes[level], path->slots[level],
last_snapshot); if (ret < 0) break; /* * swap blocks in fs tree and reloc tree.
*/
btrfs_set_node_blockptr(parent, slot, new_bytenr);
btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
/* We don't know the real owning_root, use 0. */
ref.action = BTRFS_DROP_DELAYED_REF;
ref.bytenr = new_bytenr;
ref.num_bytes = blocksize;
ref.parent = path->nodes[level]->start;
ref.owning_root = 0;
ref.ref_root = btrfs_root_id(src);
btrfs_init_tree_ref(&ref, level - 1, 0, true);
ret = btrfs_free_extent(trans, &ref); if (ret) {
btrfs_abort_transaction(trans, ret); break;
}
/* We don't know the real owning_root, use 0. */
ref.action = BTRFS_DROP_DELAYED_REF;
ref.bytenr = old_bytenr;
ref.num_bytes = blocksize;
ref.parent = 0;
ref.owning_root = 0;
ref.ref_root = btrfs_root_id(dest);
btrfs_init_tree_ref(&ref, level - 1, 0, true);
ret = btrfs_free_extent(trans, &ref); if (ret) {
btrfs_abort_transaction(trans, ret); break;
}
btrfs_unlock_up_safe(path, 0);
ret = level; break;
}
btrfs_tree_unlock(parent);
free_extent_buffer(parent); return ret;
}
/* * helper to find next relocated block in reloc tree
*/ static noinline_for_stack int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, int *level)
{ struct extent_buffer *eb; int i;
u64 last_snapshot;
u32 nritems;
list_del_init(&root->reloc_dirty_list);
root->reloc_root = NULL; /* * Need barrier to ensure clear_bit() only happens after * root->reloc_root = NULL. Pairs with have_reloc_root.
*/
smp_wmb();
clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); if (reloc_root) { /* * btrfs_drop_snapshot drops our ref we hold for * ->reloc_root. If it fails however we must * drop the ref ourselves.
*/
ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); if (ret2 < 0) {
btrfs_put_root(reloc_root); if (!ret)
ret = ret2;
}
}
btrfs_put_root(root);
} else { /* Orphan reloc tree, just clean it up */
ret2 = btrfs_drop_snapshot(root, 0, 1); if (ret2 < 0) {
btrfs_put_root(root); if (!ret)
ret = ret2;
}
}
} return ret;
}
/* * merge the relocated tree blocks in reloc tree with corresponding * fs tree.
*/ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, struct btrfs_root *root)
{ struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; struct btrfs_key key; struct btrfs_key next_key; struct btrfs_trans_handle *trans = NULL; struct btrfs_root *reloc_root; struct btrfs_root_item *root_item; struct btrfs_path *path; struct extent_buffer *leaf; int reserve_level; int level; int max_level; int replaced = 0; int ret = 0;
u32 min_reserved;
path = btrfs_alloc_path(); if (!path) return -ENOMEM;
path->reada = READA_FORWARD;
/* * In merge_reloc_root(), we modify the upper level pointer to swap the * tree blocks between reloc tree and subvolume tree. Thus for tree * block COW, we COW at most from level 1 to root level for each tree. * * Thus the needed metadata size is at most root_level * nodesize, * and * 2 since we have two trees to COW.
*/
reserve_level = max_t(int, 1, btrfs_root_level(root_item));
min_reserved = fs_info->nodesize * reserve_level * 2;
memset(&next_key, 0, sizeof(next_key));
while (1) {
ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
min_reserved,
BTRFS_RESERVE_FLUSH_LIMIT); if (ret) goto out;
trans = btrfs_start_transaction(root, 0); if (IS_ERR(trans)) {
ret = PTR_ERR(trans);
trans = NULL; goto out;
}
/* * At this point we no longer have a reloc_control, so we can't * depend on btrfs_init_reloc_root to update our last_trans. * * But that's ok, we started the trans handle on our * corresponding fs_root, which means it's been added to the * dirty list. At commit time we'll still call * btrfs_update_reloc_root() and update our root item * appropriately.
*/
btrfs_set_root_last_trans(reloc_root, trans->transid);
trans->block_rsv = rc->block_rsv;
replaced = 0;
max_level = level;
ret = walk_down_reloc_tree(reloc_root, path, &level); if (ret < 0) goto out; if (ret > 0) break;
if (!find_next_key(path, level, &key) &&
btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
ret = 0;
} else {
ret = replace_path(trans, rc, root, reloc_root, path,
&next_key, level, max_level);
} if (ret < 0) goto out; if (ret > 0) {
level = ret;
btrfs_node_key_to_cpu(path->nodes[level], &key,
path->slots[level]);
replaced = 1;
}
ret = walk_up_reloc_tree(reloc_root, path, &level); if (ret > 0) break;
BUG_ON(level == 0); /* * save the merging progress in the drop_progress. * this is OK since root refs == 1 in this case.
