#include <linux/sched.h>
#include <linux/writeback.h>
#include <linux/pagemap.h>
+#include <linux/blkdev.h>
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "ref-cache.h"
#include "tree-log.h"
-static int total_trans = 0;
-extern struct kmem_cache *btrfs_trans_handle_cachep;
-extern struct kmem_cache *btrfs_transaction_cachep;
-
#define BTRFS_ROOT_TRANS_TAG 0
static noinline void put_transaction(struct btrfs_transaction *transaction)
WARN_ON(transaction->use_count == 0);
transaction->use_count--;
if (transaction->use_count == 0) {
- WARN_ON(total_trans == 0);
- total_trans--;
list_del_init(&transaction->list);
memset(transaction, 0, sizeof(*transaction));
kmem_cache_free(btrfs_transaction_cachep, transaction);
}
}
+/*
+ * either allocate a new transaction or hop into the existing one
+ */
static noinline int join_transaction(struct btrfs_root *root)
{
struct btrfs_transaction *cur_trans;
if (!cur_trans) {
cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
GFP_NOFS);
- total_trans++;
BUG_ON(!cur_trans);
root->fs_info->generation++;
root->fs_info->last_alloc = 0;
root->fs_info->last_data_alloc = 0;
- root->fs_info->last_log_alloc = 0;
cur_trans->num_writers = 1;
cur_trans->num_joined = 0;
cur_trans->transid = root->fs_info->generation;
return 0;
}
+/*
+ * this does all the record keeping required to make sure that a reference
+ * counted root is properly recorded in a given transaction. This is required
+ * to make sure the old root from before we joined the transaction is deleted
+ * when the transaction commits
+ */
noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
{
struct btrfs_dirty_root *dirty;
spin_lock_init(&dirty->root->node_lock);
spin_lock_init(&dirty->root->list_lock);
mutex_init(&dirty->root->objectid_mutex);
+ mutex_init(&dirty->root->log_mutex);
INIT_LIST_HEAD(&dirty->root->dead_list);
dirty->root->node = root->commit_root;
dirty->root->commit_root = NULL;
return 0;
}
+/* wait for commit against the current transaction to become unblocked
+ * when this is done, it is safe to start a new transaction, but the current
+ * transaction might not be fully on disk.
+ */
static void wait_current_trans(struct btrfs_root *root)
{
struct btrfs_transaction *cur_trans;
if (cur_trans && cur_trans->blocked) {
DEFINE_WAIT(wait);
cur_trans->use_count++;
- while(1) {
+ while (1) {
prepare_to_wait(&root->fs_info->transaction_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (cur_trans->blocked) {
h->transaction = root->fs_info->running_transaction;
h->blocks_reserved = num_blocks;
h->blocks_used = 0;
- h->block_group = NULL;
+ h->block_group = 0;
h->alloc_exclude_nr = 0;
h->alloc_exclude_start = 0;
root->fs_info->running_transaction->use_count++;
return start_transaction(r, num_blocks, 2);
}
-
+/* wait for a transaction commit to be fully complete */
static noinline int wait_for_commit(struct btrfs_root *root,
struct btrfs_transaction *commit)
{
DEFINE_WAIT(wait);
mutex_lock(&root->fs_info->trans_mutex);
- while(!commit->commit_done) {
+ while (!commit->commit_done) {
prepare_to_wait(&commit->commit_wait, &wait,
TASK_UNINTERRUPTIBLE);
if (commit->commit_done)
return 0;
}
+/*
+ * rate limit against the drop_snapshot code. This helps to slow down new
+ * operations if the drop_snapshot code isn't able to keep up.
