diff options
-rw-r--r-- | fs/btrfs/ctree.h | 13 | ||||
-rw-r--r-- | fs/btrfs/disk-io.c | 79 | ||||
-rw-r--r-- | fs/btrfs/disk-io.h | 2 | ||||
-rw-r--r-- | fs/btrfs/extent-tree.c | 10 | ||||
-rw-r--r-- | fs/btrfs/file.c | 4 | ||||
-rw-r--r-- | fs/btrfs/tree-log.c | 350 |
6 files changed, 248 insertions, 210 deletions
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index e1fec636f37f..de103a8a815e 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h | |||
@@ -695,9 +695,7 @@ struct btrfs_fs_info { | |||
695 | struct btrfs_transaction *running_transaction; | 695 | struct btrfs_transaction *running_transaction; |
696 | wait_queue_head_t transaction_throttle; | 696 | wait_queue_head_t transaction_throttle; |
697 | wait_queue_head_t transaction_wait; | 697 | wait_queue_head_t transaction_wait; |
698 | |||
699 | wait_queue_head_t async_submit_wait; | 698 | wait_queue_head_t async_submit_wait; |
700 | wait_queue_head_t tree_log_wait; | ||
701 | 699 | ||
702 | struct btrfs_super_block super_copy; | 700 | struct btrfs_super_block super_copy; |
703 | struct btrfs_super_block super_for_commit; | 701 | struct btrfs_super_block super_for_commit; |
@@ -724,10 +722,6 @@ struct btrfs_fs_info { | |||
724 | atomic_t async_submit_draining; | 722 | atomic_t async_submit_draining; |
725 | atomic_t nr_async_bios; | 723 | atomic_t nr_async_bios; |
726 | atomic_t async_delalloc_pages; | 724 | atomic_t async_delalloc_pages; |
727 | atomic_t tree_log_writers; | ||
728 | atomic_t tree_log_commit; | ||
729 | unsigned long tree_log_batch; | ||
730 | u64 tree_log_transid; | ||
731 | 725 | ||
732 | /* | 726 | /* |
733 | * this is used by the balancing code to wait for all the pending | 727 | * this is used by the balancing code to wait for all the pending |
@@ -827,7 +821,14 @@ struct btrfs_root { | |||
827 | struct kobject root_kobj; | 821 | struct kobject root_kobj; |
828 | struct completion kobj_unregister; | 822 | struct completion kobj_unregister; |
829 | struct mutex objectid_mutex; | 823 | struct mutex objectid_mutex; |
824 | |||
830 | struct mutex log_mutex; | 825 | struct mutex log_mutex; |
826 | wait_queue_head_t log_writer_wait; | ||
827 | wait_queue_head_t log_commit_wait[2]; | ||
828 | atomic_t log_writers; | ||
829 | atomic_t log_commit[2]; | ||
830 | unsigned long log_transid; | ||
831 | unsigned long log_batch; | ||
831 | 832 | ||
832 | u64 objectid; | 833 | u64 objectid; |
833 | u64 last_trans; | 834 | u64 last_trans; |
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c index 3cf17257f89d..7feac5a475e9 100644 --- a/fs/btrfs/disk-io.c +++ b/fs/btrfs/disk-io.c | |||
@@ -849,6 +849,14 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, | |||
849 | spin_lock_init(&root->list_lock); | 849 | spin_lock_init(&root->list_lock); |
850 | mutex_init(&root->objectid_mutex); | 850 | mutex_init(&root->objectid_mutex); |
851 | mutex_init(&root->log_mutex); | 851 | mutex_init(&root->log_mutex); |
852 | init_waitqueue_head(&root->log_writer_wait); | ||
853 | init_waitqueue_head(&root->log_commit_wait[0]); | ||
854 | init_waitqueue_head(&root->log_commit_wait[1]); | ||
855 | atomic_set(&root->log_commit[0], 0); | ||
856 | atomic_set(&root->log_commit[1], 0); | ||
857 | atomic_set(&root->log_writers, 0); | ||
858 | root->log_batch = 0; | ||
859 | root->log_transid = 0; | ||
852 | extent_io_tree_init(&root->dirty_log_pages, | 860 | extent_io_tree_init(&root->dirty_log_pages, |
853 | fs_info->btree_inode->i_mapping, GFP_NOFS); | 861 | fs_info->btree_inode->i_mapping, GFP_NOFS); |