*/
btrfs_node_key(path->nodes[level], &root_item->drop_progress,
path->slots[level]);
btrfs_set_root_drop_level(root_item, level);
btrfs_end_transaction_throttle(trans);
trans = NULL;
btrfs_btree_balance_dirty(fs_info);
if (replaced && rc->stage == UPDATE_DATA_PTRS)
invalidate_extent_cache(root, &key, &next_key);
}
/* * handle the case only one block in the fs tree need to be * relocated and the block is tree root.
*/
leaf = btrfs_lock_root_node(root);
ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
BTRFS_NESTING_COW);
btrfs_tree_unlock(leaf);
free_extent_buffer(leaf);
out:
btrfs_free_path(path);
if (ret == 0) {
ret = insert_dirty_subvol(trans, rc, root); if (ret)
btrfs_abort_transaction(trans, ret);
}
if (trans)
btrfs_end_transaction_throttle(trans);
btrfs_btree_balance_dirty(fs_info);
if (replaced && rc->stage == UPDATE_DATA_PTRS)
invalidate_extent_cache(root, &key, &next_key);
again: if (!err) {
num_bytes = rc->merging_rsv_size;
ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
BTRFS_RESERVE_FLUSH_ALL); if (ret)
err = ret;
}
trans = btrfs_join_transaction(rc->extent_root); if (IS_ERR(trans)) { if (!err)
btrfs_block_rsv_release(fs_info, rc->block_rsv,
num_bytes, NULL); return PTR_ERR(trans);
}
if (!err) { if (num_bytes != rc->merging_rsv_size) {
btrfs_end_transaction(trans);
btrfs_block_rsv_release(fs_info, rc->block_rsv,
num_bytes, NULL); goto again;
}
}
rc->merge_reloc_tree = true;
while (!list_empty(&rc->reloc_roots)) {
reloc_root = list_first_entry(&rc->reloc_roots, struct btrfs_root, root_list);
list_del_init(&reloc_root->root_list);
root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); if (IS_ERR(root)) { /* * Even if we have an error we need this reloc root * back on our list so we can clean up properly.
*/
list_add(&reloc_root->root_list, &reloc_roots);
btrfs_abort_transaction(trans, (int)PTR_ERR(root)); if (!err)
err = PTR_ERR(root); break;
}
if (unlikely(root->reloc_root != reloc_root)) { if (root->reloc_root) {
btrfs_err(fs_info, "reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu",
btrfs_root_id(root),
btrfs_root_id(root->reloc_root),
root->reloc_root->root_key.type,
root->reloc_root->root_key.offset,
btrfs_root_generation(
&root->reloc_root->root_item),
btrfs_root_id(reloc_root),
reloc_root->root_key.type,
reloc_root->root_key.offset,
btrfs_root_generation(
&reloc_root->root_item));
} else {
btrfs_err(fs_info, "reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu",
btrfs_root_id(root),
btrfs_root_id(reloc_root),
reloc_root->root_key.type,
reloc_root->root_key.offset,
btrfs_root_generation(
&reloc_root->root_item));
}
list_add(&reloc_root->root_list, &reloc_roots);
btrfs_put_root(root);
btrfs_abort_transaction(trans, -EUCLEAN); if (!err)
err = -EUCLEAN; break;
}
/* * set reference count to 1, so btrfs_recover_relocation * knows it should resumes merging
*/ if (!err)
btrfs_set_root_refs(&reloc_root->root_item, 1);
ret = btrfs_update_reloc_root(trans, root);
/* * Even if we have an error we need this reloc root back on our * list so we can clean up properly.