+ */
static void throttle_on_drops(struct btrfs_root *root)
{
struct btrfs_fs_info *info = root->fs_info;
return __btrfs_end_transaction(trans, root, 1);
}
-
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees. This is used to make sure all of
+ * those extents are on disk for transaction or log commit
+ */
int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
struct extent_io_tree *dirty_pages)
{
u64 end;
unsigned long index;
- while(1) {
+ while (1) {
ret = find_first_extent_bit(dirty_pages, start, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
- while(start <= end) {
+ while (start <= end) {
cond_resched();
index = start >> PAGE_CACHE_SHIFT;
page_cache_release(page);
}
}
- while(1) {
+ while (1) {
ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
EXTENT_DIRTY);
if (ret)
break;
clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
- while(start <= end) {
+ while (start <= end) {
index = start >> PAGE_CACHE_SHIFT;
start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
page = find_get_page(btree_inode->i_mapping, index);
&trans->transaction->dirty_pages);
}
+/*
+ * this is used to update the root pointer in the tree of tree roots.
+ *
+ * But, in the case of the extent allocation tree, updating the root
+ * pointer may allocate blocks which may change the root of the extent
+ * allocation tree.
+ *
+ * So, this loops and repeats and makes sure the cowonly root didn't
+ * change while the root pointer was being updated in the metadata.
+ */
static int update_cowonly_root(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
u64 old_root_bytenr;
struct btrfs_root *tree_root = root->fs_info->tree_root;
+ btrfs_extent_post_op(trans, root);
btrfs_write_dirty_block_groups(trans, root);
- while(1) {
+ btrfs_extent_post_op(trans, root);
+
+ while (1) {
old_root_bytenr = btrfs_root_bytenr(&root->root_item);
if (old_root_bytenr == root->node->start)
break;
root->node->start);
btrfs_set_root_level(&root->root_item,
btrfs_header_level(root->node));
+ btrfs_set_root_generation(&root->root_item, trans->transid);
+
+ btrfs_extent_post_op(trans, root);
+
ret = btrfs_update_root(trans, tree_root,
&root->root_key,
&root->root_item);
BUG_ON(ret);
btrfs_write_dirty_block_groups(trans, root);
+ btrfs_extent_post_op(trans, root);
}
return 0;
}
+/*
+ * update all the cowonly tree roots on disk
+ */
int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
struct btrfs_fs_info *fs_info = root->fs_info;
struct list_head *next;
+ struct extent_buffer *eb;
- while(!list_empty(&fs_info->dirty_cowonly_roots)) {
+ btrfs_extent_post_op(trans, fs_info->tree_root);
+
+ eb = btrfs_lock_root_node(fs_info->tree_root);
+ btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0);
+ btrfs_tree_unlock(eb);
+ free_extent_buffer(eb);
+
+ btrfs_extent_post_op(trans, fs_info->tree_root);
+
+ while (!list_empty(&fs_info->dirty_cowonly_roots)) {
next = fs_info->dirty_cowonly_roots.next;
list_del_init(next);
root = list_entry(next, struct btrfs_root, dirty_list);
+
update_cowonly_root(trans, root);
}
return 0;
}
+/*
+ * dead roots are old snapshots that need to be deleted. This allocates
+ * a dirty root struct and adds it into the list of dead roots that need to
+ * be deleted
+ */
int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest)
{
struct btrfs_dirty_root *dirty;
return 0;
}
+/*
+ * at transaction commit time we need to schedule the old roots for
+ * deletion via btrfs_drop_snapshot. This runs through all the
+ * reference counted roots that were modified in the current
+ * transaction and puts them into the drop list
+ */
static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
struct radix_tree_root *radix,
struct list_head *list)
int err = 0;
u32 refs;
- while(1) {
+ while (1) {
ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
ARRAY_SIZE(gang),
BTRFS_ROOT_TRANS_TAG);
dirty = root->dirty_root;
btrfs_free_log(trans, root);
+ btrfs_free_reloc_root(trans, root);
if (root->commit_root == root->node) {
WARN_ON(root->node->start !=
root->node->start);
btrfs_set_root_level(&root->root_item,
btrfs_header_level(root->node));
+ btrfs_set_root_generation(&root->root_item,
+ root->root_key.offset);
+
err = btrfs_insert_root(trans, root->fs_info->tree_root,
&root->root_key,
&root->root_item);
return err;
}
+/*
+ * defrag a given btree. If cacheonly == 1, this won't read from the disk,
+ * otherwise every leaf in the btree is read and defragged.