854 | 862 | ||
@@ -933,15 +941,16 @@ int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |||
933 | return 0; | 941 | return 0; |
934 | } | 942 | } |
935 | 943 | ||
936 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | 944 | static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, |
937 | struct btrfs_fs_info *fs_info) | 945 | struct btrfs_fs_info *fs_info) |
938 | { | 946 | { |
939 | struct btrfs_root *root; | 947 | struct btrfs_root *root; |
940 | struct btrfs_root *tree_root = fs_info->tree_root; | 948 | struct btrfs_root *tree_root = fs_info->tree_root; |
949 | struct extent_buffer *leaf; | ||
941 | 950 | ||
942 | root = kzalloc(sizeof(*root), GFP_NOFS); | 951 | root = kzalloc(sizeof(*root), GFP_NOFS); |
943 | if (!root) | 952 | if (!root) |
944 | return -ENOMEM; | 953 | return ERR_PTR(-ENOMEM); |
945 | 954 | ||
946 | __setup_root(tree_root->nodesize, tree_root->leafsize, | 955 | __setup_root(tree_root->nodesize, tree_root->leafsize, |
947 | tree_root->sectorsize, tree_root->stripesize, | 956 | tree_root->sectorsize, tree_root->stripesize, |
@@ -950,12 +959,23 @@ int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | |||
950 | root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; | 959 | root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; |
951 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; | 960 | root->root_key.type = BTRFS_ROOT_ITEM_KEY; |
952 | root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; | 961 | root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; |
962 | /* | ||
963 | * log trees do not get reference counted because they go away | ||
964 | * before a real commit is actually done. They do store pointers | ||
965 | * to file data extents, and those reference counts still get | ||
966 | * updated (along with back refs to the log tree). | ||
967 | */ | ||
953 | root->ref_cows = 0; | 968 | root->ref_cows = 0; |
954 | 969 | ||
955 | root->node = btrfs_alloc_free_block(trans, root, root->leafsize, | 970 | leaf = btrfs_alloc_free_block(trans, root, root->leafsize, |
956 | 0, BTRFS_TREE_LOG_OBJECTID, | 971 | 0, BTRFS_TREE_LOG_OBJECTID, |
957 | trans->transid, 0, 0, 0); | 972 | trans->transid, 0, 0, 0); |
973 | if (IS_ERR(leaf)) { | ||
974 | kfree(root); | ||
975 | return ERR_CAST(leaf); | ||
976 | } | ||
958 | 977 | ||
978 | root->node = leaf; | ||
959 | btrfs_set_header_nritems(root->node, 0); | 979 | btrfs_set_header_nritems(root->node, 0); |
960 | btrfs_set_header_level(root->node, 0); | 980 | btrfs_set_header_level(root->node, 0); |
961 | btrfs_set_header_bytenr(root->node, root->node->start); | 981 | btrfs_set_header_bytenr(root->node, root->node->start); |
@@ -967,7 +987,48 @@ int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | |||
967 | BTRFS_FSID_SIZE); | 987 | BTRFS_FSID_SIZE); |
968 | btrfs_mark_buffer_dirty(root->node); | 988 | btrfs_mark_buffer_dirty(root->node); |
969 | btrfs_tree_unlock(root->node); | 989 | btrfs_tree_unlock(root->node); |
970 | fs_info->log_root_tree = root; | 990 | return root; |
991 | } | ||
992 | |||
993 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | ||
994 | struct btrfs_fs_info *fs_info) | ||
995 | { | ||
996 | struct btrfs_root *log_root; | ||
997 | |||
998 | log_root = alloc_log_tree(trans, fs_info); | ||
999 | if (IS_ERR(log_root)) | ||
1000 | return PTR_ERR(log_root); | ||
1001 | WARN_ON(fs_info->log_root_tree); | ||
1002 | fs_info->log_root_tree = log_root; | ||