*/
list_add(&reloc_root->root_list, &reloc_roots);
btrfs_put_root(root);
if (ret) {
btrfs_abort_transaction(trans, ret); if (!err)
err = ret; break;
}
}
list_splice(&reloc_roots, &rc->reloc_roots);
if (!err)
err = btrfs_commit_transaction(trans); else
btrfs_end_transaction(trans); return err;
}
static noinline_for_stack void merge_reloc_roots(struct reloc_control *rc)
{ struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; struct btrfs_root *root; struct btrfs_root *reloc_root;
LIST_HEAD(reloc_roots); int found = 0; int ret = 0;
again:
root = rc->extent_root;
/* * this serializes us with btrfs_record_root_in_transaction, * we have to make sure nobody is in the middle of * adding their roots to the list while we are * doing this splice
*/
mutex_lock(&fs_info->reloc_mutex);
list_splice_init(&rc->reloc_roots, &reloc_roots);
mutex_unlock(&fs_info->reloc_mutex);
while (!list_empty(&reloc_roots)) {
found = 1;
reloc_root = list_first_entry(&reloc_roots, struct btrfs_root, root_list);
root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false); if (btrfs_root_refs(&reloc_root->root_item) > 0) { if (WARN_ON(IS_ERR(root))) { /* * For recovery we read the fs roots on mount, * and if we didn't find the root then we marked * the reloc root as a garbage root. For normal * relocation obviously the root should exist in * memory. However there's no reason we can't * handle the error properly here just in case.
*/
ret = PTR_ERR(root); goto out;
} if (WARN_ON(root->reloc_root != reloc_root)) { /* * This can happen if on-disk metadata has some * corruption, e.g. bad reloc tree key offset.
*/
ret = -EINVAL; goto out;
}
ret = merge_reloc_root(rc, root);
btrfs_put_root(root); if (ret) { if (list_empty(&reloc_root->root_list))
list_add_tail(&reloc_root->root_list,
&reloc_roots); goto out;
}
} else { if (!IS_ERR(root)) { if (root->reloc_root == reloc_root) {
root->reloc_root = NULL;
btrfs_put_root(reloc_root);
}
clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
&root->state);
btrfs_put_root(root);
}
list_del_init(&reloc_root->root_list); /* Don't forget to queue this reloc root for cleanup */
list_add_tail(&reloc_root->reloc_dirty_list,
&rc->dirty_subvol_roots);
}
}
if (found) {
found = 0; goto again;
}
out: if (ret) {
btrfs_handle_fs_error(fs_info, ret, NULL);
free_reloc_roots(&reloc_roots);
/* new reloc root may be added */
mutex_lock(&fs_info->reloc_mutex);
list_splice_init(&rc->reloc_roots, &reloc_roots);
mutex_unlock(&fs_info->reloc_mutex);
free_reloc_roots(&reloc_roots);
}
/* * We used to have * * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); * * here, but it's wrong. If we fail to start the transaction in * prepare_to_merge() we will have only 0 ref reloc roots, none of which * have actually been removed from the reloc_root_tree rb tree. This is * fine because we're bailing here, and we hold a reference on the root * for the list that holds it, so these roots will be cleaned up when we * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root * will be cleaned up on unmount. * * The remaining nodes will be cleaned up by free_reloc_control.
*/
}
/* * This should succeed, since we can't have a reloc root without having * already looked up the actual root and created the reloc root for this * root. * * However if there's some sort of corruption where we have a ref to a * reloc root without a corresponding root this could return ENOENT.
*/ if (IS_ERR(root)) {
DEBUG_WARN("error %ld reading root for reloc root", PTR_ERR(root)); return PTR_ERR(root);
} if (root->reloc_root != reloc_root) {
DEBUG_WARN("unexpected reloc root found");
btrfs_err(fs_info, "root %llu has two reloc roots associated with it",
reloc_root->root_key.offset);
btrfs_put_root(root); return -EUCLEAN;
}
ret = btrfs_record_root_in_trans(trans, root);
btrfs_put_root(root);
return ret;
}
static noinline_for_stack struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_backref_node *node, struct btrfs_backref_edge *edges[])
{ struct btrfs_backref_node *next; struct btrfs_root *root; int index = 0; int ret;
next = walk_up_backref(node, edges, &index);
root = next->root;
/* * If there is no root, then our references for this block are * incomplete, as we should be able to walk all the way up to a block * that is owned by a root. * * This path is only for SHAREABLE roots, so if we come upon a * non-SHAREABLE root then we have backrefs that resolve improperly. * * Both of these cases indicate file system corruption, or a bug in the * backref walking code.
