X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=fs%2Fbtrfs%2Fextent-tree.c;h=72a2b9c28e9fc7a6a7aacce09469cb6414215635;hb=2da8ca26c6bfad685bfddf39728eac1c83906aa9;hp=0d73a53c67633e38f2805c29f838e4ee40090df5;hpb=4366211ccd050e00674b3276b5ed81981b2d7793;p=safe%2Fjmp%2Flinux-2.6 diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 0d73a53..72a2b9c 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -19,39 +19,55 @@ #include #include #include +#include +#include +#include +#include "compat.h" #include "hash.h" -#include "crc32c.h" #include "ctree.h" #include "disk-io.h" #include "print-tree.h" #include "transaction.h" #include "volumes.h" #include "locking.h" -#include "ref-cache.h" +#include "free-space-cache.h" -#define PENDING_EXTENT_INSERT 0 -#define PENDING_EXTENT_DELETE 1 -#define PENDING_BACKREF_UPDATE 2 +static int update_reserved_extents(struct btrfs_root *root, + u64 bytenr, u64 num, int reserve); +static int update_block_group(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, int alloc, + int mark_free); +static int __btrfs_free_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 owner_objectid, + u64 owner_offset, int refs_to_drop, + struct btrfs_delayed_extent_op *extra_op); +static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, + struct extent_buffer *leaf, + struct btrfs_extent_item *ei); +static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 parent, u64 root_objectid, + u64 flags, u64 owner, u64 offset, + struct btrfs_key *ins, int ref_mod); +static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 parent, u64 root_objectid, + u64 flags, struct btrfs_disk_key *key, + int level, struct btrfs_key *ins); -struct pending_extent_op { - int type; - u64 bytenr; - u64 num_bytes; - u64 parent; - u64 orig_parent; - u64 generation; - u64 orig_generation; - int level; -}; +static int do_chunk_alloc(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, u64 alloc_bytes, + u64 flags, int force); -static int finish_current_insert(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root, int all); -static int del_pending_extents(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root, int all); -static struct btrfs_block_group_cache * -__btrfs_find_block_group(struct btrfs_root *root, - struct btrfs_block_group_cache *hint, - u64 search_start, int data, int owner); +static noinline int +block_group_cache_done(struct btrfs_block_group_cache *cache) +{ + smp_mb(); + return cache->cached == BTRFS_CACHE_FINISHED; +} static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) { @@ -62,7 +78,7 @@ static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) * this adds the block group to the fs_info rb tree for the block group * cache */ -int btrfs_add_block_group_cache(struct btrfs_fs_info *info, +static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, struct btrfs_block_group_cache *block_group) { struct rb_node **p; @@ -130,27 +146,78 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, break; } } + if (ret) + atomic_inc(&ret->count); spin_unlock(&info->block_group_cache_lock); return ret; } /* + * We always set EXTENT_LOCKED for the super mirror extents so we don't + * overwrite them, so those bits need to be unset. Also, if we are unmounting + * with pinned extents still sitting there because we had a block group caching, + * we need to clear those now, since we are done. + */ +void btrfs_free_pinned_extents(struct btrfs_fs_info *info) +{ + u64 start, end, last = 0; + int ret; + + while (1) { + ret = find_first_extent_bit(&info->pinned_extents, last, + &start, &end, + EXTENT_LOCKED|EXTENT_DIRTY); + if (ret) + break; + + clear_extent_bits(&info->pinned_extents, start, end, + EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS); + last = end+1; + } +} + +static int remove_sb_from_cache(struct btrfs_root *root, + struct btrfs_block_group_cache *cache) +{ + struct btrfs_fs_info *fs_info = root->fs_info; + u64 bytenr; + u64 *logical; + int stripe_len; + int i, nr, ret; + + for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { + bytenr = btrfs_sb_offset(i); + ret = btrfs_rmap_block(&root->fs_info->mapping_tree, + cache->key.objectid, bytenr, + 0, &logical, &nr, &stripe_len); + BUG_ON(ret); + while (nr--) { + try_lock_extent(&fs_info->pinned_extents, + logical[nr], + logical[nr] + stripe_len - 1, GFP_NOFS); + } + kfree(logical); + } + + return 0; +} + +/* * this is only called by cache_block_group, since we could have freed extents * we need to check the pinned_extents for any extents that can't be used yet * since their free space will be released as soon as the transaction commits. */ -static int add_new_free_space(struct btrfs_block_group_cache *block_group, +static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, struct btrfs_fs_info *info, u64 start, u64 end) { - u64 extent_start, extent_end, size; + u64 extent_start, extent_end, size, total_added = 0; int ret; - mutex_lock(&info->pinned_mutex); while (start < end) { ret = find_first_extent_bit(&info->pinned_extents, start, &extent_start, &extent_end, - EXTENT_DIRTY); + EXTENT_DIRTY|EXTENT_LOCKED); if (ret) break; @@ -158,8 +225,9 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group, start = extent_end + 1; } else if (extent_start > start && extent_start < end) { size = extent_start - start; - ret = btrfs_add_free_space_lock(block_group, start, - size); + total_added += size; + ret = btrfs_add_free_space(block_group, start, + size); BUG_ON(ret); start = extent_end + 1; } else { @@ -169,73 +237,93 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group, if (start < end) { size = end - start; - ret = btrfs_add_free_space_lock(block_group, start, size); + total_added += size; + ret = btrfs_add_free_space(block_group, start, size); BUG_ON(ret); } - mutex_unlock(&info->pinned_mutex); - return 0; + return total_added; } -static int cache_block_group(struct btrfs_root *root, - struct btrfs_block_group_cache *block_group) +static int caching_kthread(void *data) { + struct btrfs_block_group_cache *block_group = data; + struct btrfs_fs_info *fs_info = block_group->fs_info; + u64 last = 0; struct btrfs_path *path; int ret = 0; struct btrfs_key key; struct extent_buffer *leaf; int slot; - u64 last = 0; - u64 first_free; - int found = 0; - - if (!block_group) - return 0; - - root = root->fs_info->extent_root; + u64 total_found = 0; - if (block_group->cached) - return 0; + BUG_ON(!fs_info); path = btrfs_alloc_path(); if (!path) return -ENOMEM; - path->reada = 2; + atomic_inc(&block_group->space_info->caching_threads); + last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); /* - * we get into deadlocks with paths held by callers of this function. - * since the alloc_mutex is protecting things right now, just - * skip the locking here + * We don't want to deadlock with somebody trying to allocate a new + * extent for the extent root while also trying to search the extent + * root to add free space. So we skip locking and search the commit + * root, since its read-only */ path->skip_locking = 1; - first_free = max_t(u64, block_group->key.objectid, - BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE); - key.objectid = block_group->key.objectid; + path->search_commit_root = 1; + path->reada = 2; + + key.objectid = last; key.offset = 0; btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); - ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (ret < 0) - goto err; - ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY); +again: + /* need to make sure the commit_root doesn't disappear */ + down_read(&fs_info->extent_commit_sem); + + ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); if (ret < 0) goto err; - if (ret == 0) { - leaf = path->nodes[0]; - btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); - if (key.objectid + key.offset > first_free) - first_free = key.objectid + key.offset; - } - while(1) { + + while (1) { + smp_mb(); + if (block_group->fs_info->closing > 1) { + last = (u64)-1; + break; + } + leaf = path->nodes[0]; slot = path->slots[0]; if (slot >= btrfs_header_nritems(leaf)) { - ret = btrfs_next_leaf(root, path); + ret = btrfs_next_leaf(fs_info->extent_root, path); if (ret < 0) goto err; - if (ret == 0) - continue; - else + else if (ret) break; + + if (need_resched() || + btrfs_transaction_in_commit(fs_info)) { + leaf = path->nodes[0]; + + /* this shouldn't happen, but if the + * leaf is empty just move on. + */ + if (btrfs_header_nritems(leaf) == 0) + break; + /* + * we need to copy the key out so that + * we are sure the next search advances + * us forward in the btree. + */ + btrfs_item_key_to_cpu(leaf, &key, 0); + btrfs_release_path(fs_info->extent_root, path); + up_read(&fs_info->extent_commit_sem); + schedule_timeout(1); + goto again; + } + + continue; } btrfs_item_key_to_cpu(leaf, &key, slot); if (key.objectid < block_group->key.objectid) @@ -246,40 +334,67 @@ static int cache_block_group(struct btrfs_root *root, break; if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { - if (!found) { - last = first_free; - found = 1; - } - - add_new_free_space(block_group, root->fs_info, last, - key.objectid); - + total_found += add_new_free_space(block_group, + fs_info, last, + key.objectid); last = key.objectid + key.offset; } + + if (total_found > (1024 * 1024 * 2)) { + total_found = 0; + wake_up(&block_group->caching_q); + } next: path->slots[0]++; } + ret = 0; - if (!found) - last = first_free; + total_found += add_new_free_space(block_group, fs_info, last, + block_group->key.objectid + + block_group->key.offset); - add_new_free_space(block_group, root->fs_info, last, - block_group->key.objectid + - block_group->key.offset); + spin_lock(&block_group->lock); + block_group->cached = BTRFS_CACHE_FINISHED; + spin_unlock(&block_group->lock); - block_group->cached = 1; - ret = 0; err: btrfs_free_path(path); + up_read(&fs_info->extent_commit_sem); + atomic_dec(&block_group->space_info->caching_threads); + wake_up(&block_group->caching_q); + + return 0; +} + +static int cache_block_group(struct btrfs_block_group_cache *cache) +{ + struct task_struct *tsk; + int ret = 0; + + spin_lock(&cache->lock); + if (cache->cached != BTRFS_CACHE_NO) { + spin_unlock(&cache->lock); + return ret; + } + cache->cached = BTRFS_CACHE_STARTED; + spin_unlock(&cache->lock); + + tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n", + cache->key.objectid); + if (IS_ERR(tsk)) { + ret = PTR_ERR(tsk); + printk(KERN_ERR "error running thread %d\n", ret); + BUG(); + } + return ret; } /* * return the block group that starts at or after bytenr */ -struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct - btrfs_fs_info *info, - u64 bytenr) +static struct btrfs_block_group_cache * +btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) { struct btrfs_block_group_cache *cache; @@ -289,11 +404,11 @@ struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct } /* - * return the block group that contains teh given bytenr + * return the block group that contains the given bytenr */ -struct btrfs_block_group_cache *btrfs_lookup_block_group(struct - btrfs_fs_info *info, - u64 bytenr) +struct btrfs_block_group_cache *btrfs_lookup_block_group( + struct btrfs_fs_info *info, + u64 bytenr) { struct btrfs_block_group_cache *cache; @@ -302,20 +417,44 @@ struct btrfs_block_group_cache *btrfs_lookup_block_group(struct return cache; } +void btrfs_put_block_group(struct btrfs_block_group_cache *cache) +{ + if (atomic_dec_and_test(&cache->count)) + kfree(cache); +} + static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, u64 flags) { struct list_head *head = &info->space_info; - struct list_head *cur; struct btrfs_space_info *found; - list_for_each(cur, head) { - found = list_entry(cur, struct btrfs_space_info, list); - if (found->flags == flags) + + rcu_read_lock(); + list_for_each_entry_rcu(found, head, list) { + if (found->flags == flags) { + rcu_read_unlock(); return found; + } } + rcu_read_unlock(); return NULL; } +/* + * after adding space to the filesystem, we need to clear the full flags + * on all the space infos. + */ +void btrfs_clear_space_info_full(struct btrfs_fs_info *info) +{ + struct list_head *head = &info->space_info; + struct btrfs_space_info *found; + + rcu_read_lock(); + list_for_each_entry_rcu(found, head, list) + found->full = 0; + rcu_read_unlock(); +} + static u64 div_factor(u64 num, int factor) { if (factor == 10) @@ -325,54 +464,16 @@ static u64 div_factor(u64 num, int factor) return num; } -static struct btrfs_block_group_cache * -__btrfs_find_block_group(struct btrfs_root *root, - struct btrfs_block_group_cache *hint, - u64 search_start, int data, int owner) +u64 btrfs_find_block_group(struct btrfs_root *root, + u64 search_start, u64 search_hint, int owner) { struct btrfs_block_group_cache *cache; - struct btrfs_block_group_cache *found_group = NULL; - struct btrfs_fs_info *info = root->fs_info; u64 used; - u64 last = 0; - u64 free_check; + u64 last = max(search_hint, search_start); + u64 group_start = 0; int full_search = 0; - int factor = 10; + int factor = 9; int wrapped = 0; - - if (data & BTRFS_BLOCK_GROUP_METADATA) - factor = 9; - - if (search_start) { - struct btrfs_block_group_cache *shint; - shint = btrfs_lookup_first_block_group(info, search_start); - if (shint && block_group_bits(shint, data) && !shint->ro) { - spin_lock(&shint->lock); - used = btrfs_block_group_used(&shint->item); - if (used + shint->pinned + shint->reserved < - div_factor(shint->key.offset, factor)) { - spin_unlock(&shint->lock); - return shint; - } - spin_unlock(&shint->lock); - } - } - if (hint && !hint->ro && block_group_bits(hint, data)) { - spin_lock(&hint->lock); - used = btrfs_block_group_used(&hint->item); - if (used + hint->pinned + hint->reserved < - div_factor(hint->key.offset, factor)) { - spin_unlock(&hint->lock); - return hint; - } - spin_unlock(&hint->lock); - last = hint->key.objectid + hint->key.offset; - } else { - if (hint) - last = max(hint->key.objectid, search_start); - else - last = search_start; - } again: while (1) { cache = btrfs_lookup_first_block_group(root->fs_info, last); @@ -383,16 +484,18 @@ again: last = cache->key.objectid + cache->key.offset; used = btrfs_block_group_used(&cache->item); - if (!cache->ro && block_group_bits(cache, data)) { - free_check = div_factor(cache->key.offset, factor); + if ((full_search || !cache->ro) && + block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { if (used + cache->pinned + cache->reserved < - free_check) { - found_group = cache; + div_factor(cache->key.offset, factor)) { + group_start = cache->key.objectid; spin_unlock(&cache->lock); + btrfs_put_block_group(cache); goto found; } } spin_unlock(&cache->lock); + btrfs_put_block_group(cache); cond_resched(); } if (!wrapped) { @@ -407,18 +510,7 @@ again: goto again; } found: - return found_group; -} - -struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, - struct btrfs_block_group_cache - *hint, u64 search_start, - int data, int owner) -{ - - struct btrfs_block_group_cache *ret; - ret = __btrfs_find_block_group(root, hint, search_start, data, owner); - return ret; + return group_start; } /* simple helper to search for an existing extent at a given offset */ @@ -453,424 +545,1634 @@ int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) * maintenance. This is actually the same as #2, but with a slightly * different use case. * + * There are two kinds of back refs. The implicit back refs is optimized + * for pointers in non-shared tree blocks. For a given pointer in a block, + * back refs of this kind provide information about the block's owner tree + * and the pointer's key. These information allow us to find the block by + * b-tree searching. The full back refs is for pointers in tree blocks not + * referenced by their owner trees. The location of tree block is recorded + * in the back refs. Actually the full back refs is generic, and can be + * used in all cases the implicit back refs is used. The major shortcoming + * of the full back refs is its overhead. Every time a tree block gets + * COWed, we have to update back refs entry for all pointers in it. + * + * For a newly allocated tree block, we use implicit back refs for + * pointers in it. This means most tree related operations only involve + * implicit back refs. For a tree block created in old transaction, the + * only way to drop a reference to it is COW it. So we can detect the + * event that tree block loses its owner tree's reference and do the + * back refs conversion. + * + * When a tree block is COW'd through a tree, there are four cases: + * + * The reference count of the block is one and the tree is the block's + * owner tree. Nothing to do in this case. + * + * The reference count of the block is one and the tree is not the + * block's owner tree. In this case, full back refs is used for pointers + * in the block. Remove these full back refs, add implicit back refs for + * every pointers in the new block. + * + * The reference count of the block is greater than one and the tree is + * the block's owner tree. In this case, implicit back refs is used for + * pointers in the block. Add full back refs for every pointers in the + * block, increase lower level extents' reference counts. The original + * implicit back refs are entailed to the new block. + * + * The reference count of the block is greater than one and the tree is + * not the block's owner tree. Add implicit back refs for every pointer in + * the new block, increase lower level extents' reference count. + * + * Back Reference Key composing: + * + * The key objectid corresponds to the first byte in the extent, + * The key type is used to differentiate between types of back refs. + * There are different meanings of the key offset for different types + * of back refs. + * * File extents can be referenced by: * * - multiple snapshots, subvolumes, or different generations in one subvol * - different files inside a single subvolume * - different offsets inside a file (bookend extents in file.c) * - * The extent ref structure has fields for: + * The extent ref structure for the implicit back refs has fields for: * * - Objectid of the subvolume root - * - Generation number of the tree holding the reference * - objectid of the file holding the reference - * - number of references holding by parent node (alway 1 for tree blocks) + * - original offset in the file + * - how many bookend extents * - * Btree leaf may hold multiple references to a file extent. In most cases, - * these references are from same file and the corresponding offsets inside - * the file are close together. + * The key offset for the implicit back refs is hash of the first + * three fields. * - * When a file extent is allocated the fields are filled in: - * (root_key.objectid, trans->transid, inode objectid, 1) + * The extent ref structure for the full back refs has field for: * - * When a leaf is cow'd new references are added for every file extent found - * in the leaf. It looks similar to the create case, but trans->transid will - * be different when the block is cow'd. + * - number of pointers in the tree leaf * - * (root_key.objectid, trans->transid, inode objectid, - * number of references in the leaf) + * The key offset for the implicit back refs is the first byte of + * the tree leaf * - * When a file extent is removed either during snapshot deletion or - * file truncation, we find the corresponding back reference and check - * the following fields: + * When a file extent is allocated, The implicit back refs is used. + * the fields are filled in: * - * (btrfs_header_owner(leaf), btrfs_header_generation(leaf), - * inode objectid) + * (root_key.objectid, inode objectid, offset in file, 1) * - * Btree extents can be referenced by: - * - * - Different subvolumes - * - Different generations of the same subvolume - * - * When a tree block is created, back references are inserted: + * When a file extent is removed file truncation, we find the + * corresponding implicit back refs and check the following fields: * - * (root->root_key.objectid, trans->transid, level, 1) + * (btrfs_header_owner(leaf), inode objectid, offset in file) * - * When a tree block is cow'd, new back references are added for all the - * blocks it points to. If the tree block isn't in reference counted root, - * the old back references are removed. These new back references are of - * the form (trans->transid will have increased since creation): - * - * (root->root_key.objectid, trans->transid, level, 1) - * - * When a backref is in deleting, the following fields are checked: + * Btree extents can be referenced by: * - * if backref was for a tree root: - * (btrfs_header_owner(itself), btrfs_header_generation(itself), level) - * else - * (btrfs_header_owner(parent), btrfs_header_generation(parent), level) + * - Different subvolumes * - * Back Reference Key composing: + * Both the implicit back refs and the full back refs for tree blocks + * only consist of key. The key offset for the implicit back refs is + * objectid of block's owner tree. The key offset for the full back refs + * is the first byte of parent block. * - * The key objectid corresponds to the first byte in the extent, the key - * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first - * byte of parent extent. If a extent is tree root, the key offset is set - * to the key objectid. + * When implicit back refs is used, information about the lowest key and + * level of the tree block are required. These information are stored in + * tree block info structure. */ -static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - struct btrfs_path *path, - u64 bytenr, u64 parent, - u64 ref_root, u64 ref_generation, - u64 owner_objectid, int del) +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 +static int convert_extent_item_v0(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 owner, u32 extra_size) { - struct btrfs_key key; - struct btrfs_extent_ref *ref; + struct btrfs_extent_item *item; + struct btrfs_extent_item_v0 *ei0; + struct btrfs_extent_ref_v0 *ref0; + struct btrfs_tree_block_info *bi; struct extent_buffer *leaf; - u64 ref_objectid; + struct btrfs_key key; + struct btrfs_key found_key; + u32 new_size = sizeof(*item); + u64 refs; int ret; - key.objectid = bytenr; - key.type = BTRFS_EXTENT_REF_KEY; - key.offset = parent; + leaf = path->nodes[0]; + BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); - ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1); - if (ret < 0) - goto out; - if (ret > 0) { - ret = -ENOENT; - goto out; - } + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + ei0 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_item_v0); + refs = btrfs_extent_refs_v0(leaf, ei0); - leaf = path->nodes[0]; - ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - ref_objectid = btrfs_ref_objectid(leaf, ref); - if (btrfs_ref_root(leaf, ref) != ref_root || - btrfs_ref_generation(leaf, ref) != ref_generation || - (ref_objectid != owner_objectid && - ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { - ret = -EIO; - WARN_ON(1); - goto out; + if (owner == (u64)-1) { + while (1) { + if (path->slots[0] >= btrfs_header_nritems(leaf)) { + ret = btrfs_next_leaf(root, path); + if (ret < 0) + return ret; + BUG_ON(ret > 0); + leaf = path->nodes[0]; + } + btrfs_item_key_to_cpu(leaf, &found_key, + path->slots[0]); + BUG_ON(key.objectid != found_key.objectid); + if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { + path->slots[0]++; + continue; + } + ref0 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_ref_v0); + owner = btrfs_ref_objectid_v0(leaf, ref0); + break; + } } - ret = 0; -out: - return ret; -} - -static int noinline insert_extent_backref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - struct btrfs_path *path, - u64 bytenr, u64 parent, - u64 ref_root, u64 ref_generation, - u64 owner_objectid) -{ - struct btrfs_key key; - struct extent_buffer *leaf; - struct btrfs_extent_ref *ref; - u32 num_refs; - int ret; + btrfs_release_path(root, path); - key.objectid = bytenr; - key.type = BTRFS_EXTENT_REF_KEY; - key.offset = parent; + if (owner < BTRFS_FIRST_FREE_OBJECTID) + new_size += sizeof(*bi); - ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref)); - if (ret == 0) { - leaf = path->nodes[0]; - ref = btrfs_item_ptr(leaf, path->slots[0], - struct btrfs_extent_ref); - btrfs_set_ref_root(leaf, ref, ref_root); - btrfs_set_ref_generation(leaf, ref, ref_generation); - btrfs_set_ref_objectid(leaf, ref, owner_objectid); - btrfs_set_ref_num_refs(leaf, ref, 1); - } else if (ret == -EEXIST) { - u64 existing_owner; - BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); - leaf = path->nodes[0]; - ref = btrfs_item_ptr(leaf, path->slots[0], - struct btrfs_extent_ref); - if (btrfs_ref_root(leaf, ref) != ref_root || - btrfs_ref_generation(leaf, ref) != ref_generation) { - ret = -EIO; - WARN_ON(1); - goto out; - } + new_size -= sizeof(*ei0); + ret = btrfs_search_slot(trans, root, &key, path, + new_size + extra_size, 1); + if (ret < 0) + return ret; + BUG_ON(ret); - num_refs = btrfs_ref_num_refs(leaf, ref); - BUG_ON(num_refs == 0); - btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); + ret = btrfs_extend_item(trans, root, path, new_size); + BUG_ON(ret); - existing_owner = btrfs_ref_objectid(leaf, ref); - if (existing_owner != owner_objectid && - existing_owner != BTRFS_MULTIPLE_OBJECTIDS) { - btrfs_set_ref_objectid(leaf, ref, - BTRFS_MULTIPLE_OBJECTIDS); - } - ret = 0; + leaf = path->nodes[0]; + item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + btrfs_set_extent_refs(leaf, item, refs); + /* FIXME: get real generation */ + btrfs_set_extent_generation(leaf, item, 0); + if (owner < BTRFS_FIRST_FREE_OBJECTID) { + btrfs_set_extent_flags(leaf, item, + BTRFS_EXTENT_FLAG_TREE_BLOCK | + BTRFS_BLOCK_FLAG_FULL_BACKREF); + bi = (struct btrfs_tree_block_info *)(item + 1); + /* FIXME: get first key of the block */ + memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); + btrfs_set_tree_block_level(leaf, bi, (int)owner); } else { - goto out; + btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); } - btrfs_mark_buffer_dirty(path->nodes[0]); -out: - btrfs_release_path(root, path); - return ret; + btrfs_mark_buffer_dirty(leaf); + return 0; } +#endif -static int noinline remove_extent_backref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - struct btrfs_path *path) +static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) { - struct extent_buffer *leaf; - struct btrfs_extent_ref *ref; - u32 num_refs; - int ret = 0; + u32 high_crc = ~(u32)0; + u32 low_crc = ~(u32)0; + __le64 lenum; + + lenum = cpu_to_le64(root_objectid); + high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); + lenum = cpu_to_le64(owner); + low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); + lenum = cpu_to_le64(offset); + low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); + + return ((u64)high_crc << 31) ^ (u64)low_crc; +} - leaf = path->nodes[0]; - ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - num_refs = btrfs_ref_num_refs(leaf, ref); - BUG_ON(num_refs == 0); - num_refs -= 1; - if (num_refs == 0) { - ret = btrfs_del_item(trans, root, path); - } else { - btrfs_set_ref_num_refs(leaf, ref, num_refs); - btrfs_mark_buffer_dirty(leaf); - } - btrfs_release_path(root, path); - return ret; +static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, + struct btrfs_extent_data_ref *ref) +{ + return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), + btrfs_extent_data_ref_objectid(leaf, ref), + btrfs_extent_data_ref_offset(leaf, ref)); } -static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr, - u64 orig_parent, u64 parent, - u64 orig_root, u64 ref_root, - u64 orig_generation, u64 ref_generation, - u64 owner_objectid) +static int match_extent_data_ref(struct extent_buffer *leaf, + struct btrfs_extent_data_ref *ref, + u64 root_objectid, u64 owner, u64 offset) { - int ret; - struct btrfs_root *extent_root = root->fs_info->extent_root; - struct btrfs_path *path; + if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || + btrfs_extent_data_ref_objectid(leaf, ref) != owner || + btrfs_extent_data_ref_offset(leaf, ref) != offset) + return 0; + return 1; +} - if (root == root->fs_info->extent_root) { - struct pending_extent_op *extent_op; - u64 num_bytes; - - BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL); - num_bytes = btrfs_level_size(root, (int)owner_objectid); - mutex_lock(&root->fs_info->extent_ins_mutex); - if (test_range_bit(&root->fs_info->extent_ins, bytenr, - bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { - u64 priv; - ret = get_state_private(&root->fs_info->extent_ins, - bytenr, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *) - (unsigned long)priv; - BUG_ON(extent_op->parent != orig_parent); - BUG_ON(extent_op->generation != orig_generation); +static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 parent, + u64 root_objectid, + u64 owner, u64 offset) +{ + struct btrfs_key key; + struct btrfs_extent_data_ref *ref; + struct extent_buffer *leaf; + u32 nritems; + int ret; + int recow; + int err = -ENOENT; - extent_op->parent = parent; - extent_op->generation = ref_generation; - } else { - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); - - extent_op->type = PENDING_BACKREF_UPDATE; - extent_op->bytenr = bytenr; - extent_op->num_bytes = num_bytes; - extent_op->parent = parent; - extent_op->orig_parent = orig_parent; - extent_op->generation = ref_generation; - extent_op->orig_generation = orig_generation; - extent_op->level = (int)owner_objectid; - - set_extent_bits(&root->fs_info->extent_ins, - bytenr, bytenr + num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->extent_ins, - bytenr, (unsigned long)extent_op); - } - mutex_unlock(&root->fs_info->extent_ins_mutex); - return 0; + key.objectid = bytenr; + if (parent) { + key.type = BTRFS_SHARED_DATA_REF_KEY; + key.offset = parent; + } else { + key.type = BTRFS_EXTENT_DATA_REF_KEY; + key.offset = hash_extent_data_ref(root_objectid, + owner, offset); + } +again: + recow = 0; + ret = btrfs_search_slot(trans, root, &key, path, -1, 1); + if (ret < 0) { + err = ret; + goto fail; } + if (parent) { + if (!ret) + return 0; +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + key.type = BTRFS_EXTENT_REF_V0_KEY; + btrfs_release_path(root, path); + ret = btrfs_search_slot(trans, root, &key, path, -1, 1); + if (ret < 0) { + err = ret; + goto fail; + } + if (!ret) + return 0; +#endif + goto fail; + } + + leaf = path->nodes[0]; + nritems = btrfs_header_nritems(leaf); + while (1) { + if (path->slots[0] >= nritems) { + ret = btrfs_next_leaf(root, path); + if (ret < 0) + err = ret; + if (ret) + goto fail; + + leaf = path->nodes[0]; + nritems = btrfs_header_nritems(leaf); + recow = 1; + } + + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + if (key.objectid != bytenr || + key.type != BTRFS_EXTENT_DATA_REF_KEY) + goto fail; + + ref = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_data_ref); + + if (match_extent_data_ref(leaf, ref, root_objectid, + owner, offset)) { + if (recow) { + btrfs_release_path(root, path); + goto again; + } + err = 0; + break; + } + path->slots[0]++; + } +fail: + return err; +} + +static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 parent, + u64 root_objectid, u64 owner, + u64 offset, int refs_to_add) +{ + struct btrfs_key key; + struct extent_buffer *leaf; + u32 size; + u32 num_refs; + int ret; + + key.objectid = bytenr; + if (parent) { + key.type = BTRFS_SHARED_DATA_REF_KEY; + key.offset = parent; + size = sizeof(struct btrfs_shared_data_ref); + } else { + key.type = BTRFS_EXTENT_DATA_REF_KEY; + key.offset = hash_extent_data_ref(root_objectid, + owner, offset); + size = sizeof(struct btrfs_extent_data_ref); + } + + ret = btrfs_insert_empty_item(trans, root, path, &key, size); + if (ret && ret != -EEXIST) + goto fail; + + leaf = path->nodes[0]; + if (parent) { + struct btrfs_shared_data_ref *ref; + ref = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_shared_data_ref); + if (ret == 0) { + btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); + } else { + num_refs = btrfs_shared_data_ref_count(leaf, ref); + num_refs += refs_to_add; + btrfs_set_shared_data_ref_count(leaf, ref, num_refs); + } + } else { + struct btrfs_extent_data_ref *ref; + while (ret == -EEXIST) { + ref = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_data_ref); + if (match_extent_data_ref(leaf, ref, root_objectid, + owner, offset)) + break; + btrfs_release_path(root, path); + key.offset++; + ret = btrfs_insert_empty_item(trans, root, path, &key, + size); + if (ret && ret != -EEXIST) + goto fail; + + leaf = path->nodes[0]; + } + ref = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_data_ref); + if (ret == 0) { + btrfs_set_extent_data_ref_root(leaf, ref, + root_objectid); + btrfs_set_extent_data_ref_objectid(leaf, ref, owner); + btrfs_set_extent_data_ref_offset(leaf, ref, offset); + btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); + } else { + num_refs = btrfs_extent_data_ref_count(leaf, ref); + num_refs += refs_to_add; + btrfs_set_extent_data_ref_count(leaf, ref, num_refs); + } + } + btrfs_mark_buffer_dirty(leaf); + ret = 0; +fail: + btrfs_release_path(root, path); + return ret; +} + +static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + int refs_to_drop) +{ + struct btrfs_key key; + struct btrfs_extent_data_ref *ref1 = NULL; + struct btrfs_shared_data_ref *ref2 = NULL; + struct extent_buffer *leaf; + u32 num_refs = 0; + int ret = 0; + + leaf = path->nodes[0]; + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + + if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { + ref1 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_data_ref); + num_refs = btrfs_extent_data_ref_count(leaf, ref1); + } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { + ref2 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_shared_data_ref); + num_refs = btrfs_shared_data_ref_count(leaf, ref2); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { + struct btrfs_extent_ref_v0 *ref0; + ref0 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_ref_v0); + num_refs = btrfs_ref_count_v0(leaf, ref0); +#endif + } else { + BUG(); + } + + BUG_ON(num_refs < refs_to_drop); + num_refs -= refs_to_drop; + + if (num_refs == 0) { + ret = btrfs_del_item(trans, root, path); + } else { + if (key.type == BTRFS_EXTENT_DATA_REF_KEY) + btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); + else if (key.type == BTRFS_SHARED_DATA_REF_KEY) + btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + else { + struct btrfs_extent_ref_v0 *ref0; + ref0 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_ref_v0); + btrfs_set_ref_count_v0(leaf, ref0, num_refs); + } +#endif + btrfs_mark_buffer_dirty(leaf); + } + return ret; +} + +static noinline u32 extent_data_ref_count(struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref *iref) +{ + struct btrfs_key key; + struct extent_buffer *leaf; + struct btrfs_extent_data_ref *ref1; + struct btrfs_shared_data_ref *ref2; + u32 num_refs = 0; + + leaf = path->nodes[0]; + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); + if (iref) { + if (btrfs_extent_inline_ref_type(leaf, iref) == + BTRFS_EXTENT_DATA_REF_KEY) { + ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); + num_refs = btrfs_extent_data_ref_count(leaf, ref1); + } else { + ref2 = (struct btrfs_shared_data_ref *)(iref + 1); + num_refs = btrfs_shared_data_ref_count(leaf, ref2); + } + } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { + ref1 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_data_ref); + num_refs = btrfs_extent_data_ref_count(leaf, ref1); + } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { + ref2 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_shared_data_ref); + num_refs = btrfs_shared_data_ref_count(leaf, ref2); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { + struct btrfs_extent_ref_v0 *ref0; + ref0 = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_ref_v0); + num_refs = btrfs_ref_count_v0(leaf, ref0); +#endif + } else { + WARN_ON(1); + } + return num_refs; +} + +static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 parent, + u64 root_objectid) +{ + struct btrfs_key key; + int ret; + + key.objectid = bytenr; + if (parent) { + key.type = BTRFS_SHARED_BLOCK_REF_KEY; + key.offset = parent; + } else { + key.type = BTRFS_TREE_BLOCK_REF_KEY; + key.offset = root_objectid; + } + + ret = btrfs_search_slot(trans, root, &key, path, -1, 1); + if (ret > 0) + ret = -ENOENT; +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + if (ret == -ENOENT && parent) { + btrfs_release_path(root, path); + key.type = BTRFS_EXTENT_REF_V0_KEY; + ret = btrfs_search_slot(trans, root, &key, path, -1, 1); + if (ret > 0) + ret = -ENOENT; + } +#endif + return ret; +} + +static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 parent, + u64 root_objectid) +{ + struct btrfs_key key; + int ret; + + key.objectid = bytenr; + if (parent) { + key.type = BTRFS_SHARED_BLOCK_REF_KEY; + key.offset = parent; + } else { + key.type = BTRFS_TREE_BLOCK_REF_KEY; + key.offset = root_objectid; + } + + ret = btrfs_insert_empty_item(trans, root, path, &key, 0); + btrfs_release_path(root, path); + return ret; +} + +static inline int extent_ref_type(u64 parent, u64 owner) +{ + int type; + if (owner < BTRFS_FIRST_FREE_OBJECTID) { + if (parent > 0) + type = BTRFS_SHARED_BLOCK_REF_KEY; + else + type = BTRFS_TREE_BLOCK_REF_KEY; + } else { + if (parent > 0) + type = BTRFS_SHARED_DATA_REF_KEY; + else + type = BTRFS_EXTENT_DATA_REF_KEY; + } + return type; +} + +static int find_next_key(struct btrfs_path *path, int level, + struct btrfs_key *key) + +{ + for (; level < BTRFS_MAX_LEVEL; level++) { + if (!path->nodes[level]) + break; + if (path->slots[level] + 1 >= + btrfs_header_nritems(path->nodes[level])) + continue; + if (level == 0) + btrfs_item_key_to_cpu(path->nodes[level], key, + path->slots[level] + 1); + else + btrfs_node_key_to_cpu(path->nodes[level], key, + path->slots[level] + 1); + return 0; + } + return 1; +} + +/* + * look for inline back ref. if back ref is found, *ref_ret is set + * to the address of inline back ref, and 0 is returned. + * + * if back ref isn't found, *ref_ret is set to the address where it + * should be inserted, and -ENOENT is returned. + * + * if insert is true and there are too many inline back refs, the path + * points to the extent item, and -EAGAIN is returned. + * + * NOTE: inline back refs are ordered in the same way that back ref + * items in the tree are ordered. + */ +static noinline_for_stack +int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref **ref_ret, + u64 bytenr, u64 num_bytes, + u64 parent, u64 root_objectid, + u64 owner, u64 offset, int insert) +{ + struct btrfs_key key; + struct extent_buffer *leaf; + struct btrfs_extent_item *ei; + struct btrfs_extent_inline_ref *iref; + u64 flags; + u64 item_size; + unsigned long ptr; + unsigned long end; + int extra_size; + int type; + int want; + int ret; + int err = 0; + + key.objectid = bytenr; + key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = num_bytes; + + want = extent_ref_type(parent, owner); + if (insert) { + extra_size = btrfs_extent_inline_ref_size(want); + path->keep_locks = 1; + } else + extra_size = -1; + ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); + if (ret < 0) { + err = ret; + goto out; + } + BUG_ON(ret); + + leaf = path->nodes[0]; + item_size = btrfs_item_size_nr(leaf, path->slots[0]); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + if (item_size < sizeof(*ei)) { + if (!insert) { + err = -ENOENT; + goto out; + } + ret = convert_extent_item_v0(trans, root, path, owner, + extra_size); + if (ret < 0) { + err = ret; + goto out; + } + leaf = path->nodes[0]; + item_size = btrfs_item_size_nr(leaf, path->slots[0]); + } +#endif + BUG_ON(item_size < sizeof(*ei)); + + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + flags = btrfs_extent_flags(leaf, ei); + + ptr = (unsigned long)(ei + 1); + end = (unsigned long)ei + item_size; + + if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { + ptr += sizeof(struct btrfs_tree_block_info); + BUG_ON(ptr > end); + } else { + BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); + } + + err = -ENOENT; + while (1) { + if (ptr >= end) { + WARN_ON(ptr > end); + break; + } + iref = (struct btrfs_extent_inline_ref *)ptr; + type = btrfs_extent_inline_ref_type(leaf, iref); + if (want < type) + break; + if (want > type) { + ptr += btrfs_extent_inline_ref_size(type); + continue; + } + + if (type == BTRFS_EXTENT_DATA_REF_KEY) { + struct btrfs_extent_data_ref *dref; + dref = (struct btrfs_extent_data_ref *)(&iref->offset); + if (match_extent_data_ref(leaf, dref, root_objectid, + owner, offset)) { + err = 0; + break; + } + if (hash_extent_data_ref_item(leaf, dref) < + hash_extent_data_ref(root_objectid, owner, offset)) + break; + } else { + u64 ref_offset; + ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); + if (parent > 0) { + if (parent == ref_offset) { + err = 0; + break; + } + if (ref_offset < parent) + break; + } else { + if (root_objectid == ref_offset) { + err = 0; + break; + } + if (ref_offset < root_objectid) + break; + } + } + ptr += btrfs_extent_inline_ref_size(type); + } + if (err == -ENOENT && insert) { + if (item_size + extra_size >= + BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { + err = -EAGAIN; + goto out; + } + /* + * To add new inline back ref, we have to make sure + * there is no corresponding back ref item. + * For simplicity, we just do not add new inline back + * ref if there is any kind of item for this block + */ + if (find_next_key(path, 0, &key) == 0 && + key.objectid == bytenr && + key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { + err = -EAGAIN; + goto out; + } + } + *ref_ret = (struct btrfs_extent_inline_ref *)ptr; +out: + if (insert) { + path->keep_locks = 0; + btrfs_unlock_up_safe(path, 1); + } + return err; +} + +/* + * helper to add new inline back ref + */ +static noinline_for_stack +int setup_inline_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref *iref, + u64 parent, u64 root_objectid, + u64 owner, u64 offset, int refs_to_add, + struct btrfs_delayed_extent_op *extent_op) +{ + struct extent_buffer *leaf; + struct btrfs_extent_item *ei; + unsigned long ptr; + unsigned long end; + unsigned long item_offset; + u64 refs; + int size; + int type; + int ret; + + leaf = path->nodes[0]; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + item_offset = (unsigned long)iref - (unsigned long)ei; + + type = extent_ref_type(parent, owner); + size = btrfs_extent_inline_ref_size(type); + + ret = btrfs_extend_item(trans, root, path, size); + BUG_ON(ret); + + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + refs = btrfs_extent_refs(leaf, ei); + refs += refs_to_add; + btrfs_set_extent_refs(leaf, ei, refs); + if (extent_op) + __run_delayed_extent_op(extent_op, leaf, ei); + + ptr = (unsigned long)ei + item_offset; + end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); + if (ptr < end - size) + memmove_extent_buffer(leaf, ptr + size, ptr, + end - size - ptr); + + iref = (struct btrfs_extent_inline_ref *)ptr; + btrfs_set_extent_inline_ref_type(leaf, iref, type); + if (type == BTRFS_EXTENT_DATA_REF_KEY) { + struct btrfs_extent_data_ref *dref; + dref = (struct btrfs_extent_data_ref *)(&iref->offset); + btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); + btrfs_set_extent_data_ref_objectid(leaf, dref, owner); + btrfs_set_extent_data_ref_offset(leaf, dref, offset); + btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); + } else if (type == BTRFS_SHARED_DATA_REF_KEY) { + struct btrfs_shared_data_ref *sref; + sref = (struct btrfs_shared_data_ref *)(iref + 1); + btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); + btrfs_set_extent_inline_ref_offset(leaf, iref, parent); + } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { + btrfs_set_extent_inline_ref_offset(leaf, iref, parent); + } else { + btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); + } + btrfs_mark_buffer_dirty(leaf); + return 0; +} + +static int lookup_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref **ref_ret, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 owner, u64 offset) +{ + int ret; + + ret = lookup_inline_extent_backref(trans, root, path, ref_ret, + bytenr, num_bytes, parent, + root_objectid, owner, offset, 0); + if (ret != -ENOENT) + return ret; + + btrfs_release_path(root, path); + *ref_ret = NULL; + + if (owner < BTRFS_FIRST_FREE_OBJECTID) { + ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, + root_objectid); + } else { + ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, + root_objectid, owner, offset); + } + return ret; +} + +/* + * helper to update/remove inline back ref + */ +static noinline_for_stack +int update_inline_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref *iref, + int refs_to_mod, + struct btrfs_delayed_extent_op *extent_op) +{ + struct extent_buffer *leaf; + struct btrfs_extent_item *ei; + struct btrfs_extent_data_ref *dref = NULL; + struct btrfs_shared_data_ref *sref = NULL; + unsigned long ptr; + unsigned long end; + u32 item_size; + int size; + int type; + int ret; + u64 refs; + + leaf = path->nodes[0]; + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + refs = btrfs_extent_refs(leaf, ei); + WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); + refs += refs_to_mod; + btrfs_set_extent_refs(leaf, ei, refs); + if (extent_op) + __run_delayed_extent_op(extent_op, leaf, ei); + + type = btrfs_extent_inline_ref_type(leaf, iref); + + if (type == BTRFS_EXTENT_DATA_REF_KEY) { + dref = (struct btrfs_extent_data_ref *)(&iref->offset); + refs = btrfs_extent_data_ref_count(leaf, dref); + } else if (type == BTRFS_SHARED_DATA_REF_KEY) { + sref = (struct btrfs_shared_data_ref *)(iref + 1); + refs = btrfs_shared_data_ref_count(leaf, sref); + } else { + refs = 1; + BUG_ON(refs_to_mod != -1); + } + + BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); + refs += refs_to_mod; + + if (refs > 0) { + if (type == BTRFS_EXTENT_DATA_REF_KEY) + btrfs_set_extent_data_ref_count(leaf, dref, refs); + else + btrfs_set_shared_data_ref_count(leaf, sref, refs); + } else { + size = btrfs_extent_inline_ref_size(type); + item_size = btrfs_item_size_nr(leaf, path->slots[0]); + ptr = (unsigned long)iref; + end = (unsigned long)ei + item_size; + if (ptr + size < end) + memmove_extent_buffer(leaf, ptr, ptr + size, + end - ptr - size); + item_size -= size; + ret = btrfs_truncate_item(trans, root, path, item_size, 1); + BUG_ON(ret); + } + btrfs_mark_buffer_dirty(leaf); + return 0; +} + +static noinline_for_stack +int insert_inline_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 owner, + u64 offset, int refs_to_add, + struct btrfs_delayed_extent_op *extent_op) +{ + struct btrfs_extent_inline_ref *iref; + int ret; + + ret = lookup_inline_extent_backref(trans, root, path, &iref, + bytenr, num_bytes, parent, + root_objectid, owner, offset, 1); + if (ret == 0) { + BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); + ret = update_inline_extent_backref(trans, root, path, iref, + refs_to_add, extent_op); + } else if (ret == -ENOENT) { + ret = setup_inline_extent_backref(trans, root, path, iref, + parent, root_objectid, + owner, offset, refs_to_add, + extent_op); + } + return ret; +} + +static int insert_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 bytenr, u64 parent, u64 root_objectid, + u64 owner, u64 offset, int refs_to_add) +{ + int ret; + if (owner < BTRFS_FIRST_FREE_OBJECTID) { + BUG_ON(refs_to_add != 1); + ret = insert_tree_block_ref(trans, root, path, bytenr, + parent, root_objectid); + } else { + ret = insert_extent_data_ref(trans, root, path, bytenr, + parent, root_objectid, + owner, offset, refs_to_add); + } + return ret; +} + +static int remove_extent_backref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct btrfs_extent_inline_ref *iref, + int refs_to_drop, int is_data) +{ + int ret; + + BUG_ON(!is_data && refs_to_drop != 1); + if (iref) { + ret = update_inline_extent_backref(trans, root, path, iref, + -refs_to_drop, NULL); + } else if (is_data) { + ret = remove_extent_data_ref(trans, root, path, refs_to_drop); + } else { + ret = btrfs_del_item(trans, root, path); + } + 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); +} +#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; + + /* 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); + if (!ret) { + struct btrfs_bio_stripe *stripe = multi->stripes; + int i; + + if (map_length > num_bytes) + map_length = num_bytes; + + for (i = 0; i < multi->num_stripes; i++, stripe++) { + btrfs_issue_discard(stripe->dev->bdev, + stripe->physical, + map_length); + } + kfree(multi); + } + + return ret; +#else + return 0; +#endif +} + +int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 owner, u64 offset) +{ + int ret; + BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && + root_objectid == BTRFS_TREE_LOG_OBJECTID); + + if (owner < BTRFS_FIRST_FREE_OBJECTID) { + ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, + parent, root_objectid, (int)owner, + BTRFS_ADD_DELAYED_REF, NULL); + } else { + ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, + parent, root_objectid, owner, offset, + BTRFS_ADD_DELAYED_REF, NULL); + } + return ret; +} + +static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, + u64 parent, u64 root_objectid, + u64 owner, u64 offset, int refs_to_add, + struct btrfs_delayed_extent_op *extent_op) +{ + struct btrfs_path *path; + struct extent_buffer *leaf; + struct btrfs_extent_item *item; + u64 refs; + int ret; + int err = 0; + + path = btrfs_alloc_path(); + if (!path) + return -ENOMEM; + + path->reada = 1; + path->leave_spinning = 1; + /* this will setup the path even if it fails to insert the back ref */ + ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, + path, bytenr, num_bytes, parent, + root_objectid, owner, offset, + refs_to_add, extent_op); + if (ret == 0) + goto out; + + if (ret != -EAGAIN) { + err = ret; + goto out; + } + + leaf = path->nodes[0]; + item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + refs = btrfs_extent_refs(leaf, item); + btrfs_set_extent_refs(leaf, item, refs + refs_to_add); + if (extent_op) + __run_delayed_extent_op(extent_op, leaf, item); + + btrfs_mark_buffer_dirty(leaf); + btrfs_release_path(root->fs_info->extent_root, path); + + path->reada = 1; + path->leave_spinning = 1; + + /* now insert the actual backref */ + ret = insert_extent_backref(trans, root->fs_info->extent_root, + path, bytenr, parent, root_objectid, + owner, offset, refs_to_add); + BUG_ON(ret); +out: + btrfs_free_path(path); + return err; +} + +static int run_delayed_data_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node, + struct btrfs_delayed_extent_op *extent_op, + int insert_reserved) +{ + int ret = 0; + struct btrfs_delayed_data_ref *ref; + struct btrfs_key ins; + u64 parent = 0; + u64 ref_root = 0; + u64 flags = 0; + + ins.objectid = node->bytenr; + ins.offset = node->num_bytes; + ins.type = BTRFS_EXTENT_ITEM_KEY; + + ref = btrfs_delayed_node_to_data_ref(node); + if (node->type == BTRFS_SHARED_DATA_REF_KEY) + parent = ref->parent; + else + ref_root = ref->root; + + if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { + if (extent_op) { + BUG_ON(extent_op->update_key); + flags |= extent_op->flags_to_set; + } + ret = alloc_reserved_file_extent(trans, root, + parent, ref_root, flags, + ref->objectid, ref->offset, + &ins, node->ref_mod); + update_reserved_extents(root, ins.objectid, ins.offset, 0); + } else if (node->action == BTRFS_ADD_DELAYED_REF) { + ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, + node->num_bytes, parent, + ref_root, ref->objectid, + ref->offset, node->ref_mod, + extent_op); + } else if (node->action == BTRFS_DROP_DELAYED_REF) { + ret = __btrfs_free_extent(trans, root, node->bytenr, + node->num_bytes, parent, + ref_root, ref->objectid, + ref->offset, node->ref_mod, + extent_op); + } else { + BUG(); + } + return ret; +} + +static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, + struct extent_buffer *leaf, + struct btrfs_extent_item *ei) +{ + u64 flags = btrfs_extent_flags(leaf, ei); + if (extent_op->update_flags) { + flags |= extent_op->flags_to_set; + btrfs_set_extent_flags(leaf, ei, flags); + } + + if (extent_op->update_key) { + struct btrfs_tree_block_info *bi; + BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); + bi = (struct btrfs_tree_block_info *)(ei + 1); + btrfs_set_tree_block_key(leaf, bi, &extent_op->key); + } +} + +static int run_delayed_extent_op(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node, + struct btrfs_delayed_extent_op *extent_op) +{ + struct btrfs_key key; + struct btrfs_path *path; + struct btrfs_extent_item *ei; + struct extent_buffer *leaf; + u32 item_size; + int ret; + int err = 0; + path = btrfs_alloc_path(); if (!path) return -ENOMEM; - ret = lookup_extent_backref(trans, extent_root, path, - bytenr, orig_parent, orig_root, - orig_generation, owner_objectid, 1); - if (ret) + + key.objectid = node->bytenr; + key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = node->num_bytes; + + path->reada = 1; + path->leave_spinning = 1; + ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, + path, 0, 1); + if (ret < 0) { + err = ret; goto out; - ret = remove_extent_backref(trans, extent_root, path); - if (ret) + } + if (ret > 0) { + err = -EIO; goto out; - ret = insert_extent_backref(trans, extent_root, path, bytenr, - parent, ref_root, ref_generation, - owner_objectid); - BUG_ON(ret); - finish_current_insert(trans, extent_root, 0); - del_pending_extents(trans, extent_root, 0); + } + + leaf = path->nodes[0]; + item_size = btrfs_item_size_nr(leaf, path->slots[0]); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + if (item_size < sizeof(*ei)) { + ret = convert_extent_item_v0(trans, root->fs_info->extent_root, + path, (u64)-1, 0); + if (ret < 0) { + err = ret; + goto out; + } + leaf = path->nodes[0]; + item_size = btrfs_item_size_nr(leaf, path->slots[0]); + } +#endif + BUG_ON(item_size < sizeof(*ei)); + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); + __run_delayed_extent_op(extent_op, leaf, ei); + + btrfs_mark_buffer_dirty(leaf); out: btrfs_free_path(path); + return err; +} + +static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node, + struct btrfs_delayed_extent_op *extent_op, + int insert_reserved) +{ + int ret = 0; + struct btrfs_delayed_tree_ref *ref; + struct btrfs_key ins; + u64 parent = 0; + u64 ref_root = 0; + + ins.objectid = node->bytenr; + ins.offset = node->num_bytes; + ins.type = BTRFS_EXTENT_ITEM_KEY; + + ref = btrfs_delayed_node_to_tree_ref(node); + if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) + parent = ref->parent; + else + ref_root = ref->root; + + BUG_ON(node->ref_mod != 1); + if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { + BUG_ON(!extent_op || !extent_op->update_flags || + !extent_op->update_key); + ret = alloc_reserved_tree_block(trans, root, + parent, ref_root, + extent_op->flags_to_set, + &extent_op->key, + ref->level, &ins); + update_reserved_extents(root, ins.objectid, ins.offset, 0); + } else if (node->action == BTRFS_ADD_DELAYED_REF) { + ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, + node->num_bytes, parent, ref_root, + ref->level, 0, 1, extent_op); + } else if (node->action == BTRFS_DROP_DELAYED_REF) { + ret = __btrfs_free_extent(trans, root, node->bytenr, + node->num_bytes, parent, ref_root, + ref->level, 0, 1, extent_op); + } else { + BUG(); + } return ret; } -int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr, - u64 orig_parent, u64 parent, - u64 ref_root, u64 ref_generation, - u64 owner_objectid) + +/* helper function to actually process a single delayed ref entry */ +static int run_one_delayed_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node, + struct btrfs_delayed_extent_op *extent_op, + int insert_reserved) { int ret; - if (ref_root == BTRFS_TREE_LOG_OBJECTID && - owner_objectid < BTRFS_FIRST_FREE_OBJECTID) + if (btrfs_delayed_ref_is_head(node)) { + struct btrfs_delayed_ref_head *head; + /* + * we've hit the end of the chain and we were supposed + * to insert this extent into the tree. But, it got + * deleted before we ever needed to insert it, so all + * we have to do is clean up the accounting + */ + BUG_ON(extent_op); + head = btrfs_delayed_node_to_head(node); + if (insert_reserved) { + if (head->is_data) { + ret = btrfs_del_csums(trans, root, + node->bytenr, + node->num_bytes); + BUG_ON(ret); + } + btrfs_update_pinned_extents(root, node->bytenr, + node->num_bytes, 1); + update_reserved_extents(root, node->bytenr, + node->num_bytes, 0); + } + mutex_unlock(&head->mutex); return 0; - ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, - parent, ref_root, ref_root, - ref_generation, ref_generation, - owner_objectid); + } + + if (node->type == BTRFS_TREE_BLOCK_REF_KEY || + node->type == BTRFS_SHARED_BLOCK_REF_KEY) + ret = run_delayed_tree_ref(trans, root, node, extent_op, + insert_reserved); + else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || + node->type == BTRFS_SHARED_DATA_REF_KEY) + ret = run_delayed_data_ref(trans, root, node, extent_op, + insert_reserved); + else + BUG(); return ret; } -static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr, - u64 orig_parent, u64 parent, - u64 orig_root, u64 ref_root, - u64 orig_generation, u64 ref_generation, - u64 owner_objectid) +static noinline struct btrfs_delayed_ref_node * +select_delayed_ref(struct btrfs_delayed_ref_head *head) { - struct btrfs_path *path; + struct rb_node *node; + struct btrfs_delayed_ref_node *ref; + int action = BTRFS_ADD_DELAYED_REF; +again: + /* + * select delayed ref of type BTRFS_ADD_DELAYED_REF first. + * this prevents ref count from going down to zero when + * there still are pending delayed ref. + */ + node = rb_prev(&head->node.rb_node); + while (1) { + if (!node) + break; + ref = rb_entry(node, struct btrfs_delayed_ref_node, + rb_node); + if (ref->bytenr != head->node.bytenr) + break; + if (ref->action == action) + return ref; + node = rb_prev(node); + } + if (action == BTRFS_ADD_DELAYED_REF) { + action = BTRFS_DROP_DELAYED_REF; + goto again; + } + return NULL; +} + +static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct list_head *cluster) +{ + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct btrfs_delayed_ref_head *locked_ref = NULL; + struct btrfs_delayed_extent_op *extent_op; int ret; - struct btrfs_key key; - struct extent_buffer *l; - struct btrfs_extent_item *item; - u32 refs; + int count = 0; + int must_insert_reserved = 0; - path = btrfs_alloc_path(); - if (!path) - return -ENOMEM; + delayed_refs = &trans->transaction->delayed_refs; + while (1) { + if (!locked_ref) { + /* pick a new head ref from the cluster list */ + if (list_empty(cluster)) + break; - path->reada = 1; - key.objectid = bytenr; - key.type = BTRFS_EXTENT_ITEM_KEY; - key.offset = (u64)-1; + locked_ref = list_entry(cluster->next, + struct btrfs_delayed_ref_head, cluster); - ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, - 0, 1); - if (ret < 0) - return ret; - BUG_ON(ret == 0 || path->slots[0] == 0); + /* grab the lock that says we are going to process + * all the refs for this head */ + ret = btrfs_delayed_ref_lock(trans, locked_ref); - path->slots[0]--; - l = path->nodes[0]; + /* + * we may have dropped the spin lock to get the head + * mutex lock, and that might have given someone else + * time to free the head. If that's true, it has been + * removed from our list and we can move on. + */ + if (ret == -EAGAIN) { + locked_ref = NULL; + count++; + continue; + } + } - btrfs_item_key_to_cpu(l, &key, path->slots[0]); - if (key.objectid != bytenr) { - btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]); - printk("wanted %Lu found %Lu\n", bytenr, key.objectid); - BUG(); - } - BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); + /* + * record the must insert reserved flag before we + * drop the spin lock. + */ + must_insert_reserved = locked_ref->must_insert_reserved; + locked_ref->must_insert_reserved = 0; - item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); - refs = btrfs_extent_refs(l, item); - btrfs_set_extent_refs(l, item, refs + 1); - btrfs_mark_buffer_dirty(path->nodes[0]); + extent_op = locked_ref->extent_op; + locked_ref->extent_op = NULL; - btrfs_release_path(root->fs_info->extent_root, path); + /* + * locked_ref is the head node, so we have to go one + * node back for any delayed ref updates + */ + ref = select_delayed_ref(locked_ref); + if (!ref) { + /* All delayed refs have been processed, Go ahead + * and send the head node to run_one_delayed_ref, + * so that any accounting fixes can happen + */ + ref = &locked_ref->node; - path->reada = 1; - ret = insert_extent_backref(trans, root->fs_info->extent_root, - path, bytenr, parent, - ref_root, ref_generation, - owner_objectid); - BUG_ON(ret); - finish_current_insert(trans, root->fs_info->extent_root, 0); - del_pending_extents(trans, root->fs_info->extent_root, 0); + if (extent_op && must_insert_reserved) { + kfree(extent_op); + extent_op = NULL; + } - btrfs_free_path(path); - return 0; + if (extent_op) { + spin_unlock(&delayed_refs->lock); + + ret = run_delayed_extent_op(trans, root, + ref, extent_op); + BUG_ON(ret); + kfree(extent_op); + + cond_resched(); + spin_lock(&delayed_refs->lock); + continue; + } + + list_del_init(&locked_ref->cluster); + locked_ref = NULL; + } + + ref->in_tree = 0; + rb_erase(&ref->rb_node, &delayed_refs->root); + delayed_refs->num_entries--; + + spin_unlock(&delayed_refs->lock); + + ret = run_one_delayed_ref(trans, root, ref, extent_op, + must_insert_reserved); + BUG_ON(ret); + + btrfs_put_delayed_ref(ref); + kfree(extent_op); + count++; + + cond_resched(); + spin_lock(&delayed_refs->lock); + } + return count; } -int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, u64 parent, - u64 ref_root, u64 ref_generation, - u64 owner_objectid) +/* + * this starts processing the delayed reference count updates and + * extent insertions we have queued up so far. count can be + * 0, which means to process everything in the tree at the start + * of the run (but not newly added entries), or it can be some target + * number you'd like to process. + */ +int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, unsigned long count) { + struct rb_node *node; + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct list_head cluster; int ret; - if (ref_root == BTRFS_TREE_LOG_OBJECTID && - owner_objectid < BTRFS_FIRST_FREE_OBJECTID) - return 0; - ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, - 0, ref_root, 0, ref_generation, - owner_objectid); - return ret; -} + int run_all = count == (unsigned long)-1; + int run_most = 0; -int btrfs_extent_post_op(struct btrfs_trans_handle *trans, - struct btrfs_root *root) -{ - finish_current_insert(trans, root->fs_info->extent_root, 1); - del_pending_extents(trans, root->fs_info->extent_root, 1); + if (root == root->fs_info->extent_root) + root = root->fs_info->tree_root; + + delayed_refs = &trans->transaction->delayed_refs; + INIT_LIST_HEAD(&cluster); +again: + spin_lock(&delayed_refs->lock); + if (count == 0) { + count = delayed_refs->num_entries * 2; + run_most = 1; + } + while (1) { + if (!(run_all || run_most) && + delayed_refs->num_heads_ready < 64) + break; + + /* + * go find something we can process in the rbtree. We start at + * the beginning of the tree, and then build a cluster + * of refs to process starting at the first one we are able to + * lock + */ + ret = btrfs_find_ref_cluster(trans, &cluster, + delayed_refs->run_delayed_start); + if (ret) + break; + + ret = run_clustered_refs(trans, root, &cluster); + BUG_ON(ret < 0); + + count -= min_t(unsigned long, ret, count); + + if (count == 0) + break; + } + + if (run_all) { + node = rb_first(&delayed_refs->root); + if (!node) + goto out; + count = (unsigned long)-1; + + while (node) { + ref = rb_entry(node, struct btrfs_delayed_ref_node, + rb_node); + if (btrfs_delayed_ref_is_head(ref)) { + struct btrfs_delayed_ref_head *head; + + head = btrfs_delayed_node_to_head(ref); + atomic_inc(&ref->refs); + + spin_unlock(&delayed_refs->lock); + mutex_lock(&head->mutex); + mutex_unlock(&head->mutex); + + btrfs_put_delayed_ref(ref); + cond_resched(); + goto again; + } + node = rb_next(node); + } + spin_unlock(&delayed_refs->lock); + schedule_timeout(1); + goto again; + } +out: + spin_unlock(&delayed_refs->lock); return 0; } -int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr, - u64 num_bytes, u32 *refs) +int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 flags, + int is_data) { - struct btrfs_path *path; + struct btrfs_delayed_extent_op *extent_op; int ret; - struct btrfs_key key; - struct extent_buffer *l; - struct btrfs_extent_item *item; - WARN_ON(num_bytes < root->sectorsize); - path = btrfs_alloc_path(); - path->reada = 1; - key.objectid = bytenr; - key.offset = num_bytes; - btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); - ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, - 0, 0); - if (ret < 0) + extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); + if (!extent_op) + return -ENOMEM; + + extent_op->flags_to_set = flags; + extent_op->update_flags = 1; + extent_op->update_key = 0; + extent_op->is_data = is_data ? 1 : 0; + + ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); + if (ret) + kfree(extent_op); + return ret; +} + +static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 objectid, u64 offset, u64 bytenr) +{ + struct btrfs_delayed_ref_head *head; + struct btrfs_delayed_ref_node *ref; + struct btrfs_delayed_data_ref *data_ref; + struct btrfs_delayed_ref_root *delayed_refs; + struct rb_node *node; + int ret = 0; + + ret = -ENOENT; + delayed_refs = &trans->transaction->delayed_refs; + spin_lock(&delayed_refs->lock); + head = btrfs_find_delayed_ref_head(trans, bytenr); + if (!head) goto out; - if (ret != 0) { - btrfs_print_leaf(root, path->nodes[0]); - printk("failed to find block number %Lu\n", bytenr); - BUG(); + + if (!mutex_trylock(&head->mutex)) { + atomic_inc(&head->node.refs); + spin_unlock(&delayed_refs->lock); + + btrfs_release_path(root->fs_info->extent_root, path); + + mutex_lock(&head->mutex); + mutex_unlock(&head->mutex); + btrfs_put_delayed_ref(&head->node); + return -EAGAIN; + } + + node = rb_prev(&head->node.rb_node); + if (!node) + goto out_unlock; + + ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); + + if (ref->bytenr != bytenr) + goto out_unlock; + + ret = 1; + if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) + goto out_unlock; + + data_ref = btrfs_delayed_node_to_data_ref(ref); + + node = rb_prev(node); + if (node) { + ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); + if (ref->bytenr == bytenr) + goto out_unlock; } - l = path->nodes[0]; - item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); - *refs = btrfs_extent_refs(l, item); + + if (data_ref->root != root->root_key.objectid || + data_ref->objectid != objectid || data_ref->offset != offset) + goto out_unlock; + + ret = 0; +out_unlock: + mutex_unlock(&head->mutex); out: - btrfs_free_path(path); - return 0; + spin_unlock(&delayed_refs->lock); + return ret; } -int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr) +static noinline int check_committed_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + u64 objectid, u64 offset, u64 bytenr) { struct btrfs_root *extent_root = root->fs_info->extent_root; - struct btrfs_path *path; struct extent_buffer *leaf; - struct btrfs_extent_ref *ref_item; + struct btrfs_extent_data_ref *ref; + struct btrfs_extent_inline_ref *iref; + struct btrfs_extent_item *ei; struct btrfs_key key; - struct btrfs_key found_key; - u64 ref_root; - u64 last_snapshot; - u32 nritems; + u32 item_size; int ret; key.objectid = bytenr; key.offset = (u64)-1; key.type = BTRFS_EXTENT_ITEM_KEY; - path = btrfs_alloc_path(); ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); if (ret < 0) goto out; @@ -882,54 +2184,83 @@ int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, path->slots[0]--; leaf = path->nodes[0]; - btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); + btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); - if (found_key.objectid != bytenr || - found_key.type != BTRFS_EXTENT_ITEM_KEY) + if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) goto out; - last_snapshot = btrfs_root_last_snapshot(&root->root_item); - while (1) { - leaf = path->nodes[0]; - nritems = btrfs_header_nritems(leaf); - if (path->slots[0] >= nritems) { - ret = btrfs_next_leaf(extent_root, path); - if (ret < 0) - goto out; - if (ret == 0) - continue; - break; - } - btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); - if (found_key.objectid != bytenr) - break; + ret = 1; + item_size = btrfs_item_size_nr(leaf, path->slots[0]); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + if (item_size < sizeof(*ei)) { + WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); + goto out; + } +#endif + ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); - if (found_key.type != BTRFS_EXTENT_REF_KEY) { - path->slots[0]++; - continue; - } + if (item_size != sizeof(*ei) + + btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) + goto out; - ref_item = btrfs_item_ptr(leaf, path->slots[0], - struct btrfs_extent_ref); - ref_root = btrfs_ref_root(leaf, ref_item); - if (ref_root != root->root_key.objectid && - ref_root != BTRFS_TREE_LOG_OBJECTID) { - ret = 1; - goto out; - } - if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) { - ret = 1; + if (btrfs_extent_generation(leaf, ei) <= + btrfs_root_last_snapshot(&root->root_item)) + goto out; + + iref = (struct btrfs_extent_inline_ref *)(ei + 1); + if (btrfs_extent_inline_ref_type(leaf, iref) != + BTRFS_EXTENT_DATA_REF_KEY) + goto out; + + ref = (struct btrfs_extent_data_ref *)(&iref->offset); + if (btrfs_extent_refs(leaf, ei) != + btrfs_extent_data_ref_count(leaf, ref) || + btrfs_extent_data_ref_root(leaf, ref) != + root->root_key.objectid || + btrfs_extent_data_ref_objectid(leaf, ref) != objectid || + btrfs_extent_data_ref_offset(leaf, ref) != offset) + goto out; + + ret = 0; +out: + return ret; +} + +int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 objectid, u64 offset, u64 bytenr) +{ + struct btrfs_path *path; + int ret; + int ret2; + + path = btrfs_alloc_path(); + if (!path) + return -ENOENT; + + do { + ret = check_committed_ref(trans, root, path, objectid, + offset, bytenr); + if (ret && ret != -ENOENT) goto out; - } - path->slots[0]++; + ret2 = check_delayed_ref(trans, root, path, objectid, + offset, bytenr); + } while (ret2 == -EAGAIN); + + if (ret2 && ret2 != -ENOENT) { + ret = ret2; + goto out; } - ret = 0; + + if (ret != -ENOENT || ret2 != -ENOENT) + ret = 0; out: btrfs_free_path(path); return ret; } +#if 0 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *buf, u32 nr_extents) { @@ -1011,48 +2342,80 @@ out: return ret; } -int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, - struct extent_buffer *orig_buf, struct extent_buffer *buf, - u32 *nr_extents) +/* when a block goes through cow, we update the reference counts of + * everything that block points to. The internal pointers of the block + * can be in just about any order, and it is likely to have clusters of + * things that are close together and clusters of things that are not. + * + * To help reduce the seeks that come with updating all of these reference + * counts, sort them by byte number before actual updates are done. + * + * struct refsort is used to match byte number to slot in the btree block. + * we sort based on the byte number and then use the slot to actually + * find the item. + * + * struct refsort is smaller than strcut btrfs_item and smaller than + * struct btrfs_key_ptr. Since we're currently limited to the page size + * for a btree block, there's no way for a kmalloc of refsorts for a + * single node to be bigger than a page. + */ +struct refsort { + u64 bytenr; + u32 slot; +}; + +/* + * for passing into sort() + */ +static int refsort_cmp(const void *a_void, const void *b_void) +{ + const struct refsort *a = a_void; + const struct refsort *b = b_void; + + if (a->bytenr < b->bytenr) + return -1; + if (a->bytenr > b->bytenr) + return 1; + return 0; +} +#endif + +static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct extent_buffer *buf, + int full_backref, int inc) { u64 bytenr; + u64 num_bytes; + u64 parent; u64 ref_root; - u64 orig_root; - u64 ref_generation; - u64 orig_generation; u32 nritems; - u32 nr_file_extents = 0; struct btrfs_key key; struct btrfs_file_extent_item *fi; int i; int level; int ret = 0; - int faili = 0; int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, - u64, u64, u64, u64, u64, u64, u64, u64); + u64, u64, u64, u64, u64, u64); ref_root = btrfs_header_owner(buf); - ref_generation = btrfs_header_generation(buf); - orig_root = btrfs_header_owner(orig_buf); - orig_generation = btrfs_header_generation(orig_buf); - nritems = btrfs_header_nritems(buf); level = btrfs_header_level(buf); - if (root->ref_cows) { - process_func = __btrfs_inc_extent_ref; - } else { - if (level == 0 && - root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) - goto out; - if (level != 0 && - root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) - goto out; - process_func = __btrfs_update_extent_ref; - } + if (!root->ref_cows && level == 0) + return 0; + + if (inc) + process_func = btrfs_inc_extent_ref; + else + process_func = btrfs_free_extent; + + if (full_backref) + parent = buf->start; + else + parent = 0; for (i = 0; i < nritems; i++) { - cond_resched(); if (level == 0) { btrfs_item_key_to_cpu(buf, &key, i); if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) @@ -1066,117 +2429,38 @@ int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, if (bytenr == 0) continue; - nr_file_extents++; - - ret = process_func(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - key.objectid); - - if (ret) { - faili = i; - WARN_ON(1); + num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); + key.offset -= btrfs_file_extent_offset(buf, fi); + ret = process_func(trans, root, bytenr, num_bytes, + parent, ref_root, key.objectid, + key.offset); + if (ret) goto fail; - } } else { bytenr = btrfs_node_blockptr(buf, i); - ret = process_func(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - level - 1); - if (ret) { - faili = i; - WARN_ON(1); + num_bytes = btrfs_level_size(root, level - 1); + ret = process_func(trans, root, bytenr, num_bytes, + parent, ref_root, level - 1, 0); + if (ret) goto fail; - } } } -out: - if (nr_extents) { - if (level == 0) - *nr_extents = nr_file_extents; - else - *nr_extents = nritems; - } return 0; fail: - WARN_ON(1); + BUG(); return ret; } -int btrfs_update_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, struct extent_buffer *orig_buf, - struct extent_buffer *buf, int start_slot, int nr) - +int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, + struct extent_buffer *buf, int full_backref) { - u64 bytenr; - u64 ref_root; - u64 orig_root; - u64 ref_generation; - u64 orig_generation; - struct btrfs_key key; - struct btrfs_file_extent_item *fi; - int i; - int ret; - int slot; - int level; - - BUG_ON(start_slot < 0); - BUG_ON(start_slot + nr > btrfs_header_nritems(buf)); - - ref_root = btrfs_header_owner(buf); - ref_generation = btrfs_header_generation(buf); - orig_root = btrfs_header_owner(orig_buf); - orig_generation = btrfs_header_generation(orig_buf); - level = btrfs_header_level(buf); - - if (!root->ref_cows) { - if (level == 0 && - root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) - return 0; - if (level != 0 && - root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) - return 0; - } + return __btrfs_mod_ref(trans, root, buf, full_backref, 1); +} - for (i = 0, slot = start_slot; i < nr; i++, slot++) { - cond_resched(); - if (level == 0) { - btrfs_item_key_to_cpu(buf, &key, slot); - if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) - continue; - fi = btrfs_item_ptr(buf, slot, - struct btrfs_file_extent_item); - if (btrfs_file_extent_type(buf, fi) == - BTRFS_FILE_EXTENT_INLINE) - continue; - bytenr = btrfs_file_extent_disk_bytenr(buf, fi); - if (bytenr == 0) - continue; - ret = __btrfs_update_extent_ref(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - key.objectid); - if (ret) - goto fail; - } else { - bytenr = btrfs_node_blockptr(buf, slot); - ret = __btrfs_update_extent_ref(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - level - 1); - if (ret) - goto fail; - } - } - return 0; -fail: - WARN_ON(1); - return -1; +int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, + struct extent_buffer *buf, int full_backref) +{ + return __btrfs_mod_ref(trans, root, buf, full_backref, 0); } static int write_one_cache_group(struct btrfs_trans_handle *trans, @@ -1185,7 +2469,6 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans, struct btrfs_block_group_cache *cache) { int ret; - int pending_ret; struct btrfs_root *extent_root = root->fs_info->extent_root; unsigned long bi; struct extent_buffer *leaf; @@ -1201,23 +2484,35 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans, btrfs_mark_buffer_dirty(leaf); btrfs_release_path(extent_root, path); fail: - finish_current_insert(trans, extent_root, 0); - pending_ret = del_pending_extents(trans, extent_root, 0); if (ret) return ret; - if (pending_ret) - return pending_ret; return 0; } +static struct btrfs_block_group_cache * +next_block_group(struct btrfs_root *root, + struct btrfs_block_group_cache *cache) +{ + struct rb_node *node; + spin_lock(&root->fs_info->block_group_cache_lock); + node = rb_next(&cache->cache_node); + btrfs_put_block_group(cache); + if (node) { + cache = rb_entry(node, struct btrfs_block_group_cache, + cache_node); + atomic_inc(&cache->count); + } else + cache = NULL; + spin_unlock(&root->fs_info->block_group_cache_lock); + return cache; +} + int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, struct btrfs_root *root) { - struct btrfs_block_group_cache *cache, *entry; - struct rb_node *n; + struct btrfs_block_group_cache *cache; int err = 0; - int werr = 0; struct btrfs_path *path; u64 last = 0; @@ -1225,119 +2520,388 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; - while(1) { - cache = NULL; - spin_lock(&root->fs_info->block_group_cache_lock); - for (n = rb_first(&root->fs_info->block_group_cache_tree); - n; n = rb_next(n)) { - entry = rb_entry(n, struct btrfs_block_group_cache, - cache_node); - if (entry->dirty) { - cache = entry; + while (1) { + if (last == 0) { + err = btrfs_run_delayed_refs(trans, root, + (unsigned long)-1); + BUG_ON(err); + } + + cache = btrfs_lookup_first_block_group(root->fs_info, last); + while (cache) { + if (cache->dirty) break; - } + cache = next_block_group(root, cache); + } + if (!cache) { + if (last == 0) + break; + last = 0; + continue; + } + + cache->dirty = 0; + last = cache->key.objectid + cache->key.offset; + + err = write_one_cache_group(trans, root, path, cache); + BUG_ON(err); + btrfs_put_block_group(cache); + } + + btrfs_free_path(path); + return 0; +} + +int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) +{ + struct btrfs_block_group_cache *block_group; + int readonly = 0; + + block_group = btrfs_lookup_block_group(root->fs_info, bytenr); + if (!block_group || block_group->ro) + readonly = 1; + if (block_group) + btrfs_put_block_group(block_group); + return readonly; +} + +static int update_space_info(struct btrfs_fs_info *info, u64 flags, + u64 total_bytes, u64 bytes_used, + struct btrfs_space_info **space_info) +{ + struct btrfs_space_info *found; + + found = __find_space_info(info, flags); + if (found) { + spin_lock(&found->lock); + found->total_bytes += total_bytes; + found->bytes_used += bytes_used; + found->full = 0; + spin_unlock(&found->lock); + *space_info = found; + return 0; + } + found = kzalloc(sizeof(*found), GFP_NOFS); + if (!found) + return -ENOMEM; + + INIT_LIST_HEAD(&found->block_groups); + init_rwsem(&found->groups_sem); + spin_lock_init(&found->lock); + found->flags = flags; + found->total_bytes = total_bytes; + found->bytes_used = bytes_used; + found->bytes_pinned = 0; + found->bytes_reserved = 0; + found->bytes_readonly = 0; + found->bytes_delalloc = 0; + found->full = 0; + found->force_alloc = 0; + *space_info = found; + list_add_rcu(&found->list, &info->space_info); + atomic_set(&found->caching_threads, 0); + return 0; +} + +static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) +{ + u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | + BTRFS_BLOCK_GROUP_RAID1 | + BTRFS_BLOCK_GROUP_RAID10 | + BTRFS_BLOCK_GROUP_DUP); + if (extra_flags) { + if (flags & BTRFS_BLOCK_GROUP_DATA) + fs_info->avail_data_alloc_bits |= extra_flags; + if (flags & BTRFS_BLOCK_GROUP_METADATA) + fs_info->avail_metadata_alloc_bits |= extra_flags; + if (flags & BTRFS_BLOCK_GROUP_SYSTEM) + fs_info->avail_system_alloc_bits |= extra_flags; + } +} + +static void set_block_group_readonly(struct btrfs_block_group_cache *cache) +{ + spin_lock(&cache->space_info->lock); + spin_lock(&cache->lock); + if (!cache->ro) { + cache->space_info->bytes_readonly += cache->key.offset - + btrfs_block_group_used(&cache->item); + cache->ro = 1; + } + spin_unlock(&cache->lock); + spin_unlock(&cache->space_info->lock); +} + +u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) +{ + u64 num_devices = root->fs_info->fs_devices->rw_devices; + + if (num_devices == 1) + flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); + if (num_devices < 4) + flags &= ~BTRFS_BLOCK_GROUP_RAID10; + + if ((flags & BTRFS_BLOCK_GROUP_DUP) && + (flags & (BTRFS_BLOCK_GROUP_RAID1 | + BTRFS_BLOCK_GROUP_RAID10))) { + flags &= ~BTRFS_BLOCK_GROUP_DUP; + } + + if ((flags & BTRFS_BLOCK_GROUP_RAID1) && + (flags & BTRFS_BLOCK_GROUP_RAID10)) { + flags &= ~BTRFS_BLOCK_GROUP_RAID1; + } + + if ((flags & BTRFS_BLOCK_GROUP_RAID0) && + ((flags & BTRFS_BLOCK_GROUP_RAID1) | + (flags & BTRFS_BLOCK_GROUP_RAID10) | + (flags & BTRFS_BLOCK_GROUP_DUP))) + flags &= ~BTRFS_BLOCK_GROUP_RAID0; + return flags; +} + +static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) +{ + struct btrfs_fs_info *info = root->fs_info; + u64 alloc_profile; + + if (data) { + alloc_profile = info->avail_data_alloc_bits & + info->data_alloc_profile; + data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; + } else if (root == root->fs_info->chunk_root) { + alloc_profile = info->avail_system_alloc_bits & + info->system_alloc_profile; + data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; + } else { + alloc_profile = info->avail_metadata_alloc_bits & + info->metadata_alloc_profile; + data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; + } + + return btrfs_reduce_alloc_profile(root, data); +} + +void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) +{ + u64 alloc_target; + + alloc_target = btrfs_get_alloc_profile(root, 1); + BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, + alloc_target); +} + +/* + * for now this just makes sure we have at least 5% of our metadata space free + * for use. + */ +int btrfs_check_metadata_free_space(struct btrfs_root *root) +{ + struct btrfs_fs_info *info = root->fs_info; + struct btrfs_space_info *meta_sinfo; + u64 alloc_target, thresh; + int committed = 0, ret; + + /* 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: + spin_lock(&meta_sinfo->lock); + if (!meta_sinfo->full) + thresh = meta_sinfo->total_bytes * 80; + else + thresh = meta_sinfo->total_bytes * 95; + + do_div(thresh, 100); + + 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); + + trans = btrfs_start_transaction(root, 1); + if (!trans) + return -ENOMEM; + + ret = do_chunk_alloc(trans, root->fs_info->extent_root, + 2 * 1024 * 1024, alloc_target, 0); + btrfs_end_transaction(trans, root); + goto again; + } + spin_unlock(&meta_sinfo->lock); + + if (!committed) { + committed = 1; + trans = btrfs_join_transaction(root, 1); + if (!trans) + return -ENOMEM; + ret = btrfs_commit_transaction(trans, root); + if (ret) + return ret; + goto again; } - spin_unlock(&root->fs_info->block_group_cache_lock); + return -ENOSPC; + } + spin_unlock(&meta_sinfo->lock); + + return 0; +} + +/* + * This will check the space that the inode allocates from to make sure we have + * enough space for bytes. + */ +int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + int ret = 0, committed = 0; - if (!cache) - break; + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); - cache->dirty = 0; - last += cache->key.offset; + data_sinfo = BTRFS_I(inode)->space_info; +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) { + struct btrfs_trans_handle *trans; - err = write_one_cache_group(trans, root, - path, cache); /* - * if we fail to write the cache group, we want - * to keep it marked dirty in hopes that a later - * write will work + * if we don't have enough free bytes in this space then we need + * to alloc a new chunk. */ - if (err) { - werr = err; - continue; + if (!data_sinfo->full) { + u64 alloc_target; + + data_sinfo->force_alloc = 1; + spin_unlock(&data_sinfo->lock); + + alloc_target = btrfs_get_alloc_profile(root, 1); + trans = btrfs_start_transaction(root, 1); + if (!trans) + return -ENOMEM; + + ret = do_chunk_alloc(trans, root->fs_info->extent_root, + bytes + 2 * 1024 * 1024, + alloc_target, 0); + btrfs_end_transaction(trans, root); + if (ret) + return ret; + goto again; + } + spin_unlock(&data_sinfo->lock); + + /* commit the current transaction and try again */ + if (!committed) { + committed = 1; + trans = btrfs_join_transaction(root, 1); + if (!trans) + return -ENOMEM; + ret = btrfs_commit_transaction(trans, root); + if (ret) + return ret; + goto again; } + + printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" + ", %llu bytes_used, %llu bytes_reserved, " + "%llu bytes_pinned, %llu bytes_readonly, %llu may use " + "%llu total\n", (unsigned long long)bytes, + (unsigned long long)data_sinfo->bytes_delalloc, + (unsigned long long)data_sinfo->bytes_used, + (unsigned long long)data_sinfo->bytes_reserved, + (unsigned long long)data_sinfo->bytes_pinned, + (unsigned long long)data_sinfo->bytes_readonly, + (unsigned long long)data_sinfo->bytes_may_use, + (unsigned long long)data_sinfo->total_bytes); + return -ENOSPC; } - btrfs_free_path(path); - return werr; + data_sinfo->bytes_may_use += bytes; + BTRFS_I(inode)->reserved_bytes += bytes; + spin_unlock(&data_sinfo->lock); + + return btrfs_check_metadata_free_space(root); } -static int update_space_info(struct btrfs_fs_info *info, u64 flags, - u64 total_bytes, u64 bytes_used, - struct btrfs_space_info **space_info) +/* + * if there was an error for whatever reason after calling + * btrfs_check_data_free_space, call this so we can cleanup the counters. + */ +void btrfs_free_reserved_data_space(struct btrfs_root *root, + struct inode *inode, u64 bytes) { - struct btrfs_space_info *found; + struct btrfs_space_info *data_sinfo; - found = __find_space_info(info, flags); - if (found) { - spin_lock(&found->lock); - found->total_bytes += total_bytes; - found->bytes_used += bytes_used; - found->full = 0; - spin_unlock(&found->lock); - *space_info = found; - return 0; - } - found = kmalloc(sizeof(*found), GFP_NOFS); - if (!found) - return -ENOMEM; + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); - list_add(&found->list, &info->space_info); - INIT_LIST_HEAD(&found->block_groups); - init_rwsem(&found->groups_sem); - spin_lock_init(&found->lock); - found->flags = flags; - found->total_bytes = total_bytes; - found->bytes_used = bytes_used; - found->bytes_pinned = 0; - found->bytes_reserved = 0; - found->full = 0; - found->force_alloc = 0; - *space_info = found; - return 0; + data_sinfo = BTRFS_I(inode)->space_info; + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; + spin_unlock(&data_sinfo->lock); } -static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) +/* called when we are adding a delalloc extent to the inode's io_tree */ +void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) { - u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | - BTRFS_BLOCK_GROUP_RAID1 | - BTRFS_BLOCK_GROUP_RAID10 | - BTRFS_BLOCK_GROUP_DUP); - if (extra_flags) { - if (flags & BTRFS_BLOCK_GROUP_DATA) - fs_info->avail_data_alloc_bits |= extra_flags; - if (flags & BTRFS_BLOCK_GROUP_METADATA) - fs_info->avail_metadata_alloc_bits |= extra_flags; - if (flags & BTRFS_BLOCK_GROUP_SYSTEM) - fs_info->avail_system_alloc_bits |= extra_flags; + struct btrfs_space_info *data_sinfo; + + /* get the space info for where this inode will be storing its data */ + data_sinfo = BTRFS_I(inode)->space_info; + + /* make sure we have enough space to handle the data first */ + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_delalloc += bytes; + + /* + * we are adding a delalloc extent without calling + * btrfs_check_data_free_space first. This happens on a weird + * writepage condition, but shouldn't hurt our accounting + */ + if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { + data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; + BTRFS_I(inode)->reserved_bytes = 0; + } else { + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; } + + spin_unlock(&data_sinfo->lock); } -static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags) +/* called when we are clearing an delalloc extent from the inode's io_tree */ +void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) { - u64 num_devices = root->fs_info->fs_devices->num_devices; + struct btrfs_space_info *info; - if (num_devices == 1) - flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); - if (num_devices < 4) - flags &= ~BTRFS_BLOCK_GROUP_RAID10; + info = BTRFS_I(inode)->space_info; - if ((flags & BTRFS_BLOCK_GROUP_DUP) && - (flags & (BTRFS_BLOCK_GROUP_RAID1 | - BTRFS_BLOCK_GROUP_RAID10))) { - flags &= ~BTRFS_BLOCK_GROUP_DUP; - } + spin_lock(&info->lock); + info->bytes_delalloc -= bytes; + spin_unlock(&info->lock); +} - if ((flags & BTRFS_BLOCK_GROUP_RAID1) && - (flags & BTRFS_BLOCK_GROUP_RAID10)) { - flags &= ~BTRFS_BLOCK_GROUP_RAID1; - } +static void force_metadata_allocation(struct btrfs_fs_info *info) +{ + struct list_head *head = &info->space_info; + struct btrfs_space_info *found; - if ((flags & BTRFS_BLOCK_GROUP_RAID0) && - ((flags & BTRFS_BLOCK_GROUP_RAID1) | - (flags & BTRFS_BLOCK_GROUP_RAID10) | - (flags & BTRFS_BLOCK_GROUP_DUP))) - flags &= ~BTRFS_BLOCK_GROUP_RAID0; - return flags; + rcu_read_lock(); + list_for_each_entry_rcu(found, head, list) { + if (found->flags & BTRFS_BLOCK_GROUP_METADATA) + found->force_alloc = 1; + } + rcu_read_unlock(); } static int do_chunk_alloc(struct btrfs_trans_handle *trans, @@ -1345,12 +2909,13 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, int force) { struct btrfs_space_info *space_info; + struct btrfs_fs_info *fs_info = extent_root->fs_info; u64 thresh; - u64 start; - u64 num_bytes; - int ret = 0, waited = 0; + int ret = 0; + + mutex_lock(&fs_info->chunk_mutex); - flags = reduce_alloc_profile(extent_root, flags); + flags = btrfs_reduce_alloc_profile(extent_root, flags); space_info = __find_space_info(extent_root->fs_info, flags); if (!space_info) { @@ -1370,46 +2935,33 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans, goto out; } - thresh = div_factor(space_info->total_bytes, 6); + thresh = space_info->total_bytes - space_info->bytes_readonly; + thresh = div_factor(thresh, 6); if (!force && (space_info->bytes_used + space_info->bytes_pinned + space_info->bytes_reserved + alloc_bytes) < thresh) { spin_unlock(&space_info->lock); goto out; } - spin_unlock(&space_info->lock); - ret = mutex_trylock(&extent_root->fs_info->chunk_mutex); - if (!ret && !force) { - goto out; - } else if (!ret) { - mutex_lock(&extent_root->fs_info->chunk_mutex); - waited = 1; - } - - if (waited) { - spin_lock(&space_info->lock); - if (space_info->full) { - spin_unlock(&space_info->lock); - goto out_unlock; - } - spin_unlock(&space_info->lock); + /* + * if we're doing a data chunk, go ahead and make sure that + * we keep a reasonable number of metadata chunks allocated in the + * FS as well. + */ + if (flags & BTRFS_BLOCK_GROUP_DATA) { + fs_info->data_chunk_allocations++; + if (!