1003 | return 0; | ||
1004 | } | ||
1005 | |||
1006 | int btrfs_add_log_tree(struct btrfs_trans_handle *trans, | ||
1007 | struct btrfs_root *root) | ||
1008 | { | ||
1009 | struct btrfs_root *log_root; | ||
1010 | struct btrfs_inode_item *inode_item; | ||
1011 | |||
1012 | log_root = alloc_log_tree(trans, root->fs_info); | ||
1013 | if (IS_ERR(log_root)) | ||
1014 | return PTR_ERR(log_root); | ||
1015 | |||
1016 | log_root->last_trans = trans->transid; | ||
1017 | log_root->root_key.offset = root->root_key.objectid; | ||
1018 | |||
1019 | inode_item = &log_root->root_item.inode; | ||
1020 | inode_item->generation = cpu_to_le64(1); | ||
1021 | inode_item->size = cpu_to_le64(3); | ||
1022 | inode_item->nlink = cpu_to_le32(1); | ||
1023 | inode_item->nbytes = cpu_to_le64(root->leafsize); | ||
1024 | inode_item->mode = cpu_to_le32(S_IFDIR | 0755); | ||
1025 | |||
1026 | btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start); | ||
1027 | btrfs_set_root_generation(&log_root->root_item, trans->transid); | ||
1028 | |||
1029 | WARN_ON(root->log_root); | ||
1030 | root->log_root = log_root; | ||
1031 | root->log_transid = 0; | ||
971 | return 0; | 1032 | return 0; |
972 | } | 1033 | } |
973 | 1034 | ||
@@ -1530,10 +1591,6 @@ struct btrfs_root *open_ctree(struct super_block *sb, | |||
1530 | init_waitqueue_head(&fs_info->transaction_throttle); | 1591 | init_waitqueue_head(&fs_info->transaction_throttle); |
1531 | init_waitqueue_head(&fs_info->transaction_wait); | 1592 | init_waitqueue_head(&fs_info->transaction_wait); |
1532 | init_waitqueue_head(&fs_info->async_submit_wait); | 1593 | init_waitqueue_head(&fs_info->async_submit_wait); |
1533 | init_waitqueue_head(&fs_info->tree_log_wait); | ||
1534 | atomic_set(&fs_info->tree_log_commit, 0); | ||
1535 | atomic_set(&fs_info->tree_log_writers, 0); | ||
1536 | fs_info->tree_log_transid = 0; | ||
1537 | 1594 | ||
1538 | __setup_root(4096, 4096, 4096, 4096, tree_root, | 1595 | __setup_root(4096, 4096, 4096, 4096, tree_root, |
1539 | fs_info, BTRFS_ROOT_TREE_OBJECTID); | 1596 | fs_info, BTRFS_ROOT_TREE_OBJECTID); |
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h index c0ff404c31b7..494a56eb2986 100644 --- a/fs/btrfs/disk-io.h +++ b/fs/btrfs/disk-io.h | |||
@@ -98,5 +98,7 @@ int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans, | |||
98 | struct btrfs_fs_info *fs_info); | 98 | struct btrfs_fs_info *fs_info); |
99 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, | 99 | int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, |
100 | struct btrfs_fs_info *fs_info); | 100 | struct btrfs_fs_info *fs_info); |
101 | int btrfs_add_log_tree(struct btrfs_trans_handle *trans, | ||
102 | struct btrfs_root *root); | ||
101 | int btree_lock_page_hook(struct page *page); | 103 | int btree_lock_page_hook(struct page *page); |
102 | #endif | 104 | #endif |
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 1d7f043152b0..3b26f0980946 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c | |||
@@ -2698,13 +2698,9 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans, | |||
2698 | /* if metadata always pin */ | 2698 | /* if metadata always pin */ |
2699 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { | 2699 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
2700 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { | 2700 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
2701 | struct btrfs_block_group_cache *cache; | 2701 | mutex_lock(&root->fs_info->pinned_mutex); |