+ */
int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
{
struct btrfs_fs_info *info = root->fs_info;
return 0;
}
+/*
+ * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
+ * all of them
+ */
static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
struct list_head *list)
{
int ret = 0;
int err;
- while(!list_empty(list)) {
+ while (!list_empty(list)) {
struct btrfs_root *root;
dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
root = dirty->latest_root;
atomic_inc(&root->fs_info->throttles);
- mutex_lock(&root->fs_info->drop_mutex);
- while(1) {
+ while (1) {
trans = btrfs_start_transaction(tree_root, 1);
+ mutex_lock(&root->fs_info->drop_mutex);
ret = btrfs_drop_snapshot(trans, dirty->root);
- if (ret != -EAGAIN) {
+ if (ret != -EAGAIN)
break;
- }
+ mutex_unlock(&root->fs_info->drop_mutex);
err = btrfs_update_root(trans,
tree_root,
ret = btrfs_end_transaction(trans, tree_root);
BUG_ON(ret);
- mutex_unlock(&root->fs_info->drop_mutex);
btrfs_btree_balance_dirty(tree_root, nr);
cond_resched();
- mutex_lock(&root->fs_info->drop_mutex);
}
BUG_ON(ret);
atomic_dec(&root->fs_info->throttles);
wake_up(&root->fs_info->transaction_throttle);
- mutex_lock(&root->fs_info->alloc_mutex);
num_bytes -= btrfs_root_used(&dirty->root->root_item);
bytes_used = btrfs_root_used(&root->root_item);
if (num_bytes) {
btrfs_set_root_used(&root->root_item,
bytes_used - num_bytes);
}
- mutex_unlock(&root->fs_info->alloc_mutex);
ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
if (ret) {
ret = btrfs_end_transaction(trans, tree_root);
BUG_ON(ret);
- ret = btrfs_remove_leaf_refs(root, max_useless);
+ ret = btrfs_remove_leaf_refs(root, max_useless, 0);
BUG_ON(ret);
free_extent_buffer(dirty->root->node);
return ret;
}
+/*
+ * new snapshots need to be created at a very specific time in the
+ * transaction commit. This does the actual creation
+ */
static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info,
struct btrfs_pending_snapshot *pending)
struct extent_buffer *tmp;
struct extent_buffer *old;
int ret;
- int namelen;
u64 objectid;
new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
if (ret)
goto fail;
+ btrfs_record_root_in_trans(root);
+ btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
key.objectid = objectid;
- key.offset = 1;
+ key.offset = trans->transid;
btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
old = btrfs_lock_root_node(root);
btrfs_set_root_bytenr(new_root_item, tmp->start);
btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
+ btrfs_set_root_generation(new_root_item, trans->transid);
ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
new_root_item);
btrfs_tree_unlock(tmp);
if (ret)
goto fail;
+ key.offset = (u64)-1;
+ memcpy(&pending->root_key, &key, sizeof(key));
+fail:
+ kfree(new_root_item);
+ return ret;
+}
+
+static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
+ struct btrfs_pending_snapshot *pending)
+{
+ int ret;
+ int namelen;
+ u64 index = 0;
+ struct btrfs_trans_handle *trans;
+ struct inode *parent_inode;
+ struct inode *inode;
+ struct btrfs_root *parent_root;
+
+ parent_inode = pending->dentry->d_parent->d_inode;
+ parent_root = BTRFS_I(parent_inode)->root;
+ trans = btrfs_start_transaction(parent_root, 1);
+
/*
* insert the directory item
*/
- key.offset = (u64)-1;
namelen = strlen(pending->name);
- ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
- pending->name, namelen,
- root->fs_info->sb->s_root->d_inode->i_ino,
- &key, BTRFS_FT_DIR, 0);
+ ret = btrfs_set_inode_index(parent_inode, &index);
+ ret = btrfs_insert_dir_item(trans, parent_root,
+ pending->name, namelen,
+ parent_inode->i_ino,
+ &pending->root_key, BTRFS_FT_DIR, index);
if (ret)
goto fail;
- ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
- pending->name, strlen(pending->name), objectid,
- root->fs_info->sb->s_root->d_inode->i_ino, 0);
+ btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+ ret = btrfs_update_inode(trans, parent_root, parent_inode);
+ BUG_ON(ret);
+
+ /* add the backref first */
+ ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+ pending->root_key.objectid,
+ BTRFS_ROOT_BACKREF_KEY,
+ parent_root->root_key.objectid,
+ parent_inode->i_ino, index, pending->name,
+ namelen);
- /* Invalidate existing dcache entry for new snapshot. */
- btrfs_invalidate_dcache_root(root, pending->name, namelen);
+ BUG_ON(ret);
+
+ /* now add the forward ref */
+ ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+ parent_root->root_key.objectid,
+ BTRFS_ROOT_REF_KEY,
+ pending->root_key.objectid,
+ parent_inode->i_ino, index, pending->name,
+ namelen);
+ inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
+ d_instantiate(pending->dentry, inode);
fail:
- kfree(new_root_item);
+ btrfs_end_transaction(trans, fs_info->fs_root);
return ret;
}
+/*
+ * create all the snapshots we've scheduled for creation
+ */
static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
struct btrfs_fs_info *fs_info)
+{
+ struct btrfs_pending_snapshot *pending;
+ struct list_head *head = &trans->transaction->pending_snapshots;
+ struct list_head *cur;
+ int ret;
+
+ list_for_each(cur, head) {
+ pending = list_entry(cur, struct btrfs_pending_snapshot, list);
+ ret = create_pending_snapshot(trans, fs_info, pending);
+ BUG_ON(ret);
+ }
+ return 0;
+}
+
+static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
+ struct btrfs_fs_info *fs_info)
{
struct btrfs_pending_snapshot *pending;
struct list_head *head = &trans->transaction->pending_snapshots;
int ret;
- while(!list_empty(head)) {
+ while (!list_empty(head)) {
pending = list_entry(head->next,
struct btrfs_pending_snapshot, list);
- ret = create_pending_snapshot(trans, fs_info, pending);
+ ret = finish_pending_snapshot(fs_info, pending);
BUG_ON(ret);
list_del(&pending->list);
kfree(pending->name);
* with the tree-log code.
*/
mutex_lock(&root->fs_info->tree_log_mutex);
+ /*
+ * keep tree reloc code from adding new reloc trees
+ */
+ mutex_lock(&root->fs_info->tree_reloc_mutex);
+
ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
&dirty_fs_roots);
chunk_root->node->start);
btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
btrfs_header_level(chunk_root->node));
+ btrfs_set_super_chunk_root_generation(&root->fs_info->super_copy,
+ btrfs_header_generation(chunk_root->node));
if (!root->fs_info->log_root_recovering) {
btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
mutex_unlock(&root->fs_info->trans_mutex);
ret = btrfs_write_and_wait_transaction(trans, root);
BUG_ON(ret);
- write_ctree_super(trans, root);
+ write_ctree_super(trans, root, 0);
/*
* the super is written, we can safely allow the tree-loggers
mutex_unlock(&root->fs_info->tree_log_mutex);
btrfs_finish_extent_commit(trans, root, pinned_copy);
- mutex_lock(&root->fs_info->trans_mutex);
-
kfree(pinned_copy);
+ btrfs_drop_dead_reloc_roots(root);
+ mutex_unlock(&root->fs_info->tree_reloc_mutex);
+
+ /* do the directory inserts of any pending snapshot creations */
+ finish_pending_snapshots(trans, root->fs_info);
+
+ mutex_lock(&root->fs_info->trans_mutex);
+
cur_trans->commit_done = 1;
root->fs_info->last_trans_committed = cur_trans->transid;
wake_up(&cur_trans->commit_wait);
+
put_transaction(cur_trans);
put_transaction(cur_trans);
list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
mutex_unlock(&root->fs_info->trans_mutex);
+
kmem_cache_free(btrfs_trans_handle_cachep, trans);
- if (root->fs_info->closing) {
+ if (root->fs_info->closing)
drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
- }
return ret;
}
+/*
+ * interface function to delete all the snapshots we have scheduled for deletion
+ */
int btrfs_clean_old_snapshots(struct btrfs_root *root)
{
struct list_head dirty_roots;