(fs_info->data_chunk_allocations % + fs_info->metadata_ratio)) + force_metadata_allocation(fs_info); } - ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags); - if (ret) { -printk("space info full %Lu\n", flags); + ret = btrfs_alloc_chunk(trans, extent_root, flags); + if (ret) space_info->full = 1; - goto out_unlock; - } - - ret = btrfs_make_block_group(trans, extent_root, 0, flags, - BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes); - BUG_ON(ret); -out_unlock: - mutex_unlock(&extent_root->fs_info->chunk_mutex); out: + mutex_unlock(&extent_root->fs_info->chunk_mutex); return ret; } @@ -1424,11 +2976,28 @@ static int update_block_group(struct btrfs_trans_handle *trans, u64 old_val; u64 byte_in_group; - while(total) { + /* block accounting for super block */ + spin_lock(&info->delalloc_lock); + old_val = btrfs_super_bytes_used(&info->super_copy); + if (alloc) + old_val += num_bytes; + 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) { cache = btrfs_lookup_block_group(info, bytenr); - if (!cache) { + if (!cache) return -1; - } byte_in_group = bytenr - cache->key.objectid; WARN_ON(byte_in_group > cache->key.offset); @@ -1440,23 +3009,32 @@ static int update_block_group(struct btrfs_trans_handle *trans, if (alloc) { old_val += num_bytes; cache->space_info->bytes_used += num_bytes; + if (cache->ro) + cache->space_info->bytes_readonly -= num_bytes; btrfs_set_block_group_used(&cache->item, old_val); spin_unlock(&cache->lock); spin_unlock(&cache->space_info->lock); } else { old_val -= num_bytes; cache->space_info->bytes_used -= num_bytes; + if (cache->ro) + cache->space_info->bytes_readonly += num_bytes; btrfs_set_block_group_used(&cache->item, old_val); spin_unlock(&cache->lock); spin_unlock(&cache->space_info->lock); if (mark_free) { int ret; + + ret = btrfs_discard_extent(root, bytenr, + num_bytes); + WARN_ON(ret); + ret = btrfs_add_free_space(cache, bytenr, num_bytes); - if (ret) - return -1; + WARN_ON(ret); } } + btrfs_put_block_group(cache); total -= num_bytes; bytenr += num_bytes; } @@ -1466,12 +3044,16 @@ static int update_block_group(struct btrfs_trans_handle *trans, static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) { struct btrfs_block_group_cache *cache; + u64 bytenr; cache = btrfs_lookup_first_block_group(root->fs_info, search_start); if (!cache) return 0; - return cache->key.objectid; + bytenr = cache->key.objectid; + btrfs_put_block_group(cache); + + return bytenr; } int btrfs_update_pinned_extents(struct btrfs_root *root, @@ -1481,14 +3063,10 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, struct btrfs_block_group_cache *cache; struct btrfs_fs_info *fs_info = root->fs_info; - WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex)); - if (pin) { + if (pin) set_extent_dirty(&fs_info->pinned_extents, bytenr, bytenr + num - 1, GFP_NOFS); - } else { - clear_extent_dirty(&fs_info->pinned_extents, - bytenr, bytenr + num - 1, GFP_NOFS); - } + while (num > 0) { cache = btrfs_lookup_block_group(fs_info, bytenr); BUG_ON(!cache); @@ -1503,14 +3081,37 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, spin_unlock(&cache->space_info->lock); fs_info->total_pinned += len; } else { + int unpin = 0; + + /* + * in order to not race with the block group caching, we + * only want to unpin the extent if we are cached. If + * we aren't cached, we want to start async caching this + * block group so we can free the extent the next time + * around. + */ spin_lock(&cache->space_info->lock); spin_lock(&cache->lock); - cache->pinned -= len; - cache->space_info->bytes_pinned -= len; + unpin = (cache->cached == BTRFS_CACHE_FINISHED); + if (likely(unpin)) { + cache->pinned -= len; + cache->space_info->bytes_pinned -= len; + fs_info->total_pinned -= len; + } spin_unlock(&cache->lock); spin_unlock(&cache->space_info->lock); - fs_info->total_pinned -= len; + + if (likely(unpin)) + clear_extent_dirty(&fs_info->pinned_extents, + bytenr, bytenr + len -1, + GFP_NOFS); + else + cache_block_group(cache); + + if (unpin) + btrfs_add_free_space(cache, bytenr, len); } + btrfs_put_block_group(cache); bytenr += len; num -= len; } @@ -1541,6 +3142,7 @@ static int update_reserved_extents(struct btrfs_root *root, } spin_unlock(&cache->lock); spin_unlock(&cache->space_info->lock); + btrfs_put_block_group(cache); bytenr += len; num -= len; } @@ -1555,16 +3157,15 @@ int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy) struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents; int ret; - mutex_lock(&root->fs_info->pinned_mutex); - while(1) { + while (1) { ret = find_first_extent_bit(pinned_extents, last, &start, &end, EXTENT_DIRTY); if (ret) break; + set_extent_dirty(copy, start, end, GFP_NOFS); last = end + 1; } - mutex_unlock(&root->fs_info->pinned_mutex); return 0; } @@ -1575,139 +3176,30 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, u64 start; u64 end; int ret; - struct btrfs_block_group_cache *cache; - mutex_lock(&root->fs_info->pinned_mutex); - while(1) { + while (1) { ret = find_first_extent_bit(unpin, 0, &start, &end, EXTENT_DIRTY); if (ret) break; - btrfs_update_pinned_extents(root, start, end + 1 - start, 0); - clear_extent_dirty(unpin, start, end, GFP_NOFS); - cache = btrfs_lookup_block_group(root->fs_info, start); - if (cache->cached) - btrfs_add_free_space(cache, start, end - start + 1); - if (need_resched()) { - mutex_unlock(&root->fs_info->pinned_mutex); - cond_resched(); - mutex_lock(&root->fs_info->pinned_mutex); - } - } - mutex_unlock(&root->fs_info->pinned_mutex); - return 0; -} - -static int finish_current_insert(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, int all) -{ - u64 start; - u64 end; - u64 priv; - u64 search = 0; - struct btrfs_fs_info *info = extent_root->fs_info; - struct btrfs_path *path; - struct btrfs_extent_ref *ref; - struct pending_extent_op *extent_op; - struct btrfs_key key; - struct btrfs_extent_item extent_item; - int ret; - int err = 0; - - btrfs_set_stack_extent_refs(&extent_item, 1); - path = btrfs_alloc_path(); - - while(1) { - mutex_lock(&info->extent_ins_mutex); - ret = find_first_extent_bit(&info->extent_ins, search, &start, - &end, EXTENT_WRITEBACK); - if (ret) { - mutex_unlock(&info->extent_ins_mutex); - if (search && all) { - search = 0; - continue; - } - break; - } - - ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS); - if (!ret) { - search = end + 1; - mutex_unlock(&info->extent_ins_mutex); - cond_resched(); - continue; - } - BUG_ON(ret < 0); - - ret = get_state_private(&info->extent_ins, start, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *)(unsigned long)priv; - - mutex_unlock(&info->extent_ins_mutex); - - if (extent_op->type == PENDING_EXTENT_INSERT) { - key.objectid = start; - key.offset = end + 1 - start; - key.type = BTRFS_EXTENT_ITEM_KEY; - err = btrfs_insert_item(trans, extent_root, &key, - &extent_item, sizeof(extent_item)); - BUG_ON(err); - - mutex_lock(&info->extent_ins_mutex); - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_WRITEBACK, GFP_NOFS); - mutex_unlock(&info->extent_ins_mutex); - err = insert_extent_backref(trans, extent_root, path, - start, extent_op->parent, - extent_root->root_key.objectid, - extent_op->generation, - extent_op->level); - BUG_ON(err); - } else if (extent_op->type == PENDING_BACKREF_UPDATE) { - err = lookup_extent_backref(trans, extent_root, path, - start, extent_op->orig_parent, - extent_root->root_key.objectid, - extent_op->orig_generation, - extent_op->level, 0); - BUG_ON(err); + ret = btrfs_discard_extent(root, start, end + 1 - start); - mutex_lock(&info->extent_ins_mutex); - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_WRITEBACK, GFP_NOFS); - mutex_unlock(&info->extent_ins_mutex); - - key.objectid = start; - key.offset = extent_op->parent; - key.type = BTRFS_EXTENT_REF_KEY; - err = btrfs_set_item_key_safe(trans, extent_root, path, - &key); - BUG_ON(err); - ref = btrfs_item_ptr(path->nodes[0], path->slots[0], - struct btrfs_extent_ref); - btrfs_set_ref_generation(path->nodes[0], ref, - extent_op->generation); - btrfs_mark_buffer_dirty(path->nodes[0]); - btrfs_release_path(extent_root, path); - } else { - BUG_ON(1); - } - kfree(extent_op); - unlock_extent(&info->extent_ins, start, end, GFP_NOFS); - if (all) - search = 0; - else - search = end + 1; + /* unlocks the pinned mutex */ + btrfs_update_pinned_extents(root, start, end + 1 - start, 0); + clear_extent_dirty(unpin, start, end, GFP_NOFS); cond_resched(); } - btrfs_free_path(path); - return 0; + + return ret; } static int pin_down_bytes(struct btrfs_trans_handle *trans, struct btrfs_root *root, - u64 bytenr, u64 num_bytes, int is_data) + struct btrfs_path *path, + u64 bytenr, u64 num_bytes, int is_data, + struct extent_buffer **must_clean) { int err = 0; struct extent_buffer *buf; @@ -1729,363 +3221,344 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans, u64 header_owner = btrfs_header_owner(buf); u64 header_transid = btrfs_header_generation(buf); if (header_owner != BTRFS_TREE_LOG_OBJECTID && - header_owner != BTRFS_TREE_RELOC_OBJECTID && header_transid == trans->transid && !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { - clean_tree_block(NULL, root, buf); - btrfs_tree_unlock(buf); - free_extent_buffer(buf); + *must_clean = buf; return 1; } btrfs_tree_unlock(buf); } free_extent_buffer(buf); pinit: + btrfs_set_path_blocking(path); + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); BUG_ON(err < 0); return 0; } -/* - * remove an extent from the root, returns 0 on success - */ -static int __free_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner_objectid, int pin, int mark_free) + +static int __btrfs_free_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 owner_objectid, + u64 owner_offset, int refs_to_drop, + struct btrfs_delayed_extent_op *extent_op) { - struct btrfs_path *path; struct btrfs_key key; + struct btrfs_path *path; struct btrfs_fs_info *info = root->fs_info; struct btrfs_root *extent_root = info->extent_root; struct extent_buffer *leaf; + struct btrfs_extent_item *ei; + struct btrfs_extent_inline_ref *iref; int ret; + int is_data; int extent_slot = 0; int found_extent = 0; int num_to_del = 1; - struct btrfs_extent_item *ei; - u32 refs; + u32 item_size; + u64 refs; - key.objectid = bytenr; - btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); - key.offset = num_bytes; path = btrfs_alloc_path(); if (!path) return -ENOMEM; path->reada = 1; - ret = lookup_extent_backref(trans, extent_root, path, - bytenr, parent, root_objectid, - ref_generation, owner_objectid, 1); + path->leave_spinning = 1; + + is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; + BUG_ON(!is_data && refs_to_drop != 1); + + ret = lookup_extent_backref(trans, extent_root, path, &iref, + bytenr, num_bytes, parent, + root_objectid, owner_objectid, + owner_offset); if (ret == 0) { - struct btrfs_key found_key; extent_slot = path->slots[0]; - while(extent_slot > 0) { - extent_slot--; - btrfs_item_key_to_cpu(path->nodes[0], &found_key, + while (extent_slot >= 0) { + btrfs_item_key_to_cpu(path->nodes[0], &key, extent_slot); - if (found_key.objectid != bytenr) + if (key.objectid != bytenr) break; - if (found_key.type == BTRFS_EXTENT_ITEM_KEY && - found_key.offset == num_bytes) { + if (key.type == BTRFS_EXTENT_ITEM_KEY && + key.offset == num_bytes) { found_extent = 1; break; } if (path->slots[0] - extent_slot > 5) break; + extent_slot--; } +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); + if (found_extent && item_size < sizeof(*ei)) + found_extent = 0; +#endif if (!found_extent) { - ret = remove_extent_backref(trans, extent_root, path); + BUG_ON(iref); + ret = remove_extent_backref(trans, extent_root, path, + NULL, refs_to_drop, + is_data); BUG_ON(ret); btrfs_release_path(extent_root, path); + path->leave_spinning = 1; + + key.objectid = bytenr; + key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = num_bytes; + ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); + if (ret) { + printk(KERN_ERR "umm, got %d back from search" + ", was looking for %llu\n", ret, + (unsigned long long)bytenr); + btrfs_print_leaf(extent_root, path->nodes[0]); + } BUG_ON(ret); extent_slot = path->slots[0]; } } else { btrfs_print_leaf(extent_root, path->nodes[0]); WARN_ON(1); - printk("Unable to find ref byte nr %Lu root %Lu " - "gen %Lu owner %Lu\n", bytenr, - root_objectid, ref_generation, owner_objectid); + printk(KERN_ERR "btrfs unable to find ref byte nr %llu " + "parent %llu root %llu owner %llu offset %llu\n", + (unsigned long long)bytenr, + (unsigned long long)parent, + (unsigned long long)root_objectid, + (unsigned long long)owner_objectid, + (unsigned long long)owner_offset); } leaf = path->nodes[0]; - ei = btrfs_item_ptr(leaf, extent_slot, - struct btrfs_extent_item); - refs = btrfs_extent_refs(leaf, ei); - BUG_ON(refs == 0); - refs -= 1; - btrfs_set_extent_refs(leaf, ei, refs); + item_size = btrfs_item_size_nr(leaf, extent_slot); +#ifdef BTRFS_COMPAT_EXTENT_TREE_V0 + if (item_size < sizeof(*ei)) { + BUG_ON(found_extent || extent_slot != path->slots[0]); + ret = convert_extent_item_v0(trans, extent_root, path, + owner_objectid, 0); + BUG_ON(ret < 0); - btrfs_mark_buffer_dirty(leaf); + btrfs_release_path(extent_root, path); + path->leave_spinning = 1; - if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { - struct btrfs_extent_ref *ref; - ref = btrfs_item_ptr(leaf, path->slots[0], - struct btrfs_extent_ref); - BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); - /* if the back ref and the extent are next to each other - * they get deleted below in one shot - */ - path->slots[0] = extent_slot; - num_to_del = 2; - } else if (found_extent) { - /* otherwise delete the extent back ref */ - ret = remove_extent_backref(trans, extent_root, path); - BUG_ON(ret); - /* if refs are 0, we need to setup the path for deletion */ - if (refs == 0) { - btrfs_release_path(extent_root, path); - ret = btrfs_search_slot(trans, extent_root, &key, path, - -1, 1); - BUG_ON(ret); + key.objectid = bytenr; + key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = num_bytes; + + ret = btrfs_search_slot(trans, extent_root, &key, path, + -1, 1); + if (ret) { + printk(KERN_ERR "umm, got %d back from search" + ", was looking for %llu\n", ret, + (unsigned long long)bytenr); + btrfs_print_leaf(extent_root, path->nodes[0]); } + BUG_ON(ret); + extent_slot = path->slots[0]; + leaf = path->nodes[0]; + item_size = btrfs_item_size_nr(leaf, extent_slot); } - - if (refs == 0) { - u64 super_used; - u64 root_used; -#ifdef BIO_RW_DISCARD - u64 map_length = num_bytes; - struct btrfs_multi_bio *multi = NULL; #endif + BUG_ON(item_size < sizeof(*ei)); + ei = btrfs_item_ptr(leaf, extent_slot, + struct btrfs_extent_item); + if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { + struct btrfs_tree_block_info *bi; + BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); + bi = (struct btrfs_tree_block_info *)(ei + 1); + WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); + } - if (pin) { - mutex_lock(&root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, root, bytenr, num_bytes, - owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); - mutex_unlock(&root->fs_info->pinned_mutex); - if (ret > 0) - mark_free = 1; - BUG_ON(ret < 0); + refs = btrfs_extent_refs(leaf, ei); + BUG_ON(refs < refs_to_drop); + refs -= refs_to_drop; + + if (refs > 0) { + if (extent_op) + __run_delayed_extent_op(extent_op, leaf, ei); + /* + * In the case of inline back ref, reference count will + * be updated by remove_extent_backref + */ + if (iref) { + BUG_ON(!found_extent); + } else { + btrfs_set_extent_refs(leaf, ei, refs); + btrfs_mark_buffer_dirty(leaf); + } + if (found_extent) { + ret = remove_extent_backref(trans, extent_root, path, + iref, refs_to_drop, + is_data); + BUG_ON(ret); + } + } else { + int mark_free = 0; + struct extent_buffer *must_clean = NULL; + + if (found_extent) { + BUG_ON(is_data && refs_to_drop != + extent_data_ref_count(root, path, iref)); + if (iref) { + BUG_ON(path->slots[0] != extent_slot); + } else { + BUG_ON(path->slots[0] != extent_slot + 1); + path->slots[0] = extent_slot; + num_to_del = 2; + } } - /* block accounting for super block */ - spin_lock_irq(&info->delalloc_lock); - super_used = btrfs_super_bytes_used(&info->super_copy); - btrfs_set_super_bytes_used(&info->super_copy, - super_used - num_bytes); - spin_unlock_irq(&info->delalloc_lock); + ret = pin_down_bytes(trans, root, path, bytenr, + num_bytes, is_data, &must_clean); + if (ret > 0) + mark_free = 1; + BUG_ON(ret < 0); + /* + * it is going to be very rare for someone to be waiting + * on the block we're freeing. del_items might need to + * schedule, so rather than get fancy, just force it + * to blocking here + */ + if (must_clean) + btrfs_set_lock_blocking(must_clean); - /* block accounting for root item */ - root_used = btrfs_root_used(&root->root_item); - btrfs_set_root_used(&root->root_item, - root_used - num_bytes); ret = btrfs_del_items(trans, extent_root, path, path->slots[0], num_to_del); BUG_ON(ret); btrfs_release_path(extent_root, path); - ret = update_block_group(trans, root, bytenr, num_bytes, 0, - mark_free); - BUG_ON(ret); - -#ifdef BIO_RW_DISCARD - /* 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); - if (!ret) { - struct btrfs_bio_stripe *stripe = multi->stripes; - int i; - if (map_length > num_bytes) - map_length = num_bytes; + if (must_clean) { + clean_tree_block(NULL, root, must_clean); + btrfs_tree_unlock(must_clean); + free_extent_buffer(must_clean); + } - for (i = 0; i < multi->num_stripes; i++, stripe++) { - blkdev_issue_discard(stripe->dev->bdev, - stripe->physical >> 9, - map_length >> 9); - } - kfree(multi); + if (is_data) { + ret = btrfs_del_csums(trans, root, bytenr, num_bytes); + BUG_ON(ret); + } else { + invalidate_mapping_pages(info->btree_inode->i_mapping, + bytenr >> PAGE_CACHE_SHIFT, + (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); } -#endif + + ret = update_block_group(trans, root, bytenr, num_bytes, 0, + mark_free); + BUG_ON(ret); } btrfs_free_path(path); - finish_current_insert(trans, extent_root, 0); return ret; } /* - * find all the blocks marked as pending in the radix tree and remove - * them from the extent map + * when we free an extent, it is possible (and likely) that we free the last + * delayed ref for that extent as well. This searches the delayed ref tree for + * a given extent, and if there are no other delayed refs to be processed, it + * removes it from the tree. */ -static int del_pending_extents(struct btrfs_trans_handle *trans, struct - btrfs_root *extent_root, int all) +static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 bytenr) { + struct btrfs_delayed_ref_head *head; + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct rb_node *node; int ret; - int err = 0; - u64 start; - u64 end; - u64 priv; - u64 search = 0; - struct extent_io_tree *pending_del; - struct extent_io_tree *extent_ins; - struct pending_extent_op *extent_op; - struct btrfs_fs_info *info = extent_root->fs_info; - - extent_ins = &extent_root->fs_info->extent_ins; - pending_del = &extent_root->fs_info->pending_del; - - while(1) { - mutex_lock(&info->extent_ins_mutex); - ret = find_first_extent_bit(pending_del, search, &start, &end, - EXTENT_WRITEBACK); - if (ret) { - mutex_unlock(&info->extent_ins_mutex); - if (all && search) { - search = 0; - continue; - } - break; - } - - ret = try_lock_extent(extent_ins, start, end, GFP_NOFS); - if (!ret) { - search = end+1; - mutex_unlock(&info->extent_ins_mutex); - cond_resched(); - continue; - } - BUG_ON(ret < 0); - - ret = get_state_private(pending_del, start, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *)(unsigned long)priv; - - clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK, - GFP_NOFS); - if (!test_range_bit(extent_ins, start, end, - EXTENT_WRITEBACK, 0)) { - mutex_unlock(&info->extent_ins_mutex); -free_extent: - ret = __free_extent(trans, extent_root, - start, end + 1 - start, - extent_op->orig_parent, - extent_root->root_key.objectid, - extent_op->orig_generation, - extent_op->level, 1, 0); - kfree(extent_op); - } else { - kfree(extent_op); - - ret = get_state_private(&info->extent_ins, start, - &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *) - (unsigned long)priv; - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_WRITEBACK, GFP_NOFS); - - mutex_unlock(&info->extent_ins_mutex); - - if (extent_op->type == PENDING_BACKREF_UPDATE) - goto free_extent; + delayed_refs = &trans->transaction->delayed_refs; + spin_lock(&delayed_refs->lock); + head = btrfs_find_delayed_ref_head(trans, bytenr); + if (!head) + goto out; - mutex_lock(&extent_root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, extent_root, start, - end + 1 - start, 0); - mutex_unlock(&extent_root->fs_info->pinned_mutex); + node = rb_prev(&head->node.rb_node); + if (!node) + goto out; - ret = update_block_group(trans, extent_root, start, - end + 1 - start, 0, ret > 0); + ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); - BUG_ON(ret); - kfree(extent_op); - } - if (ret) - err = ret; - unlock_extent(extent_ins, start, end, GFP_NOFS); + /* there are still entries for this ref, we can't drop it */ + if (ref->bytenr == bytenr) + goto out; - if (all) - search = 0; - else - search = end + 1; - cond_resched(); + if (head->extent_op) { + if (!head->must_insert_reserved) + goto out; + kfree(head->extent_op); + head->extent_op = NULL; } - return err; -} - -/* - * remove an extent from the root, returns 0 on success - */ -static int __btrfs_free_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner_objectid, int pin) -{ - struct btrfs_root *extent_root = root->fs_info->extent_root; - int pending_ret; - int ret; - WARN_ON(num_bytes < root->sectorsize); - if (root == extent_root) { - struct pending_extent_op *extent_op; + /* + * waiting for the lock here would deadlock. If someone else has it + * locked they are already in the process of dropping it anyway + */ + if (!mutex_trylock(&head->mutex)) + goto out; - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); + /* + * at this point we have a head with no other entries. Go + * ahead and process it. + */ + head->node.in_tree = 0; + rb_erase(&head->node.rb_node, &delayed_refs->root); - extent_op->type = PENDING_EXTENT_DELETE; - extent_op->bytenr = bytenr; - extent_op->num_bytes = num_bytes; - extent_op->parent = parent; - extent_op->orig_parent = parent; - extent_op->generation = ref_generation; - extent_op->orig_generation = ref_generation; - extent_op->level = (int)owner_objectid; - - mutex_lock(&root->fs_info->extent_ins_mutex); - set_extent_bits(&root->fs_info->pending_del, - bytenr, bytenr + num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->pending_del, - bytenr, (unsigned long)extent_op); - mutex_unlock(&root->fs_info->extent_ins_mutex); - return 0; - } - /* if metadata always pin */ - if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { - if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { - struct btrfs_block_group_cache *cache; - - /* btrfs_free_reserved_extent */ - cache = btrfs_lookup_block_group(root->fs_info, bytenr); - BUG_ON(!cache); - btrfs_add_free_space(cache, bytenr, num_bytes); - update_reserved_extents(root, bytenr, num_bytes, 0); - return 0; - } - pin = 1; - } + delayed_refs->num_entries--; - /* if data pin when any transaction has committed this */ - if (ref_generation != trans->transid) - pin = 1; + /* + * we don't take a ref on the node because we're removing it from the + * tree, so we just steal the ref the tree was holding. + */ + delayed_refs->num_heads--; + if (list_empty(&head->cluster)) + delayed_refs->num_heads_ready--; - ret = __free_extent(trans, root, bytenr, num_bytes, parent, - root_objectid, ref_generation, - owner_objectid, pin, pin == 0); + list_del_init(&head->cluster); + spin_unlock(&delayed_refs->lock); - finish_current_insert(trans, root->fs_info->extent_root, 0); - pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); - return ret ? ret : pending_ret; + ret = run_one_delayed_ref(trans, root->fs_info->tree_root, + &head->node, head->extent_op, + head->must_insert_reserved); + BUG_ON(ret); + btrfs_put_delayed_ref(&head->node); + return 0; +out: + spin_unlock(&delayed_refs->lock); + return 0; } int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner_objectid, int pin) + u64 root_objectid, u64 owner, u64 offset) { int ret; - ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, - root_objectid, ref_generation, - owner_objectid, pin); + /* + * tree log blocks never actually go into the extent allocation + * tree, just update pinning info and exit early. + */ + if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { + WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); + /* unlocks the pinned mutex */ + btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); + update_reserved_extents(root, bytenr, num_bytes, 0); + ret = 0; + } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { + ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, + parent, root_objectid, (int)owner, + BTRFS_DROP_DELAYED_REF, NULL); + BUG_ON(ret); + ret = check_ref_cleanup(trans, root, bytenr); + BUG_ON(ret); + } else { + ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, + parent, root_objectid, owner, + offset, BTRFS_DROP_DELAYED_REF, NULL); + BUG_ON(ret); + } return ret; } @@ -2097,6 +3570,45 @@ static u64 stripe_align(struct btrfs_root *root, u64 val) } /* + * when we wait for progress in the block group caching, its because + * our allocation attempt failed at least once. So, we must sleep + * and let some progress happen before we try again. + * + * This function will sleep at least once waiting for new free space to + * show up, and then it will check the block group free space numbers + * for our min num_bytes. Another option is to have it go ahead + * and look in the rbtree for a free extent of a given size, but this + * is a good start. + */ +static noinline int +wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, + u64 num_bytes) +{ + DEFINE_WAIT(wait); + + prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE); + + if (block_group_cache_done(cache)) { + finish_wait(&cache->caching_q, &wait); + return 0; + } + schedule(); + finish_wait(&cache->caching_q, &wait); + + wait_event(cache->caching_q, block_group_cache_done(cache) || + (cache->free_space >= num_bytes)); + return 0; +} + +enum btrfs_loop_type { + LOOP_CACHED_ONLY = 0, + LOOP_CACHING_NOWAIT = 1, + LOOP_CACHING_WAIT = 2, + LOOP_ALLOC_CHUNK = 3, + LOOP_NO_EMPTY_SIZE = 4, +}; + +/* * walks the btree of allocated extents and find a hole of a given size. * The key ins is changed to record the hole: * ins->objectid == block start @@ -2104,7 +3616,7 @@ static