2702 | 2702 | btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); | |
2703 | /* btrfs_free_reserved_extent */ | 2703 | mutex_unlock(&root->fs_info->pinned_mutex); |
2704 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | ||
2705 | BUG_ON(!cache); | ||
2706 | btrfs_add_free_space(cache, bytenr, num_bytes); | ||
2707 | put_block_group(cache); | ||
2708 | update_reserved_extents(root, bytenr, num_bytes, 0); | 2704 | update_reserved_extents(root, bytenr, num_bytes, 0); |
2709 | return 0; | 2705 | return 0; |
2710 | } | 2706 | } |
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c index fbcbf43f5114..3e8023efaff7 100644 --- a/fs/btrfs/file.c +++ b/fs/btrfs/file.c | |||
@@ -1214,10 +1214,10 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync) | |||
1214 | } | 1214 | } |
1215 | mutex_unlock(&root->fs_info->trans_mutex); | 1215 | mutex_unlock(&root->fs_info->trans_mutex); |
1216 | 1216 | ||
1217 | root->fs_info->tree_log_batch++; | 1217 | root->log_batch++; |
1218 | filemap_fdatawrite(inode->i_mapping); | 1218 | filemap_fdatawrite(inode->i_mapping); |
1219 | btrfs_wait_ordered_range(inode, 0, (u64)-1); | 1219 | btrfs_wait_ordered_range(inode, 0, (u64)-1); |
1220 | root->fs_info->tree_log_batch++; | 1220 | root->log_batch++; |
1221 | 1221 | ||
1222 | /* | 1222 | /* |
1223 | * ok we haven't committed the transaction yet, lets do a commit | 1223 | * ok we haven't committed the transaction yet, lets do a commit |
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index d81cda2e077c..4f26f3ed0c87 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c | |||
@@ -78,104 +78,6 @@ static int link_to_fixup_dir(struct btrfs_trans_handle *trans, | |||
78 | */ | 78 | */ |
79 | 79 | ||
80 | /* | 80 | /* |
81 | * btrfs_add_log_tree adds a new per-subvolume log tree into the | ||
82 | * tree of log tree roots. This must be called with a tree log transaction | ||
83 | * running (see start_log_trans). | ||
84 | */ | ||
85 | static int btrfs_add_log_tree(struct btrfs_trans_handle *trans, | ||
86 | struct btrfs_root *root) | ||
87 | { | ||
88 | struct btrfs_key key; | ||
89 | struct btrfs_root_item root_item; | ||
90 | struct btrfs_inode_item *inode_item; | ||
91 | struct extent_buffer *leaf; | ||
92 | struct btrfs_root *new_root = root; | ||
93 | int ret; | ||
94 | u64 objectid = root->root_key.objectid; | ||
95 | |||
96 | leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | ||
97 | BTRFS_TREE_LOG_OBJECTID, | ||
98 | trans->transid, 0, 0, 0); | ||
99 | if (IS_ERR(leaf)) { | ||
100 | ret = PTR_ERR(leaf); | ||
101 | return ret; | ||
102 | } | ||
103 | |||
104 | btrfs_set_header_nritems(leaf, 0); | ||
105 | btrfs_set_header_level(leaf, 0); | ||
106 | btrfs_set_header_bytenr(leaf, leaf->start); | ||
107 | btrfs_set_header_generation(leaf, trans->transid); | ||
108 | btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); | ||
109 | |||
110 | write_extent_buffer(leaf, root->fs_info->fsid, | ||
111 | (unsigned long)btrfs_header_fsid(leaf), | ||
112 | BTRFS_FSID_SIZE); | ||
113 | btrfs_mark_buffer_dirty(leaf); | ||
114 | |||
115 | inode_item = &root_item.inode; | ||
116 | memset(inode_item, 0, sizeof(*inode_item)); | ||
117 | inode_item->generation = cpu_to_le64(1); | ||
118 | inode_item->size = cpu_to_le64(3); | ||
119 | inode_item->nlink = cpu_to_le32(1); | ||
120 | inode_item->nbytes = cpu_to_le64(root->leafsize); | ||
121 | inode_item->mode = cpu_to_le32(S_IFDIR | 0755); | ||
122 | |||