*/ if (unlikely(!root)) {
btrfs_err(trans->fs_info, "bytenr %llu doesn't have a backref path ending in a root",
node->bytenr); return ERR_PTR(-EUCLEAN);
} if (unlikely(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))) {
btrfs_err(trans->fs_info, "bytenr %llu has multiple refs with one ending in a non-shareable root",
node->bytenr); return ERR_PTR(-EUCLEAN);
}
if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) {
ret = record_reloc_root_in_trans(trans, root); if (ret) return ERR_PTR(ret); goto found;
}
ret = btrfs_record_root_in_trans(trans, root); if (ret) return ERR_PTR(ret);
root = root->reloc_root;
/* * We could have raced with another thread which failed, so * root->reloc_root may not be set, return ENOENT in this case.
*/ if (!root) return ERR_PTR(-ENOENT);
if (next->new_bytenr) { /* * We just created the reloc root, so we shouldn't have * ->new_bytenr set yet. If it is then we have multiple roots * pointing at the same bytenr which indicates corruption, or * we've made a mistake in the backref walking code.
*/
ASSERT(next->new_bytenr == 0);
btrfs_err(trans->fs_info, "bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
node->bytenr, next->bytenr); return ERR_PTR(-EUCLEAN);
}
next->new_bytenr = root->node->start;
btrfs_put_root(next->root);
next->root = btrfs_grab_root(root);
ASSERT(next->root);
mark_block_processed(rc, next);
found:
next = node; /* setup backref node path for btrfs_reloc_cow_block */ while (1) {
rc->backref_cache.path[next->level] = next; if (--index < 0) break;
next = edges[index]->node[UPPER];
} return root;
}
/* * Select a tree root for relocation. * * Return NULL if the block is not shareable. We should use do_relocation() in * this case. * * Return a tree root pointer if the block is shareable. * Return -ENOENT if the block is root of reloc tree.
*/ static noinline_for_stack struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
{ struct btrfs_backref_node *next; struct btrfs_root *root; struct btrfs_root *fs_root = NULL; struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; int index = 0;
next = node; while (1) {
cond_resched();
next = walk_up_backref(next, edges, &index);
root = next->root;
/* * This can occur if we have incomplete extent refs leading all * the way up a particular path, in this case return -EUCLEAN.
*/ if (!root) return ERR_PTR(-EUCLEAN);
/* No other choice for non-shareable tree */ if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) return root;
if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID)
fs_root = root;
if (next != node) return NULL;
next = walk_down_backref(edges, &index); if (!next || next->level <= node->level) break;
}
if (!fs_root) return ERR_PTR(-ENOENT); return fs_root;
}
/* * We are under a transaction here so we can only do limited flushing. * If we get an enospc just kick back -EAGAIN so we know to drop the * transaction and try to refill when we can flush all the things.
*/
ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
BTRFS_RESERVE_FLUSH_LIMIT); if (ret) {
u64 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
while (tmp <= rc->reserved_bytes)
tmp <<= 1; /* * only one thread can access block_rsv at this point, * so we don't need hold lock to protect block_rsv. * we expand more reservation size here to allow enough * space for relocation and we will return earlier in * enospc case.
*/
rc->block_rsv->size = tmp + fs_info->nodesize *
RELOCATION_RESERVED_NODES; return -EAGAIN;
}
/* * relocate a block tree, and then update pointers in upper level * blocks that reference the block to point to the new location. * * if called by link_to_upper, the block has already been relocated. * in that case this function just updates pointers.
*/ staticint do_relocation(struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_backref_node *node, struct btrfs_key *key, struct btrfs_path *path, int lowest)
{ struct btrfs_backref_node *upper; struct btrfs_backref_edge *edge; struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1]; struct btrfs_root *root; struct extent_buffer *eb;
u32 blocksize;
u64 bytenr; int slot; int ret = 0;
/* * If we are lowest then this is the first time we're processing this * block, and thus shouldn't have an eb associated with it yet.
*/
ASSERT(!lowest || !node->eb);
/* * helper function to relocate a tree block
*/ staticint relocate_tree_block(struct btrfs_trans_handle *trans, struct reloc_control *rc, struct btrfs_backref_node *node, struct btrfs_key *key, struct btrfs_path *path)
{ struct btrfs_root *root; int ret = 0;
if (!node) return 0;
/* * If we fail here we want to drop our backref_node because we are going * to start over and regenerate the tree for it.
*/
ret = reserve_metadata_space(trans, rc, node); if (ret) goto out;
BUG_ON(node->processed);
root = select_one_root(node); if (IS_ERR(root)) {
ret = PTR_ERR(root);
/* See explanation in select_one_root for the -EUCLEAN case. */
ASSERT(ret == -ENOENT); if (ret == -ENOENT) {
ret = 0;
update_processed_blocks(rc, node);
} goto out;
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.