u64 stripe_align(struct btrfs_root *root, u64 val) * ins->offset == number of blocks * Any available blocks before search_start are skipped. */ -static int noinline find_free_extent(struct btrfs_trans_handle *trans, +static noinline int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *orig_root, u64 num_bytes, u64 empty_size, u64 search_start, u64 search_end, @@ -2113,225 +3625,347 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans, int data) { int ret = 0; - struct btrfs_root * root = orig_root->fs_info->extent_root; - u64 total_needed = num_bytes; - u64 *last_ptr = NULL; - u64 last_wanted = 0; + struct btrfs_root *root = orig_root->fs_info->extent_root; + struct btrfs_free_cluster *last_ptr = NULL; struct btrfs_block_group_cache *block_group = NULL; - int chunk_alloc_done = 0; int empty_cluster = 2 * 1024 * 1024; int allowed_chunk_alloc = 0; - struct list_head *head = NULL, *cur = NULL; - int loop = 0; struct btrfs_space_info *space_info; + int last_ptr_loop = 0; + int loop = 0; + bool found_uncached_bg = false; WARN_ON(num_bytes < root->sectorsize); btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); ins->objectid = 0; ins->offset = 0; + space_info = __find_space_info(root->fs_info, data); + if (orig_root->ref_cows || empty_size) allowed_chunk_alloc = 1; if (data & BTRFS_BLOCK_GROUP_METADATA) { - last_ptr = &root->fs_info->last_alloc; - empty_cluster = 64 * 1024; + last_ptr = &root->fs_info->meta_alloc_cluster; + if (!btrfs_test_opt(root, SSD)) + empty_cluster = 64 * 1024; } - if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) - last_ptr = &root->fs_info->last_data_alloc; + if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { + last_ptr = &root->fs_info->data_alloc_cluster; + } if (last_ptr) { - if (*last_ptr) { - hint_byte = *last_ptr; - last_wanted = *last_ptr; - } else - empty_size += empty_cluster; - } else { - empty_cluster = 0; + spin_lock(&last_ptr->lock); + if (last_ptr->block_group) + hint_byte = last_ptr->window_start; + spin_unlock(&last_ptr->lock); } + search_start = max(search_start, first_logical_byte(root, 0)); search_start = max(search_start, hint_byte); - total_needed += empty_size; - if (search_start != last_wanted) - last_wanted = 0; - - block_group = btrfs_lookup_block_group(root->fs_info, search_start); - if (!block_group) - block_group = btrfs_lookup_first_block_group(root->fs_info, - search_start); - space_info = __find_space_info(root->fs_info, data); + if (!last_ptr) + empty_cluster = 0; - down_read(&space_info->groups_sem); - while (1) { - struct btrfs_free_space *free_space; + if (search_start == hint_byte) { + block_group = btrfs_lookup_block_group(root->fs_info, + search_start); /* - * the only way this happens if our hint points to a block - * group thats not of the proper type, while looping this - * should never happen + * we don't want to use the block group if it doesn't match our + * allocation bits, or if its not cached. */ - WARN_ON(!block_group); - mutex_lock(&block_group->alloc_mutex); - if (unlikely(!block_group_bits(block_group, data))) - goto new_group; + if (block_group && block_group_bits(block_group, data) && + block_group_cache_done(block_group)) { + down_read(&space_info->groups_sem); + if (list_empty(&block_group->list) || + block_group->ro) { + /* + * someone is removing this block group, + * we can't jump into the have_block_group + * target because our list pointers are not + * valid + */ + btrfs_put_block_group(block_group); + up_read(&space_info->groups_sem); + } else + goto have_block_group; + } else if (block_group) { + btrfs_put_block_group(block_group); + } + } - ret = cache_block_group(root, block_group); - if (ret) { - mutex_unlock(&block_group->alloc_mutex); - break; +search: + down_read(&space_info->groups_sem); + list_for_each_entry(block_group, &space_info->block_groups, list) { + u64 offset; + int cached; + + atomic_inc(&block_group->count); + search_start = block_group->key.objectid; + +have_block_group: + if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { + /* + * 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. + */ + if (loop > LOOP_CACHING_NOWAIT || + atomic_read(&space_info->caching_threads) < 2) { + ret = cache_block_group(block_group); + BUG_ON(ret); + } } - if (block_group->ro) - goto new_group; + cached = block_group_cache_done(block_group); + if (unlikely(!cached)) { + found_uncached_bg = true; - free_space = btrfs_find_free_space(block_group, search_start, - total_needed); - if (free_space) { - u64 start = block_group->key.objectid; - u64 end = block_group->key.objectid + - block_group->key.offset; + /* if we only want cached bgs, loop */ + if (loop == LOOP_CACHED_ONLY) + goto loop; + } + + if (unlikely(block_group->ro)) + goto loop; + + if (last_ptr) { + /* + * the refill lock keeps out other + * people trying to start a new cluster + */ + spin_lock(&last_ptr->refill_lock); + if (last_ptr->block_group && + (last_ptr->block_group->ro || + !block_group_bits(last_ptr->block_group, data))) { + offset = 0; + goto refill_cluster; + } - search_start = stripe_align(root, free_space->offset); + offset = btrfs_alloc_from_cluster(block_group, last_ptr, + num_bytes, search_start); + if (offset) { + /* we have a block, we're done */ + spin_unlock(&last_ptr->refill_lock); + goto checks; + } - /* move on to the next group */ - if (search_start + num_bytes >= search_end) - goto new_group; + spin_lock(&last_ptr->lock); + /* + * whoops, this cluster doesn't actually point to + * this block group. Get a ref on the block + * group is does point to and try again + */ + if (!last_ptr_loop && last_ptr->block_group && + last_ptr->block_group != block_group) { - /* move on to the next group */ - if (search_start + num_bytes > end) - goto new_group; + btrfs_put_block_group(block_group); + block_group = last_ptr->block_group; + atomic_inc(&block_group->count); + spin_unlock(&last_ptr->lock); + spin_unlock(&last_ptr->refill_lock); - if (last_wanted && search_start != last_wanted) { - total_needed += empty_cluster; - last_wanted = 0; + last_ptr_loop = 1; + search_start = block_group->key.objectid; /* - * if search_start is still in this block group - * then we just re-search this block group + * we know this block group is properly + * in the list because + * btrfs_remove_block_group, drops the + * cluster before it removes the block + * group from the list */ - if (search_start >= start && - search_start < end) { - mutex_unlock(&block_group->alloc_mutex); - continue; - } - - /* else we go to the next block group */ - goto new_group; + goto have_block_group; } + spin_unlock(&last_ptr->lock); +refill_cluster: + /* + * this cluster didn't work out, free it and + * start over + */ + btrfs_return_cluster_to_free_space(NULL, last_ptr); - if (exclude_nr > 0 && - (search_start + num_bytes > exclude_start && - search_start < exclude_start + exclude_nr)) { - search_start = exclude_start + exclude_nr; + last_ptr_loop = 0; + + /* allocate a cluster in this block group */ + ret = btrfs_find_space_cluster(trans, root, + block_group, last_ptr, + offset, num_bytes, + empty_cluster + empty_size); + if (ret == 0) { /* - * if search_start is still in this block group - * then we just re-search this block group + * now pull our allocation out of this + * cluster */ - if (search_start >= start && - search_start < end) { - mutex_unlock(&block_group->alloc_mutex); - last_wanted = 0; - continue; + offset = btrfs_alloc_from_cluster(block_group, + last_ptr, num_bytes, + search_start); + if (offset) { + /* we found one, proceed */ + spin_unlock(&last_ptr->refill_lock); + goto checks; } + } else if (!cached && loop > LOOP_CACHING_NOWAIT) { + spin_unlock(&last_ptr->refill_lock); - /* else we go to the next block group */ - goto new_group; + wait_block_group_cache_progress(block_group, + num_bytes + empty_cluster + empty_size); + goto have_block_group; } - ins->objectid = search_start; - ins->offset = num_bytes; + /* + * at this point we either didn't find a cluster + * or we weren't able to allocate a block from our + * 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; + } + spin_unlock(&last_ptr->refill_lock); + } - btrfs_remove_free_space_lock(block_group, search_start, - num_bytes); - /* we are all good, lets return */ - mutex_unlock(&block_group->alloc_mutex); - break; + 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)) { + wait_block_group_cache_progress(block_group, + num_bytes + empty_size); + goto have_block_group; } -new_group: - last_wanted = 0; - if (loop > 0) { - total_needed -= empty_cluster; - empty_cluster = 0; +checks: + search_start = stripe_align(root, offset); + /* move on to the next group */ + if (search_start + num_bytes >= search_end) { + btrfs_add_free_space(block_group, offset, num_bytes); + goto loop; } - mutex_unlock(&block_group->alloc_mutex); - /* - * Here's how this works. - * loop == 0: we were searching a block group via a hint - * and didn't find anything, so we start at - * the head of the block groups and keep searching - * loop == 1: we're searching through all of the block groups - * if we hit the head again we have searched - * all of the block groups for this space and we - * need to try and allocate, if we cant error out. - * loop == 2: we allocated more space and are looping through - * all of the block groups again. - */ - if (loop == 0) { - head = &space_info->block_groups; - cur = head->next; - loop++; - } else if (loop == 1 && cur == head) { - if (allowed_chunk_alloc && !chunk_alloc_done) { - up_read(&space_info->groups_sem); - ret = do_chunk_alloc(trans, root, num_bytes + - 2 * 1024 * 1024, data, 1); - if (ret < 0) - break; - down_read(&space_info->groups_sem); + + /* move on to the next group */ + if (search_start + num_bytes > + block_group->key.objectid + block_group->key.offset) { + btrfs_add_free_space(block_group, offset, num_bytes); + goto loop; + } + + if (exclude_nr > 0 && + (search_start + num_bytes > exclude_start && + search_start < exclude_start + exclude_nr)) { + search_start = exclude_start + exclude_nr; + + btrfs_add_free_space(block_group, offset, num_bytes); + /* + * if search_start is still in this block group + * then we just re-search this block group + */ + if (search_start >= block_group->key.objectid && + search_start < (block_group->key.objectid + + block_group->key.offset)) + goto have_block_group; + goto loop; + } + + ins->objectid = search_start; + ins->offset = num_bytes; + + if (offset < search_start) + btrfs_add_free_space(block_group, offset, + search_start - offset); + BUG_ON(offset > search_start); + + /* we are all good, lets return */ + break; +loop: + 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_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 + * again + */ + if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE && + (found_uncached_bg || empty_size || empty_cluster || + allowed_chunk_alloc)) { + if (found_uncached_bg) { + found_uncached_bg = false; + if (loop < LOOP_CACHING_WAIT) { loop++; - head = &space_info->block_groups; - cur = head->next; - chunk_alloc_done = 1; - } else if (!allowed_chunk_alloc) { - space_info->force_alloc = 1; - break; - } else { - break; + goto search; } - } else if (cur == head) { - break; } - block_group = list_entry(cur, struct btrfs_block_group_cache, - list); - search_start = block_group->key.objectid; - cur = cur->next; + if (loop == LOOP_ALLOC_CHUNK) { + empty_size = 0; + empty_cluster = 0; + } + + if (allowed_chunk_alloc) { + ret = do_chunk_alloc(trans, root, num_bytes + + 2 * 1024 * 1024, data, 1); + allowed_chunk_alloc = 0; + } else { + space_info->force_alloc = 1; + } + + if (loop < LOOP_NO_EMPTY_SIZE) { + loop++; + goto search; + } + ret = -ENOSPC; + } else if (!ins->objectid) { + ret = -ENOSPC; } /* we found what we needed */ if (ins->objectid) { if (!(data & BTRFS_BLOCK_GROUP_DATA)) - trans->block_group = block_group; + trans->block_group = block_group->key.objectid; - if (last_ptr) - *last_ptr = ins->objectid + ins->offset; + btrfs_put_block_group(block_group); ret = 0; - } else if (!ret) { - ret = -ENOSPC; } - up_read(&space_info->groups_sem); return ret; } static void dump_space_info(struct btrfs_space_info *info, u64 bytes) { struct btrfs_block_group_cache *cache; - struct list_head *l; - printk(KERN_INFO "space_info has %Lu free, is %sfull\n", - info->total_bytes - info->bytes_used - info->bytes_pinned - - info->bytes_reserved, (info->full) ? "" : "not "); + 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->full) ? "" : "not "); + printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," + " may_use=%llu, used=%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); down_read(&info->groups_sem); - list_for_each(l, &info->block_groups) { - cache = list_entry(l, struct btrfs_block_group_cache, list); + list_for_each_entry(cache, &info->block_groups, list) { spin_lock(&cache->lock); - printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used " - "%Lu pinned %Lu reserved\n", - cache->key.objectid, cache->key.offset, - btrfs_block_group_used(&cache->item), - cache->pinned, cache->reserved); + printk(KERN_INFO "block group %llu has %llu bytes, %llu used " + "%llu pinned %llu reserved\n", + (unsigned long long)cache->key.objectid, + (unsigned long long)cache->key.offset, + (unsigned long long)btrfs_block_group_used(&cache->item), + (unsigned long long)cache->pinned, + (unsigned long long)cache->reserved); btrfs_dump_free_space(cache, bytes); spin_unlock(&cache->lock); } @@ -2347,24 +3981,10 @@ static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, { int ret; u64 search_start = 0; - u64 alloc_profile; struct btrfs_fs_info *info = root->fs_info; - if (data) { - alloc_profile = info->avail_data_alloc_bits & - info->data_alloc_profile; - data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; - } else if (root == root->fs_info->chunk_root) { - alloc_profile = info->avail_system_alloc_bits & - info->system_alloc_profile; - data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; - } else { - alloc_profile = info->avail_metadata_alloc_bits & - info->metadata_alloc_profile; - data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; - } + data = btrfs_get_alloc_profile(root, data); again: - data = reduce_alloc_profile(root, data); /* * the only place that sets empty_size is btrfs_realloc_node, which * is not called recursively on allocations @@ -2395,14 +4015,14 @@ again: num_bytes, data, 1); goto again; } - if (ret) { + if (ret == -ENOSPC) { struct btrfs_space_info *sinfo; sinfo = __find_space_info(root->fs_info, data); - printk("allocation failed flags %Lu, wanted %Lu\n", - data, num_bytes); + 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); - BUG(); } return ret; @@ -2411,15 +4031,22 @@ again: int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) { struct btrfs_block_group_cache *cache; + int ret = 0; cache = btrfs_lookup_block_group(root->fs_info, start); if (!cache) { - printk(KERN_ERR "Unable to find block group for %Lu\n", start); + printk(KERN_ERR "Unable to find block group for %llu\n", + (unsigned long long)start); return -ENOSPC; } + + ret = btrfs_discard_extent(root, start, len); + btrfs_add_free_space(cache, start, len); + btrfs_put_block_group(cache); update_reserved_extents(root, start, len, 0); - return 0; + + return ret; } int btrfs_reserve_extent(struct btrfs_trans_handle *trans, @@ -2433,129 +4060,153 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans, ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, empty_size, hint_byte, search_end, ins, data); - update_reserved_extents(root, ins->objectid, ins->offset, 1); + if (!ret) + update_reserved_extents(root, ins->objectid, ins->offset, 1); + return ret; } -static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner, struct btrfs_key *ins) +static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 parent, u64 root_objectid, + u64 flags, u64 owner, u64 offset, + struct btrfs_key *ins, int ref_mod) { int ret; - int pending_ret; - u64 super_used; - u64 root_used; - u64 num_bytes = ins->offset; - u32 sizes[2]; - struct btrfs_fs_info *info = root->fs_info; - struct btrfs_root *extent_root = info->extent_root; + struct btrfs_fs_info *fs_info = root->fs_info; struct btrfs_extent_item *extent_item; - struct btrfs_extent_ref *ref; + struct btrfs_extent_inline_ref *iref; struct btrfs_path *path; - struct btrfs_key keys[2]; - - if (parent == 0) - parent = ins->objectid; - - /* block accounting for super block */ - spin_lock_irq(&info->delalloc_lock); - super_used = btrfs_super_bytes_used(&info->super_copy); - btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); - spin_unlock_irq(&info->delalloc_lock); - - /* block accounting for root item */ - root_used = btrfs_root_used(&root->root_item); - btrfs_set_root_used(&root->root_item, root_used + num_bytes); - - if (root == extent_root) { - struct pending_extent_op *extent_op; + struct extent_buffer *leaf; + int type; + u32 size; - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); + if (parent > 0) + type = BTRFS_SHARED_DATA_REF_KEY; + else + type = BTRFS_EXTENT_DATA_REF_KEY; - extent_op->type = PENDING_EXTENT_INSERT; - extent_op->bytenr = ins->objectid; - extent_op->num_bytes = ins->offset; - extent_op->parent = parent; - extent_op->orig_parent = 0; - extent_op->generation = ref_generation; - extent_op->orig_generation = 0; - extent_op->level = (int)owner; - - mutex_lock(&root->fs_info->extent_ins_mutex); - set_extent_bits(&root->fs_info->extent_ins, ins->objectid, - ins->objectid + ins->offset - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->extent_ins, - ins->objectid, (unsigned long)extent_op); - mutex_unlock(&root->fs_info->extent_ins_mutex); - goto update_block; - } - - memcpy(&keys[0], ins, sizeof(*ins)); - keys[1].objectid = ins->objectid; - keys[1].type = BTRFS_EXTENT_REF_KEY; - keys[1].offset = parent; - sizes[0] = sizeof(*extent_item); - sizes[1] = sizeof(*ref); + size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); path = btrfs_alloc_path(); BUG_ON(!path); - ret = btrfs_insert_empty_items(trans, extent_root, path, keys, - sizes, 2); + path->leave_spinning = 1; + ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, + ins, size); BUG_ON(ret); - extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], + leaf = path->nodes[0]; + extent_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); - btrfs_set_extent_refs(path->nodes[0], extent_item, 1); - ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, - struct btrfs_extent_ref); - - btrfs_set_ref_root(path->nodes[0], ref, root_objectid); - btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); - btrfs_set_ref_objectid(path->nodes[0], ref, owner); - btrfs_set_ref_num_refs(path->nodes[0], ref, 1); + btrfs_set_extent_refs(leaf, extent_item, ref_mod); + btrfs_set_extent_generation(leaf, extent_item, trans->transid); + btrfs_set_extent_flags(leaf, extent_item, + flags | BTRFS_EXTENT_FLAG_DATA); + + iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); + btrfs_set_extent_inline_ref_type(leaf, iref, type); + if (parent > 0) { + struct btrfs_shared_data_ref *ref; + ref = (struct btrfs_shared_data_ref *)(iref + 1); + btrfs_set_extent_inline_ref_offset(leaf, iref, parent); + btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); + } else { + struct btrfs_extent_data_ref *ref; + ref = (struct btrfs_extent_data_ref *)(&iref->offset); + btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); + btrfs_set_extent_data_ref_objectid(leaf, ref, owner); + btrfs_set_extent_data_ref_offset(leaf, ref, offset); + btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); + } btrfs_mark_buffer_dirty(path->nodes[0]); - - trans->alloc_exclude_start = 0; - trans->alloc_exclude_nr = 0; btrfs_free_path(path); - finish_current_insert(trans, extent_root, 0); - pending_ret = del_pending_extents(trans, extent_root, 0); - if (ret) - goto out; - if (pending_ret) { - ret = pending_ret; - goto out; + ret = update_block_group(trans, root, ins->objectid, ins->offset, + 1, 0); + if (ret) { + printk(KERN_ERR "btrfs update block group failed for %llu " + "%llu\n", (unsigned long long)ins->objectid, + (unsigned long long)ins->offset); + BUG(); + } + return ret; +} + +static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 parent, u64 root_objectid, + u64 flags, struct btrfs_disk_key *key, + int level, struct btrfs_key *ins) +{ + int ret; + struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_extent_item *extent_item; + struct btrfs_tree_block_info *block_info; + struct btrfs_extent_inline_ref *iref; + struct btrfs_path *path; + struct extent_buffer *leaf; + u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); + + path = btrfs_alloc_path(); + BUG_ON(!path); + + path->leave_spinning = 1; + ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, + ins, size); + BUG_ON(ret); + + leaf = path->nodes[0]; + extent_item = btrfs_item_ptr(leaf, path->slots[0], + struct btrfs_extent_item); + btrfs_set_extent_refs(leaf, extent_item, 1); + btrfs_set_extent_generation(leaf, extent_item, trans->transid); + btrfs_set_extent_flags(leaf, extent_item, + flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); + block_info = (struct btrfs_tree_block_info *)(extent_item + 1); + + btrfs_set_tree_block_key(leaf, block_info, key); + btrfs_set_tree_block_level(leaf, block_info, level); + + iref = (struct btrfs_extent_inline_ref *)(block_info + 1); + if (parent > 0) { + BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); + btrfs_set_extent_inline_ref_type(leaf, iref, + BTRFS_SHARED_BLOCK_REF_KEY); + btrfs_set_extent_inline_ref_offset(leaf, iref, parent); + } else { + btrfs_set_extent_inline_ref_type(leaf, iref, + BTRFS_TREE_BLOCK_REF_KEY); + btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); } -update_block: - ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); + btrfs_mark_buffer_dirty(leaf); + btrfs_free_path(path); + + ret = update_block_group(trans, root, ins->objectid, ins->offset, + 1, 0); if (ret) { - printk("update block group failed for %Lu %Lu\n", - ins->objectid, ins->offset); + printk(KERN_ERR "btrfs update block group failed for %llu " + "%llu\n", (unsigned long long)ins->objectid, + (unsigned long long)ins->offset); BUG(); } -out: return ret; } -int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner, struct btrfs_key *ins) +int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 root_objectid, u64 owner, + u64 offset, struct btrfs_key *ins) { int ret; - if (root_objectid == BTRFS_TREE_LOG_OBJECTID) - return 0; - ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, - ref_generation, owner, ins); - update_reserved_extents(root, ins->objectid, ins->offset, 0); + BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); + + ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, + 0, root_objectid, owner, offset, + BTRFS_ADD_DELAYED_EXTENT, NULL); return ret; } @@ -2564,24 +4215,25 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, * an extent has been allocated and makes sure to clear the free * space cache bits as well */ -int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner, struct btrfs_key *ins) +int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 root_objectid, u64 owner, u64 offset, + struct btrfs_key *ins) { int ret; struct btrfs_block_group_cache *block_group; block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); - mutex_lock(&block_group->alloc_mutex); - cache_block_group(root, block_group); + cache_block_group(block_group); + wait_event(block_group->caching_q, + block_group_cache_done(block_group)); - ret = btrfs_remove_free_space_lock(block_group, ins->objectid, - ins->offset); - mutex_unlock(&block_group->alloc_mutex); + ret = btrfs_remove_free_space(block_group, ins->objectid, + ins->offset); BUG_ON(ret); - ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, - ref_generation, owner, ins); + btrfs_put_block_group(block_group); + ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, + 0, owner, offset, ins, 1); return ret; } @@ -2592,34 +4244,56 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, * * returns 0 if everything worked, non-zero otherwise. */ -int btrfs_alloc_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 num_bytes, u64 parent, u64 min_alloc_size, - u64 root_objectid, u64 ref_generation, - u64 owner_objectid, u64 empty_size, u64 hint_byte, - u64 search_end, struct btrfs_key *ins, u64 data) +static int alloc_tree_block(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 num_bytes, u64 parent, u64 root_objectid, + struct btrfs_disk_key *key, int level, + u64 empty_size, u64 hint_byte, u64 search_end, + struct btrfs_key *ins) { int ret; + u64 flags = 0; - ret = __btrfs_reserve_extent(trans, root, num_bytes, - min_alloc_size, empty_size, hint_byte, - search_end, ins, data); - BUG_ON(ret); + ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes, + empty_size, hint_byte, search_end, + ins, 0); + if (ret) + return ret; + + if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { + if (parent == 0) + parent = ins->objectid; + flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; + } else + BUG_ON(parent > 0); + + update_reserved_extents(root, ins->objectid, ins->offset, 1); if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { - ret = __btrfs_alloc_reserved_extent(trans, root, parent, - root_objectid, ref_generation, - owner_objectid, ins); + struct btrfs_delayed_extent_op *extent_op; + extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); + BUG_ON(!extent_op); + if (key) + memcpy(&extent_op->key, key, sizeof(extent_op->key)); + else + memset(&extent_op->key, 0, sizeof(extent_op->key)); + extent_op->flags_to_set = flags; + extent_op->update_key = 1; + extent_op->update_flags = 1; + extent_op->is_data = 0; + + ret = btrfs_add_delayed_tree_ref(trans, ins->objectid, + ins->offset, parent, root_objectid, + level, BTRFS_ADD_DELAYED_EXTENT, + extent_op); BUG_ON(ret); - - } else { - update_reserved_extents(root, ins->objectid, ins->offset, 1); } return ret; } struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, - u64 bytenr, u32 blocksize) + u64 bytenr, u32 blocksize, + int level) { struct extent_buffer *buf; @@ -2627,9 +4301,13 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, if (!buf) return ERR_PTR(-ENOMEM); btrfs_set_header_generation(buf, trans->transid); + btrfs_set_buffer_lockdep_class(buf, level); btrfs_tree_lock(buf); clean_tree_block(trans, root, buf); + + btrfs_set_lock_blocking(buf); 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); @@ -2638,6 +4316,7 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, buf->start + buf->len - 1, GFP_NOFS); } trans->blocks_used++; + /* this returns a buffer locked for blocking */ return buf; } @@ -2646,88 +4325,97 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, * returns the tree buffer or NULL. */ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u32 blocksize, u64 parent, - u64 root_objectid, - u64 ref_generation, - int level, - u64 hint, - u64 empty_size) + struct btrfs_root *root, u32 blocksize, + u64 parent, u64 root_objectid, + struct btrfs_disk_key *key, int level, + u64 hint, u64 empty_size) { struct btrfs_key ins; int ret; struct extent_buffer *buf; - ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize, - root_objectid, ref_generation, level, - empty_size, hint, (u64)-1, &ins, 0); + ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid, + key, level, empty_size, hint, (u64)-1, &ins); if (ret) { BUG_ON(ret > 0); return ERR_PTR(ret); } - buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize); + buf = btrfs_init_new_buffer(trans, root, ins.objectid, + blocksize, level); return buf; } +#if 0 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *leaf) { - u64 leaf_owner; - u64 leaf_generation; + u64 disk_bytenr; + u64 num_bytes; struct btrfs_key key; struct btrfs_file_extent_item *fi; + u32 nritems; int i; - int nritems; int ret; BUG_ON(!btrfs_is_leaf(leaf)); nritems = btrfs_header_nritems(leaf); - leaf_owner = btrfs_header_owner(leaf); - leaf_generation = btrfs_header_generation(leaf); for (i = 0; i < nritems; i++) { - u64 disk_bytenr; 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; - /* - * FIXME make sure to insert a trans record that - * repeats the snapshot del on crash - */ + disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); + + /* holes don't have refs */ if (disk_bytenr == 0) continue; - ret = __btrfs_free_extent(trans, root, disk_bytenr, - btrfs_file_extent_disk_num_bytes(leaf, fi), - leaf->start, leaf_owner, leaf_generation, - key.objectid, 0); + 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); - - atomic_inc(&root->fs_info->throttle_gen); - wake_up(&root->fs_info->transaction_throttle); - cond_resched(); } return 0; } -static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, +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 = ref->extents; + 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++) { - ret = __btrfs_free_extent(trans, root, info->bytenr, + 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); @@ -2740,18 +4428,21 @@ static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans, info++; } + kfree(sorted); return 0; } -int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, - u32 *refs) + +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_ref(NULL, root, start, len, refs); + 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 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. @@ -2772,8 +4463,8 @@ int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, free_extent_buffer(eb); } if (*refs == 1) { - printk("block %llu went down to one during drop_snap\n", - (unsigned long long)start); + printk(KERN_ERR "btrfs block %llu went down to one " + "during drop_snap\n", (unsigned long long)start); } } @@ -2783,11 +4474,159 @@ int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len, 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 int noinline walk_down_tree(struct btrfs_trans_handle *trans, +static noinline int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, int *level) { @@ -2798,14 +4637,13 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, struct extent_buffer *next; struct extent_buffer *cur; struct extent_buffer *parent; - struct btrfs_leaf_ref *ref; u32 blocksize; int ret; u32 refs; WARN_ON(*level < 0); WARN_ON(*level >= BTRFS_MAX_LEVEL); - ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, + ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start, path->nodes[*level]->len, &refs); BUG_ON(ret); if (refs > 1) @@ -2814,7 +4652,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, /* * walk down to the last node level and free all the leaves */ - while(*level >= 0) { + while (*level >= 0) { WARN_ON(*level < 0); WARN_ON(*level >= BTRFS_MAX_LEVEL); cur = path->nodes[*level]; @@ -2825,24 +4663,54 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, 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(root, bytenr, blocksize, &refs); + 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, + ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, root_owner, root_gen, *level - 1, 1); @@ -2854,50 +4722,12 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans, continue; } + /* - * at this point, we have a single ref, and since the - * only place referencing this extent is a dead root - * the reference count should never go higher. - * So, we don't need to check it again + * we need to keep freeing things in the next level down. + * read the block and loop around to process it */ - if (*level == 1) { - 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); - *level = 0; - break; - } - if (printk_ratelimit()) { - printk("leaf ref miss for bytenr %llu\n", - (unsigned long long)bytenr); - } - } - next = btrfs_find_tree_block(root, bytenr, blocksize); - if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) { - free_extent_buffer(next); - - next = read_tree_block(root, bytenr, blocksize, - ptr_gen); - cond_resched(); -#if 0 - /* - * this is a debugging check and can go away - * the ref should never go all the way down to 1 - * at this point - */ - ret = lookup_extent_ref(NULL, root, bytenr, blocksize, - &refs); - BUG_ON(ret); - WARN_ON(refs != 1); -#endif - } + 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]); @@ -2922,303 +4752,536 @@ out: root_owner = btrfs_header_owner(parent); root_gen = btrfs_header_generation(parent); - 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; + /* + * 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]; + struct btrfs_key update_progress; + int stage; + int level; + int shared_level; + int update_ref; + int keep_locks; +}; + +#define DROP_REFERENCE 1 +#define UPDATE_BACKREF 2 + +/* + * hepler to process tree block while walking down the tree. + * + * when wc->stage == DROP_REFERENCE, this function checks + * reference count of the block. if the block is shared and + * we need update back refs for the subtree rooted at the + * block, this function changes wc->stage to UPDATE_BACKREF + * + * when wc->stage == UPDATE_BACKREF, this function updates + * back refs for pointers in the block. + * + * NOTE: return value 1 means we should stop walking down. + */ +static noinline int walk_down_proc(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct walk_control *wc) +{ + int level = wc->level; + struct extent_buffer *eb = path->nodes[level]; + struct btrfs_key key; + u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; + int ret; + + if (wc->stage == UPDATE_BACKREF && + btrfs_header_owner(eb) != root->root_key.objectid) + return 1; + + /* + * 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))) { + BUG_ON(!path->locks[level]); + ret = btrfs_lookup_extent_info(trans, root, + eb->start, eb->len, + &wc->refs[level], + &wc->flags[level]); + BUG_ON(ret); + BUG_ON(wc->refs[level] == 0); + } + + if (wc->stage == DROP_REFERENCE && + wc->update_ref && wc->refs[level] > 1) { + BUG_ON(eb == root->node); + BUG_ON(path->slots[level] > 0); + if (level == 0) + btrfs_item_key_to_cpu(eb, &key, path->slots[level]); + else + btrfs_node_key_to_cpu(eb, &key, path->slots[level]); + if (btrfs_header_owner(eb) == root->root_key.objectid && + btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) { + wc->stage = UPDATE_BACKREF; + wc->shared_level = level; + } + } + + if (wc->stage == DROP_REFERENCE) { + if (wc->refs[level] > 1) + return 1; + + if (path->locks[level] && !wc->keep_locks) { + btrfs_tree_unlock(eb); + path->locks[level] = 0; + } + return 0; + } + + /* wc->stage == UPDATE_BACKREF */ + if (!(wc->flags[level] & flag)) { + BUG_ON(!path->locks[level]); + ret = btrfs_inc_ref(trans, root, eb, 1); + BUG_ON(ret); + ret = btrfs_dec_ref(trans, root, eb, 0); + BUG_ON(ret); + ret = btrfs_set_disk_extent_flags(trans, root, eb->start, + eb->len, flag, 0); + BUG_ON(ret); + wc->flags[level] |= flag; + } + + /* + * the block is shared by multiple trees, so it's not good to + * keep the tree lock + */ + if (path->locks[level] && level > 0) { + btrfs_tree_unlock(eb); + path->locks[level] = 0; + } + return 0; +} + +/* + * hepler to process tree block while walking up the tree. + * + * when wc->stage == DROP_REFERENCE, this function drops + * reference count on the block. + * + * when wc->stage == UPDATE_BACKREF, this function changes + * wc->stage back to DROP_REFERENCE if we changed wc->stage + * to UPDATE_BACKREF previously while processing the block. + * + * NOTE: return value 1 means we should stop walking up. + */ +static noinline int walk_up_proc(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct walk_control *wc) +{ + int ret = 0; + int level = wc->level; + struct extent_buffer *eb = path->nodes[level]; + u64 parent = 0; + + if (wc->stage == UPDATE_BACKREF) { + BUG_ON(wc->shared_level < level); + if (level < wc->shared_level) + goto out; + + BUG_ON(wc->refs[level] <= 1); + ret = find_next_key(path, level + 1, &wc->update_progress); + if (ret > 0) + wc->update_ref = 0; + + wc->stage = DROP_REFERENCE; + wc->shared_level = -1; + path->slots[level] = 0; + + /* + * check reference count again if the block isn't locked. + * we should start walking down the tree again if reference + * count is one. + */ + if (!path->locks[level]) { + BUG_ON(level == 0); + btrfs_tree_lock(eb); + btrfs_set_lock_blocking(eb); + path->locks[level] = 1; + + ret = btrfs_lookup_extent_info(trans, root, + eb->start, eb->len, + &wc->refs[level], + &wc->flags[level]); + BUG_ON(ret); + BUG_ON(wc->refs[level] == 0); + if (wc->refs[level] == 1) { + btrfs_tree_unlock(eb); + path->locks[level] = 0; + return 1; + } + } else { + BUG_ON(level != 0); + } + } + + /* wc->stage == DROP_REFERENCE */ + BUG_ON(wc->refs[level] > 1 && !path->locks[level]); + + if (wc->refs[level] == 1) { + if (level == 0) { + if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) + ret = btrfs_dec_ref(trans, root, eb, 1); + else + ret = btrfs_dec_ref(trans, root, eb, 0); + BUG_ON(ret); + } + /* make block locked assertion in clean_tree_block happy */ + if (!path->locks[level] && + btrfs_header_generation(eb) == trans->transid) { + btrfs_tree_lock(eb); + btrfs_set_lock_blocking(eb); + path->locks[level] = 1; + } + clean_tree_block(trans, root, eb); + } + + if (eb == root->node) { + if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) + parent = eb->start; + else + BUG_ON(root->root_key.objectid != + btrfs_header_owner(eb)); + } else { + if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) + parent = path->nodes[level + 1]->start; + else + BUG_ON(root->root_key.objectid != + btrfs_header_owner(path->nodes[level + 1])); + } + + ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent, + root->root_key.objectid, level, 0); BUG_ON(ret); - - cond_resched(); - return 0; +out: + wc->refs[level] = 0; + wc->flags[level] = 0; + return ret; } -/* - * helper function for drop_subtree, this function is similar to - * walk_down_tree. The main difference is that it checks reference - * counts while tree blocks are locked. - */ -static int noinline walk_down_subtree(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - struct btrfs_path *path, int *level) +static noinline int walk_down_tree(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_path *path, + struct walk_control *wc) { struct extent_buffer *next; struct extent_buffer *cur; - struct extent_buffer *parent; u64 bytenr; u64 ptr_gen; u32 blocksize; - u32 refs; + int level = wc->level; int ret; - cur = path->nodes[*level]; - ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len, - &refs); - BUG_ON(ret); - if (refs > 1) - goto out; + while (level >= 0) { + cur = path->nodes[level]; + BUG_ON(path->slots[level] >= btrfs_header_nritems(cur)); - while (*level >= 0) { - cur = path->nodes[*level]; - if (*level == 0) { - ret = btrfs_drop_leaf_ref(trans, root, cur); - BUG_ON(ret); - clean_tree_block(trans, root, cur); + ret = walk_down_proc(trans, root, path, wc); + if (ret > 0) break; - } - if (path->slots[*level] >= btrfs_header_nritems(cur)) { - clean_tree_block(trans, root, cur); + + if (level == 0) break; - } - bytenr = btrfs_node_blockptr(cur, path->slots[*level]); - blocksize = btrfs_level_size(root, *level - 1); - ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); + bytenr = btrfs_node_blockptr(cur, path->slots[level]); + blocksize = btrfs_level_size(root, level - 1); + ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]); next = read_tree_block(root, bytenr, blocksize, ptr_gen); btrfs_tree_lock(next); + btrfs_set_lock_blocking(next); - ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize, - &refs); - BUG_ON(ret); - if (refs > 1) { - parent = path->nodes[*level]; - ret = btrfs_free_extent(trans, root, bytenr, - blocksize, parent->start, - btrfs_header_owner(parent), - btrfs_header_generation(parent), - *level - 1, 1); - BUG_ON(ret); - path->slots[*level]++; - btrfs_tree_unlock(next); - free_extent_buffer(next); - continue; - } - - *level = btrfs_header_level(next); - path->nodes[*level] = next; - path->slots[*level] = 0; - path->locks[*level] = 1; - cond_resched(); - } -out: - parent = path->nodes[*level + 1]; - bytenr = path->nodes[*level]->start; - blocksize = path->nodes[*level]->len; - - ret = btrfs_free_extent(trans, root, bytenr, blocksize, - parent->start, btrfs_header_owner(parent), - btrfs_header_generation(parent), *level, 1); - BUG_ON(ret); - - if (path->locks[*level]) { - btrfs_tree_unlock(path->nodes[*level]); - path->locks[*level] = 0; + level--; + BUG_ON(level != btrfs_header_level(next)); + path->nodes[level] = next; + path->slots[level] = 0; + path->locks[level] = 1; + wc->level = level; } - free_extent_buffer(path->nodes[*level]); - path->nodes[*level] = NULL; - *level += 1; - cond_resched(); return 0; } -/* - * helper for dropping snapshots. This walks back up the tree in the path - * to find the first node higher up where we haven't yet gone through - * all the slots - */ -static int noinline walk_up_tree(struct btrfs_trans_handle *trans, +static noinline int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, - int *level, int max_level) + struct walk_control *wc, int max_level) { - u64 root_owner; - u64 root_gen; - struct btrfs_root_item *root_item = &root->root_item; - int i; - int slot; + int level = wc->level; int ret; - for (i = *level; i < max_level && path->nodes[i]; i++) { - slot = path->slots[i]; - if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { - struct extent_buffer *node; - struct btrfs_disk_key disk_key; - node = path->nodes[i]; - path->slots[i]++; - *level = i; - WARN_ON(*level == 0); - btrfs_node_key(node, &disk_key, path->slots[i]); - memcpy(&root_item->drop_progress, - &disk_key, sizeof(disk_key)); - root_item->drop_level = i; + path->slots[level] = btrfs_header_nritems(path->nodes[level]); + while (level < max_level && path->nodes[level]) { + wc->level = level; + if (path->slots[level] + 1 < + btrfs_header_nritems(path->nodes[level])) { + path->slots[level]++; return 0; } else { - struct extent_buffer *parent; - if (path->nodes[*level] == root->node) - parent = path->nodes[*level]; - else - parent = path->nodes[*level + 1]; - - root_owner = btrfs_header_owner(parent); - root_gen = btrfs_header_generation(parent); + ret = walk_up_proc(trans, root, path, wc); + if (ret > 0) + return 0; - clean_tree_block(trans, root, path->nodes[*level]); - ret = btrfs_free_extent(trans, root, - path->nodes[*level]->start, - path->nodes[*level]->len, - parent->start, root_owner, - root_gen, *level, 1); - BUG_ON(ret); - if (path->locks[*level]) { - btrfs_tree_unlock(path->nodes[*level]); - path->locks[*level] = 0; + if (path->locks[level]) { + btrfs_tree_unlock(path->nodes[level]); + path->locks[level] = 0; } - free_extent_buffer(path->nodes[*level]); - path->nodes[*level] = NULL; - *level = i + 1; + free_extent_buffer(path->nodes[level]); + path->nodes[level] = NULL; + level++; } } return 1; } /* - * drop the reference count on the tree rooted at 'snap'. This traverses - * the tree freeing any blocks that have a ref count of zero after being - * decremented. + * drop a subvolume tree. + * + * this function traverses the tree freeing any blocks that only + * referenced by the tree. + * + * when a shared tree block is found. this function decreases its + * reference count by one. if update_ref is true, this function + * also make sure backrefs for the shared block and all lower level + * blocks are properly updated. */ -int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root - *root) +int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref) { - int ret = 0; - int wret; - int level; struct btrfs_path *path; - int i; - int orig_level; + struct btrfs_trans_handle *trans; + struct btrfs_root *tree_root = root->fs_info->tree_root; struct btrfs_root_item *root_item = &root->root_item; + struct walk_control *wc; + struct btrfs_key key; + int err = 0; + int ret; + int level; - WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); path = btrfs_alloc_path(); BUG_ON(!path); - level = btrfs_header_level(root->node); - orig_level = level; + wc = kzalloc(sizeof(*wc), GFP_NOFS); + BUG_ON(!wc); + + trans = btrfs_start_transaction(tree_root, 1); + if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { - path->nodes[level] = root->node; - extent_buffer_get(root->node); + level = btrfs_header_level(root->node); + path->nodes[level] = btrfs_lock_root_node(root); + btrfs_set_lock_blocking(path->nodes[level]); path->slots[level] = 0; + path->locks[level] = 1; + memset(&wc->update_progress, 0, + sizeof(wc->update_progress)); } else { - struct btrfs_key key; - struct btrfs_disk_key found_key; - struct extent_buffer *node; - btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); + memcpy(&wc->update_progress, &key, + sizeof(wc->update_progress)); + level = root_item->drop_level; + BUG_ON(level == 0); path->lowest_level = level; - wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); - if (wret < 0) { - ret = wret; + ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); + path->lowest_level = 0; + if (ret < 0) { + err = ret; goto out; } - node = path->nodes[level]; - btrfs_node_key(node, &found_key, path->slots[level]); - WARN_ON(memcmp(&found_key, &root_item->drop_progress, - sizeof(found_key))); + btrfs_node_key_to_cpu(path->nodes[level], &key, + path->slots[level]); + WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key))); + /* * unlock our path, this is safe because only this * function is allowed to delete this snapshot */ - for (i = 0; i < BTRFS_MAX_LEVEL; i++) { - if (path->nodes[i] && path->locks[i]) { - path->locks[i] = 0; - btrfs_tree_unlock(path->nodes[i]); - } + btrfs_unlock_up_safe(path, 0); + + level = btrfs_header_level(root->node); + while (1) { + btrfs_tree_lock(path->nodes[level]); + btrfs_set_lock_blocking(path->nodes[level]); + + ret = btrfs_lookup_extent_info(trans, root, + path->nodes[level]->start, + path->nodes[level]->len, + &wc->refs[level], + &wc->flags[level]); + BUG_ON(ret); + BUG_ON(wc->refs[level] == 0); + + if (level == root_item->drop_level) + break; + + btrfs_tree_unlock(path->nodes[level]); + WARN_ON(wc->refs[level] != 1); + level--; } } - while(1) { - wret = walk_down_tree(trans, root, path, &level); - if (wret > 0) + + wc->level = level; + wc->shared_level = -1; + wc->stage = DROP_REFERENCE; + wc->update_ref = update_ref; + wc->keep_locks = 0; + + while (1) { + ret = walk_down_tree(trans, root, path, wc); + if (ret < 0) { + err = ret; break; - if (wret < 0) - ret = wret; + } - wret = walk_up_tree(trans, root, path, &level, - BTRFS_MAX_LEVEL); - if (wret > 0) + ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); + if (ret < 0) { + err = ret; break; - if (wret < 0) - ret = wret; - if (trans->transaction->in_commit) { - ret = -EAGAIN; + } + + if (ret > 0) { + BUG_ON(wc->stage != DROP_REFERENCE); break; } - atomic_inc(&root->fs_info->throttle_gen); - wake_up(&root->fs_info->transaction_throttle); - } - for (i = 0; i <= orig_level; i++) { - if (path->nodes[i]) { - free_extent_buffer(path->nodes[i]); - path->nodes[i] = NULL; + + if (wc->stage == DROP_REFERENCE) { + level = wc->level; + btrfs_node_key(path->nodes[level], + &root_item->drop_progress, + path->slots[level]); + root_item->drop_level = level; + } + + BUG_ON(wc->level == 0); + if (trans->transaction->in_commit || + trans->transaction->delayed_refs.flushing) { + ret = btrfs_update_root(trans, tree_root, + &root->root_key, + root_item); + BUG_ON(ret); + + btrfs_end_transaction(trans, tree_root); + trans = btrfs_start_transaction(tree_root, 1); + } else { + unsigned long update; + update = trans->delayed_ref_updates; + trans->delayed_ref_updates = 0; + if (update) + btrfs_run_delayed_refs(trans, tree_root, + update); } } + btrfs_release_path(root, path); + BUG_ON(err); + + ret = btrfs_del_root(trans, tree_root, &root->root_key); + BUG_ON(ret); + + free_extent_buffer(root->node); + free_extent_buffer(root->commit_root); + kfree(root); out: + btrfs_end_transaction(trans, tree_root); + kfree(wc); btrfs_free_path(path); - return ret; + return err; } +/* + * drop subtree rooted at tree block 'node'. + * + * NOTE: this function will unlock and release tree block 'node' + */ int btrfs_drop_subtree(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *node, struct extent_buffer *parent) { struct btrfs_path *path; + struct walk_control *wc; int level; int parent_level; int ret = 0; int wret; + BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); + path = btrfs_alloc_path(); BUG_ON(!path); - BUG_ON(!btrfs_tree_locked(parent)); + wc = kzalloc(sizeof(*wc), GFP_NOFS); + BUG_ON(!wc); + + btrfs_assert_tree_locked(parent); parent_level = btrfs_header_level(parent); extent_buffer_get(parent); path->nodes[parent_level] = parent; path->slots[parent_level] = btrfs_header_nritems(parent); - BUG_ON(!btrfs_tree_locked(node)); + btrfs_assert_tree_locked(node); level = btrfs_header_level(node); - extent_buffer_get(node); path->nodes[level] = node; path->slots[level] = 0; + path->locks[level] = 1; + + wc->refs[parent_level] = 1; + wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; + wc->level = level; + wc->shared_level = -1; + wc->stage = DROP_REFERENCE; + wc->update_ref = 0; + wc->keep_locks = 1; while (1) { - wret = walk_down_subtree(trans, root, path, &level); - if (wret < 0) + wret = walk_down_tree(trans, root, path, wc); + if (wret < 0) { ret = wret; - if (wret != 0) break; + } - wret = walk_up_tree(trans, root, path, &level, parent_level); + wret = walk_up_tree(trans, root, path, wc, parent_level); if (wret < 0) ret = wret; if (wret != 0) break; } + kfree(wc); btrfs_free_path(path); return ret; } +#if 0 static unsigned long calc_ra(unsigned long start, unsigned long last, unsigned long nr) { return min(last, start + nr - 1); } -static int noinline relocate_inode_pages(struct inode *inode, u64 start, +static noinline int relocate_inode_pages(struct inode *inode, u64 start, u64 len) { u64 page_start; @@ -3289,10 +5352,10 @@ again: } set_page_extent_mapped(page); - btrfs_set_extent_delalloc(inode, page_start, page_end); if (i == first_index) set_extent_bits(io_tree, page_start, page_end, EXTENT_BOUNDARY, GFP_NOFS); + btrfs_set_extent_delalloc(inode, page_start, page_end); set_page_dirty(page); total_dirty++; @@ -3309,7 +5372,7 @@ out_unlock: return ret; } -static int noinline relocate_data_extent(struct inode *reloc_inode, +static noinline int relocate_data_extent(struct inode *reloc_inode, struct btrfs_key *extent_key, u64 offset) { @@ -3379,12 +5442,13 @@ static int is_cowonly_root(u64 root_objectid) root_objectid == BTRFS_EXTENT_TREE_OBJECTID || root_objectid == BTRFS_CHUNK_TREE_OBJECTID || root_objectid == BTRFS_DEV_TREE_OBJECTID || - root_objectid == BTRFS_TREE_LOG_OBJECTID) + root_objectid == BTRFS_TREE_LOG_OBJECTID || + root_objectid == BTRFS_CSUM_TREE_OBJECTID) return 1; return 0; } -static int noinline __next_ref_path(struct btrfs_trans_handle *trans, +static noinline int __next_ref_path(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root, struct btrfs_ref_path *ref_path, int first_time) @@ -3416,11 +5480,10 @@ walk_down: if (level < ref_path->lowest_level) break; - if (level >= 0) { + if (level >= 0) bytenr = ref_path->nodes[level]; - } else { + else bytenr = ref_path->extent_start; - } BUG_ON(bytenr == 0); parent = ref_path->nodes[level + 1]; @@ -3467,11 +5530,12 @@ walk_up: level = ref_path->current_level; while (level < BTRFS_MAX_LEVEL - 1) { u64 ref_objectid; - if (level >= 0) { + + if (level >= 0) bytenr = ref_path->nodes[level]; - } else { + else bytenr = ref_path->extent_start; - } + BUG_ON(bytenr == 0); key.objectid = bytenr; @@ -3596,7 +5660,7 @@ static int btrfs_next_ref_path(struct btrfs_trans_handle *trans, return __next_ref_path(trans, extent_root, ref_path, 0); } -static int noinline get_new_locations(struct inode *reloc_inode, +static noinline int get_new_locations(struct inode *reloc_inode, struct btrfs_key *extent_key, u64 offset, int no_fragment, struct disk_extent **extents, @@ -3699,7 +5763,7 @@ static int noinline get_new_locations(struct inode *reloc_inode, path->slots[0]++; } - WARN_ON(cur_pos + offset > last_byte); + BUG_ON(cur_pos + offset > last_byte); if (cur_pos + offset < last_byte) { ret = -ENOENT; goto out; @@ -3717,7 +5781,7 @@ out: return ret; } -static int noinline replace_one_extent(struct btrfs_trans_handle *trans, +static noinline int replace_one_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *extent_key, @@ -3734,7 +5798,7 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans, u64 lock_end = 0; u64 num_bytes; u64 ext_offset; - u64 first_pos; + u64 search_end = (u64)-1; u32 nritems; int nr_scaned = 0; int extent_locked = 0; @@ -3742,7 +5806,6 @@ static int noinline replace_one_extent(struct btrfs_trans_handle *trans, int ret; memcpy(&key, leaf_key, sizeof(key)); - first_pos = INT_LIMIT(loff_t) - extent_key->offset; if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) { if (key.objectid < ref_path->owner_objectid || (key.objectid == ref_path->owner_objectid && @@ -3791,7 +5854,7 @@ next: if ((key.objectid > ref_path->owner_objectid) || (key.objectid == ref_path->owner_objectid && key.type > BTRFS_EXTENT_DATA_KEY) || - (key.offset >= first_pos + extent_key->offset)) + key.offset >= search_end) break; } @@ -3824,8 +5887,10 @@ next: num_bytes = btrfs_file_extent_num_bytes(leaf, fi); ext_offset = btrfs_file_extent_offset(leaf, fi); - if (first_pos > key.offset - ext_offset) - first_pos = key.offset - ext_offset; + if (search_end == (u64)-1) { + search_end = key.offset - ext_offset + + btrfs_file_extent_ram_bytes(leaf, fi); + } if (!extent_locked) { lock_start = key.offset; @@ -4014,7 +6079,7 @@ next: } skip: if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS && - key.offset >= first_pos + extent_key->offset) + key.offset >= search_end) break; cond_resched(); @@ -4068,6 +6133,7 @@ int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, ref->bytenr = buf->start; ref->owner = btrfs_header_owner(buf); ref->generation = btrfs_header_generation(buf); + ret = btrfs_add_leaf_ref(root, ref, 0); WARN_ON(ret); btrfs_free_leaf_ref(root, ref); @@ -4075,7 +6141,7 @@ int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans, return 0; } -static int noinline invalidate_extent_cache(struct btrfs_root *root, +static noinline int invalidate_extent_cache(struct btrfs_root *root, struct extent_buffer *leaf, struct btrfs_block_group_cache *group, struct btrfs_root *target_root) @@ -4123,7 +6189,7 @@ static int noinline invalidate_extent_cache(struct btrfs_root *root, return 0; } -static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, +static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct extent_buffer *leaf, struct btrfs_block_group_cache *group, @@ -4197,6 +6263,7 @@ static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans, root->root_key.objectid, trans->transid, key.objectid); BUG_ON(ret); + ret = btrfs_free_extent(trans, root, bytenr, num_bytes, leaf->start, btrfs_header_owner(leaf), @@ -4332,7 +6399,7 @@ int btrfs_cleanup_reloc_trees(struct btrfs_root *root) return 0; } -static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, +static noinline int init_reloc_tree(struct btrfs_trans_handle *trans, struct btrfs_root *root) { struct btrfs_root *reloc_root; @@ -4399,7 +6466,7 @@ static int noinline init_reloc_tree(struct btrfs_trans_handle *trans, * tree blocks are shared between reloc trees, so they are also shared * between subvols. */ -static int noinline relocate_one_path(struct btrfs_trans_handle *trans, +static noinline int relocate_one_path(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *first_key, @@ -4458,7 +6525,8 @@ static int noinline relocate_one_path(struct btrfs_trans_handle *trans, else btrfs_node_key_to_cpu(eb, &keys[level], 0); } - if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + if (nodes[0] && + ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { eb = path->nodes[0]; ret = replace_extents_in_leaf(trans, reloc_root, eb, group, reloc_inode); @@ -4495,7 +6563,7 @@ static int noinline relocate_one_path(struct btrfs_trans_handle *trans, return 0; } -static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, +static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, struct btrfs_key *first_key, @@ -4507,13 +6575,10 @@ static int noinline relocate_tree_block(struct btrfs_trans_handle *trans, ref_path, NULL, NULL); BUG_ON(ret); - if (root == root->fs_info->extent_root) - btrfs_extent_post_op(trans, root); - return 0; } -static int noinline del_extent_zero(struct btrfs_trans_handle *trans, +static noinline int del_extent_zero(struct btrfs_trans_handle *trans, struct btrfs_root *extent_root, struct btrfs_path *path, struct btrfs_key *extent_key) @@ -4529,7 +6594,7 @@ out: return ret; } -static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, +static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info, struct btrfs_ref_path *ref_path) { struct btrfs_key root_key; @@ -4544,7 +6609,7 @@ static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info, return btrfs_read_fs_root_no_name(fs_info, &root_key); } -static int noinline relocate_one_extent(struct btrfs_root *extent_root, +static noinline int relocate_one_extent(struct btrfs_root *extent_root, struct btrfs_path *path, struct btrfs_key *extent_key, struct btrfs_block_group_cache *group, @@ -4572,8 +6637,8 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS); if (!ref_path) { - ret = -ENOMEM; - goto out; + ret = -ENOMEM; + goto out; } for (loops = 0; ; loops++) { @@ -4640,8 +6705,22 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, prev_block = block_start; } - if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && - pass >= 2) { + mutex_lock(&extent_root->fs_info->trans_mutex); + btrfs_record_root_in_trans(found_root); + mutex_unlock(&extent_root->fs_info->trans_mutex); + if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { + /* + * try to update data extent references while + * keeping metadata shared between snapshots. + */ + if (pass == 1) { + ret = relocate_one_path(trans, found_root, + path, &first_key, ref_path, + group, reloc_inode); + if (ret < 0) + goto out; + continue; + } /* * use fallback method to process the remaining * references. @@ -4659,28 +6738,13 @@ static int noinline relocate_one_extent(struct btrfs_root *extent_root, if (ret) goto out; } - btrfs_record_root_in_trans(found_root); ret = replace_one_extent(trans, found_root, path, extent_key, &first_key, ref_path, new_extents, nr_extents); - if (ret < 0) - goto out; - continue; - } - - btrfs_record_root_in_trans(found_root); - if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { + } else { ret = relocate_tree_block(trans, found_root, path, &first_key, ref_path); - } else { - /* - * try to update data extent references while - * keeping metadata shared between snapshots. - */ - ret = relocate_one_path(trans, found_root, path, - &first_key, ref_path, - group, reloc_inode); } if (ret < 0) goto out; @@ -4692,6 +6756,7 @@ out: kfree(ref_path); return ret; } +#endif static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) { @@ -4699,7 +6764,7 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; - num_devices = root->fs_info->fs_devices->num_devices; + num_devices = root->fs_info->fs_devices->rw_devices; if (num_devices == 1) { stripped |= BTRFS_BLOCK_GROUP_DUP; stripped = flags & ~stripped; @@ -4731,7 +6796,7 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) return flags; } -int __alloc_chunk_for_shrink(struct btrfs_root *root, +static int __alloc_chunk_for_shrink(struct btrfs_root *root, struct btrfs_block_group_cache *shrink_block_group, int force) { @@ -4740,7 +6805,8 @@ int __alloc_chunk_for_shrink(struct btrfs_root *root, u64 calc; spin_lock(&shrink_block_group->lock); - if (btrfs_block_group_used(&shrink_block_group->item) > 0) { + if (btrfs_block_group_used(&shrink_block_group->item) + + shrink_block_group->reserved > 0) { spin_unlock(&shrink_block_group->lock); trans = btrfs_start_transaction(root, 1); @@ -4765,6 +6831,17 @@ int __alloc_chunk_for_shrink(struct btrfs_root *root, return 0; } + +int btrfs_prepare_block_group_relocation(struct btrfs_root *root, + struct btrfs_block_group_cache *group) + +{ + __alloc_chunk_for_shrink(root, group, 1); + set_block_group_readonly(group); + return 0; +} + +#if 0 static int __insert_orphan_inode(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 objectid, u64 size) @@ -4778,6 +6855,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; ret = btrfs_insert_empty_inode(trans, root, path, objectid); if (ret) goto out; @@ -4788,8 +6866,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 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_NODATASUM | - BTRFS_INODE_NOCOMPRESS); + btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS); btrfs_mark_buffer_dirty(leaf); btrfs_release_path(root, path); out: @@ -4797,7 +6874,7 @@ out: return ret; } -static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, +static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, struct btrfs_block_group_cache *group) { struct inode *inode = NULL; @@ -4841,6 +6918,7 @@ static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info, } else { BUG_ON(1); } + BTRFS_I(inode)->index_cnt = group->key.objectid; err = btrfs_orphan_add(trans, inode); out: @@ -4853,6 +6931,47 @@ out: 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); + + sector_sum = sums->sums; + sums->bytenr = ordered->start; + + 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); + } + btrfs_put_ordered_extent(ordered); + return 0; +} + int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) { struct btrfs_trans_handle *trans; @@ -4875,7 +6994,7 @@ int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) block_group = btrfs_lookup_block_group(info, group_start); BUG_ON(!block_group); - printk("btrfs relocating block group %llu flags %llu\n", + printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n", (unsigned long long)block_group->key.objectid, (unsigned long long)block_group->flags); @@ -4886,8 +7005,7 @@ int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start) BUG_ON(IS_ERR(reloc_inode)); __alloc_chunk_for_shrink(root, block_group, 1); - block_group->ro = 1; - block_group->space_info->total_bytes -= block_group->key.offset; + set_block_group_readonly(block_group); btrfs_start_delalloc_inodes(info->tree_root); btrfs_wait_ordered_extents(info->tree_root, 0); @@ -4908,7 +7026,10 @@ again: btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); mutex_unlock(&root->fs_info->cleaner_mutex); - while(1) { + trans = btrfs_start_transaction(info->tree_root, 1); + btrfs_commit_transaction(trans, info->tree_root); + + while (1) { ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) goto out; @@ -4968,11 +7089,10 @@ next: if (pass == 0) { btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1); invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1); - WARN_ON(reloc_inode->i_mapping->nrpages); } if (total_found > 0) { - printk("btrfs found %llu extents in pass %d\n", + printk(KERN_INFO "btrfs found %llu extents in pass %d\n", (unsigned long long)total_found, pass); pass++; if (total_found == skipped && pass > 2) { @@ -4995,14 +7115,16 @@ next: 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; out: btrfs_free_path(path); return ret; } +#endif -int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, - struct btrfs_key *key) +static int find_first_block_group(struct btrfs_root *root, + struct btrfs_path *path, struct btrfs_key *key) { int ret = 0; struct btrfs_key found_key; @@ -5013,7 +7135,7 @@ int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path, if (ret < 0) goto out; - while(1) { + while (1) { slot = path->slots[0]; leaf = path->nodes[0]; if (slot >= btrfs_header_nritems(leaf)) { @@ -5041,6 +7163,7 @@ out: int btrfs_free_block_groups(struct btrfs_fs_info *info) { struct btrfs_block_group_cache *block_group; + struct btrfs_space_info *space_info; struct rb_node *n; spin_lock(&info->block_group_cache_lock); @@ -5051,15 +7174,39 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info) &info->block_group_cache_tree); spin_unlock(&info->block_group_cache_lock); - btrfs_remove_free_space_cache(block_group); down_write(&block_group->space_info->groups_sem); list_del(&block_group->list); up_write(&block_group->space_info->groups_sem); + + if (block_group->cached == BTRFS_CACHE_STARTED) + wait_event(block_group->caching_q, + block_group_cache_done(block_group)); + + btrfs_remove_free_space_cache(block_group); + + WARN_ON(atomic_read(&block_group->count) != 1); kfree(block_group); spin_lock(&info->block_group_cache_lock); } spin_unlock(&info->block_group_cache_lock); + + /* now that all the block groups are freed, go through and + * free all the space_info structs. This is only called during + * the final stages of unmount, and so we know nobody is + * using them. We call synchronize_rcu() once before we start, + * just to be on the safe side. + */ + synchronize_rcu(); + + while(!list_empty(&info->space_info)) { + space_info = list_entry(info->space_info.next, + struct btrfs_space_info, + list); + + list_del(&space_info->list); + kfree(space_info); + } return 0; } @@ -5082,7 +7229,7 @@ int btrfs_read_block_groups(struct btrfs_root *root) if (!path) return -ENOMEM; - while(1) { + while (1) { ret = find_first_block_group(root, path, &key); if (ret > 0) { ret = 0; @@ -5099,9 +7246,22 @@ int btrfs_read_block_groups(struct btrfs_root *root) break; } + atomic_set(&cache->count, 1); spin_lock_init(&cache->lock); - mutex_init(&cache->alloc_mutex); + spin_lock_init(&cache->tree_lock); + cache->fs_info = info; + init_waitqueue_head(&cache->caching_q); INIT_LIST_HEAD(&cache->list); + INIT_LIST_HEAD(&cache->cluster_list); + + /* + * we only want to have 32k of ram per block group for keeping + * track of free space, and if we pass 1/2 of that we want to + * start converting things over to using bitmaps + */ + cache->extents_thresh = ((1024 * 32) / 2) / + sizeof(struct btrfs_free_space); + read_extent_buffer(leaf, &cache->item, btrfs_item_ptr_offset(leaf, path->slots[0]), sizeof(cache->item)); @@ -5110,6 +7270,26 @@ int btrfs_read_block_groups(struct btrfs_root *root) key.objectid = found_key.objectid + found_key.offset; btrfs_release_path(root, path); cache->flags = btrfs_block_group_flags(&cache->item); + cache->sectorsize = root->sectorsize; + + remove_sb_from_cache(root, cache); + + /* + * check for two cases, either we are full, and therefore + * don't need to bother with the caching work since we won't + * find any space, or we are empty, and we can just add all + * the space in and be done with it. This saves us _alot_ of + * time, particularly in the full case. + */ + if (found_key.offset == btrfs_block_group_used(&cache->item)) { + cache->cached = BTRFS_CACHE_FINISHED; + } else if (btrfs_block_group_used(&cache->item) == 0) { + cache->cached = BTRFS_CACHE_FINISHED; + add_new_free_space(cache, root->fs_info, + found_key.objectid, + found_key.objectid + + found_key.offset); + } ret = update_space_info(info, cache->flags, found_key.offset, btrfs_block_group_used(&cache->item), @@ -5124,6 +7304,8 @@ int btrfs_read_block_groups(struct btrfs_root *root) BUG_ON(ret); set_avail_alloc_bits(root->fs_info, cache->flags); + if (btrfs_chunk_readonly(root, cache->key.objectid)) + set_block_group_readonly(cache); } ret = 0; error: @@ -5142,7 +7324,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, extent_root = root->fs_info->extent_root; - root->fs_info->last_trans_new_blockgroup = trans->transid; + root->fs_info->last_trans_log_full_commit = trans->transid; cache = kzalloc(sizeof(*cache), GFP_NOFS); if (!cache) @@ -5150,16 +7332,34 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, cache->key.objectid = chunk_offset; cache->key.offset = size; + cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; + cache->sectorsize = root->sectorsize; + + /* + * we only want to have 32k of ram per block group for keeping track + * of free space, and if we pass 1/2 of that we want to start + * converting things over to using bitmaps + */ + cache->extents_thresh = ((1024 * 32) / 2) / + sizeof(struct btrfs_free_space); + atomic_set(&cache->count, 1); spin_lock_init(&cache->lock); - mutex_init(&cache->alloc_mutex); + spin_lock_init(&cache->tree_lock); + init_waitqueue_head(&cache->caching_q); INIT_LIST_HEAD(&cache->list); - btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY); + INIT_LIST_HEAD(&cache->cluster_list); btrfs_set_block_group_used(&cache->item, bytes_used); btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); cache->flags = type; btrfs_set_block_group_flags(&cache->item, type); + cache->cached = BTRFS_CACHE_FINISHED; + remove_sb_from_cache(root, cache); + + add_new_free_space(cache, root->fs_info, chunk_offset, + chunk_offset + size); + ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, &cache->space_info); BUG_ON(ret); @@ -5174,9 +7374,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, sizeof(cache->item)); BUG_ON(ret); - finish_current_insert(trans, extent_root, 0); - ret = del_pending_extents(trans, extent_root, 0); - BUG_ON(ret); set_avail_alloc_bits(extent_root->fs_info, type); return 0; @@ -5187,6 +7384,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans, { struct btrfs_path *path; struct btrfs_block_group_cache *block_group; + struct btrfs_free_cluster *cluster; struct btrfs_key key; int ret; @@ -5194,23 +7392,56 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans, block_group = btrfs_lookup_block_group(root->fs_info, group_start); BUG_ON(!block_group); + BUG_ON(!block_group->ro); memcpy(&key, &block_group->key, sizeof(key)); + /* make sure this block group isn't part of an allocation cluster */ + cluster = &root->fs_info->data_alloc_cluster; + spin_lock(&cluster->refill_lock); + btrfs_return_cluster_to_free_space(block_group, cluster); + spin_unlock(&cluster->refill_lock); + + /* + * make sure this block group isn't part of a metadata + * allocation cluster + */ + cluster = &root->fs_info->meta_alloc_cluster; + spin_lock(&cluster->refill_lock); + btrfs_return_cluster_to_free_space(block_group, cluster); + spin_unlock(&cluster->refill_lock); + path = btrfs_alloc_path(); BUG_ON(!path); - btrfs_remove_free_space_cache(block_group); + spin_lock(&root->fs_info->block_group_cache_lock); rb_erase(&block_group->cache_node, &root->fs_info->block_group_cache_tree); + spin_unlock(&root->fs_info->block_group_cache_lock); + down_write(&block_group->space_info->groups_sem); - list_del(&block_group->list); + /* + * we must use list_del_init so people can check to see if they + * are still on the list after taking the semaphore + */ + list_del_init(&block_group->list); up_write(&block_group->space_info->groups_sem); - /* - memset(shrink_block_group, 0, sizeof(*shrink_block_group)); - kfree(shrink_block_group); - */ + if (block_group->cached == BTRFS_CACHE_STARTED) + wait_event(block_group->caching_q, + block_group_cache_done(block_group)); + + btrfs_remove_free_space_cache(block_group); + + spin_lock(&block_group->space_info->lock); + block_group->space_info->total_bytes -= block_group->key.offset; + block_group->space_info->bytes_readonly -= block_group->key.offset; + spin_unlock(&block_group->space_info->lock); + + btrfs_clear_space_info_full(root->fs_info); + + btrfs_put_block_group(block_group); + btrfs_put_block_group(block_group); ret = btrfs_search_slot(trans, root, &key, path, -1, 1); if (ret > 0)