123 | btrfs_set_root_bytenr(&root_item, leaf->start); | ||
124 | btrfs_set_root_generation(&root_item, trans->transid); | ||
125 | btrfs_set_root_level(&root_item, 0); | ||
126 | btrfs_set_root_refs(&root_item, 0); | ||
127 | btrfs_set_root_used(&root_item, 0); | ||
128 | |||
129 | memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress)); | ||
130 | root_item.drop_level = 0; | ||
131 | |||
132 | btrfs_tree_unlock(leaf); | ||
133 | free_extent_buffer(leaf); | ||
134 | leaf = NULL; | ||
135 | |||
136 | btrfs_set_root_dirid(&root_item, 0); | ||
137 | |||
138 | key.objectid = BTRFS_TREE_LOG_OBJECTID; | ||
139 | key.offset = objectid; | ||
140 | btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); | ||
141 | ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key, | ||
142 | &root_item); | ||
143 | if (ret) | ||
144 | goto fail; | ||
145 | |||
146 | new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree, | ||
147 | &key); | ||
148 | BUG_ON(!new_root); | ||
149 | |||
150 | WARN_ON(root->log_root); | ||
151 | root->log_root = new_root; | ||
152 | |||
153 | /* | ||
154 | * log trees do not get reference counted because they go away | ||
155 | * before a real commit is actually done. They do store pointers | ||
156 | * to file data extents, and those reference counts still get | ||
157 | * updated (along with back refs to the log tree). | ||
158 | */ | ||
159 | new_root->ref_cows = 0; | ||
160 | new_root->last_trans = trans->transid; | ||
161 | |||
162 | /* | ||
163 | * we need to make sure the root block for this new tree | ||
164 | * is marked as dirty in the dirty_log_pages tree. This | ||
165 | * is how it gets flushed down to disk at tree log commit time. | ||
166 | * | ||
167 | * the tree logging mutex keeps others from coming in and changing | ||
168 | * the new_root->node, so we can safely access it here | ||
169 | */ | ||
170 | set_extent_dirty(&new_root->dirty_log_pages, new_root->node->start, | ||
171 | new_root->node->start + new_root->node->len - 1, | ||
172 | GFP_NOFS); | ||
173 | |||
174 | fail: | ||
175 | return ret; | ||
176 | } | ||
177 | |||
178 | /* | ||
179 | * start a sub transaction and setup the log tree | 81 | * start a sub transaction and setup the log tree |
180 | * this increments the log tree writer count to make the people | 82 | * this increments the log tree writer count to make the people |
181 | * syncing the tree wait for us to finish | 83 | * syncing the tree wait for us to finish |
@@ -184,6 +86,14 @@ static int start_log_trans(struct btrfs_trans_handle *trans, | |||
184 | struct btrfs_root *root) | 86 | struct btrfs_root *root) |
185 | { | 87 | { |
186 | int ret; | 88 | int ret; |
89 | |||
90 | mutex_lock(&root->log_mutex); | ||
91 | if (root->log_root) { | ||
92 | root->log_batch++; | ||
93 | atomic_inc(&root->log_writers); | ||
94 | mutex_unlock(&root->log_mutex); | ||
95 | return 0; | ||
96 | } | ||
187 | mutex_lock(&root->fs_info->tree_log_mutex); | 97 | mutex_lock(&root->fs_info->tree_log_mutex); |
188 | if (!root->fs_info->log_root_tree) { | 98 | if (!root->fs_info->log_root_tree) { |
189 | ret = btrfs_init_log_root_tree(trans, root->fs_info); | 99 | ret = btrfs_init_log_root_tree(trans, root->fs_info); |
@@ -193,9 +103,10 @@ static int start_log_trans(struct btrfs_trans_handle *trans, | |||
193 | ret = btrfs_add_log_tree(trans, root); | 103 | ret = btrfs_add_log_tree(trans, root); |
194 | BUG_ON(ret); | 104 | BUG_ON(ret); |
195 | } | 105 | } |
196 | atomic_inc(&root->fs_info->tree_log_writers); | ||
197 | root->fs_info->tree_log_batch++; | ||
198 | mutex_unlock(&root->fs_info->tree_log_mutex); | 106 | mutex_unlock(&root->fs_info->tree_log_mutex); |
107 | root->log_batch++; | ||
108 | atomic_inc(&root->log_writers); | ||
109 | mutex_unlock(&root->log_mutex); | ||
199 | return 0; | 110 | return 0; |
200 | } | 111 | } |
201 | 112 | ||
@@ -212,13 +123,12 @@ static int join_running_log_trans(struct btrfs_root *root) | |||
212 | if (!root->log_root) | 123 | if (!root->log_root) |
213 | return -ENOENT; | 124 | return -ENOENT; |
214 | 125 | ||
215 | mutex_lock(&root->fs_info->tree_log_mutex); | 126 | mutex_lock(&root->log_mutex); |
216 | if (root->log_root) { | 127 | if (root->log_root) { |
217 | ret = 0; | 128 | ret = 0; |
218 | atomic_inc(&root->fs_info->tree_log_writers); | 129 | atomic_inc(&root->log_writers); |
219 | root->fs_info->tree_log_batch++; | ||
220 | } | 130 | } |
221 | mutex_unlock(&root->fs_info->tree_log_mutex); | 131 | mutex_unlock(&root->log_mutex); |
222 | return ret; | 132 | return ret; |
223 | } | 133 | } |
224 | 134 | ||
@@ -228,10 +138,11 @@ static int join_running_log_trans(struct btrfs_root *root) | |||
228 | */ | 138 | */ |
229 | static int end_log_trans(struct btrfs_root *root) | 139 | static int end_log_trans(struct btrfs_root *root) |
230 | { | 140 | { |
231 | atomic_dec(&root->fs_info->tree_log_writers); | 141 | if (atomic_dec_and_test(&root->log_writers)) { |
232 | smp_mb(); | 142 | smp_mb(); |
233 | if (waitqueue_active(&root->fs_info->tree_log_wait)) | 143 | if (waitqueue_active(&root->log_writer_wait)) |
234 | wake_up(&root->fs_info->tree_log_wait); | 144 | wake_up(&root->log_writer_wait); |
145 | } | ||
235 | return 0; | 146 | return 0; |
236 | } | 147 | } |
237 | 148 | ||
@@ -1902,26 +1813,65 @@ static int walk_log_tree(struct btrfs_trans_handle *trans, | |||
1902 | } | 1813 | } |
1903 | } | 1814 | } |
1904 | btrfs_free_path(path); | 1815 | btrfs_free_path(path); |
1905 | if (wc->free) | ||
1906 | free_extent_buffer(log->node); | ||
1907 | return ret; | 1816 | return ret; |
1908 | } | 1817 | } |
1909 | 1818 | ||
1910 | static int wait_log_commit(struct btrfs_root *log) | 1819 | /* |
1820 | * helper function to update the item for a given subvolumes log root | ||
1821 | * in the tree of log roots | ||
1822 | */ | ||
1823 | static int update_log_root(struct btrfs_trans_handle *trans, | ||
1824 | struct btrfs_root *log) | ||
1825 | { | ||
1826 | int ret; | ||
1827 | |||
1828 | if (log->log_transid == 1) { | ||
1829 | /* insert root item on the first sync */ | ||
1830 | ret = btrfs_insert_root(trans, log->fs_info->log_root_tree, | ||
1831 | &log->root_key, &log->root_item); | ||
1832 | } else { | ||
1833 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | ||
1834 | &log->root_key, &log->root_item); | ||
1835 | } | ||
1836 | return ret; | ||
1837 | } | ||
1838 | |||
1839 | static int wait_log_commit(struct btrfs_root *root, unsigned long transid) | ||
1911 | { | 1840 | { |
1912 | DEFINE_WAIT(wait); | 1841 | DEFINE_WAIT(wait); |
1913 | u64 transid = log->fs_info->tree_log_transid; | 1842 | int index = transid % 2; |
1914 | 1843 | ||
1844 | /* | ||
1845 | * we only allow two pending log transactions at a time, | ||
1846 | * so we know that if ours is more than 2 older than the | ||
1847 | * current transaction, we're done | ||
1848 | */ | ||
1915 | do { | 1849 | do { |
1916 | prepare_to_wait(&log->fs_info->tree_log_wait, &wait, | 1850 | prepare_to_wait(&root->log_commit_wait[index], |
1917 | TASK_UNINTERRUPTIBLE); | 1851 | &wait, TASK_UNINTERRUPTIBLE); |
1918 | mutex_unlock(&log->fs_info->tree_log_mutex); | 1852 | mutex_unlock(&root->log_mutex); |
1919 | if (atomic_read(&log->fs_info->tree_log_commit)) | 1853 | if (root->log_transid < transid + 2 && |
1854 | atomic_read(&root->log_commit[index])) | ||
1920 | schedule(); | 1855 | schedule(); |
1921 | finish_wait(&log->fs_info->tree_log_wait, &wait); | 1856 | finish_wait(&root->log_commit_wait[index], &wait); |
1922 | mutex_lock(&log->fs_info->tree_log_mutex); | 1857 | mutex_lock(&root->log_mutex); |
1923 | } while (transid == log->fs_info->tree_log_transid && | 1858 | } while (root->log_transid < transid + 2 && |
1924 | atomic_read(&log->fs_info->tree_log_commit)); | 1859 | atomic_read(&root->log_commit[index])); |
1860 | return 0; | ||
1861 | } | ||
1862 | |||
1863 | static int wait_for_writer(struct btrfs_root *root) | ||
1864 | { | ||
1865 | DEFINE_WAIT(wait); | ||
1866 | while (atomic_read(&root->log_writers)) { | ||
1867 | prepare_to_wait(&root->log_writer_wait, | ||
1868 | &wait, TASK_UNINTERRUPTIBLE); | ||
1869 | mutex_unlock(&root->log_mutex); | ||
1870 | if (atomic_read(&root->log_writers)) | ||
1871 | schedule(); | ||
1872 | mutex_lock(&root->log_mutex); | ||
1873 | finish_wait(&root->log_writer_wait, &wait); | ||
1874 | } | ||
1925 | return 0; | 1875 | return 0; |
1926 | } | 1876 | } |
1927 | 1877 | ||
@@ -1933,57 +1883,114 @@ static int wait_log_commit(struct btrfs_root *log) | |||
1933 | int btrfs_sync_log(struct btrfs_trans_handle *trans, | 1883 | int btrfs_sync_log(struct btrfs_trans_handle *trans, |
1934 | struct btrfs_root *root) | 1884 | struct btrfs_root *root) |
1935 | { | 1885 | { |
1886 | int index1; | ||
1887 | int index2; | ||
1936 | int ret; | 1888 | int ret; |
1937 | unsigned long batch; | ||
1938 | struct btrfs_root *log = root->log_root; | 1889 | struct btrfs_root *log = root->log_root; |
1890 | struct btrfs_root *log_root_tree = root->fs_info->log_root_tree; | ||
1939 | 1891 | ||
1940 | mutex_lock(&log->fs_info->tree_log_mutex); | 1892 | mutex_lock(&root->log_mutex); |
1941 | if (atomic_read(&log->fs_info->tree_log_commit)) { | 1893 | index1 = root->log_transid % 2; |
1942 | wait_log_commit(log); | 1894 | if (atomic_read(&root->log_commit[index1])) { |
1943 | goto out; | 1895 | wait_log_commit(root, root->log_transid); |
1896 | mutex_unlock(&root->log_mutex); | ||
1897 | return 0; | ||
1944 | } | 1898 | } |
1945 | atomic_set(&log->fs_info->tree_log_commit, 1); | 1899 | atomic_set(&root->log_commit[index1], 1); |
1900 | |||
1901 | /* wait for previous tree log sync to complete */ | ||
1902 | if (atomic_read(&root->log_commit[(index1 + 1) % 2])) | ||
1903 | wait_log_commit(root, root->log_transid - 1); | ||
1946 | 1904 | ||
1947 | while (1) { | 1905 | while (1) { |
1948 | batch = log->fs_info->tree_log_batch; | 1906 | unsigned long batch = root->log_batch; |
1949 | mutex_unlock(&log->fs_info->tree_log_mutex); | 1907 | mutex_unlock(&root->log_mutex); |
1950 | schedule_timeout_uninterruptible(1); | 1908 | schedule_timeout_uninterruptible(1); |
1951 | mutex_lock(&log->fs_info->tree_log_mutex); | 1909 | mutex_lock(&root->log_mutex); |
1952 | 1910 | wait_for_writer(root); | |
1953 | while (atomic_read(&log->fs_info->tree_log_writers)) { | 1911 | if (batch == root->log_batch) |
1954 | DEFINE_WAIT(wait); | ||
1955 | prepare_to_wait(&log->fs_info->tree_log_wait, &wait, | ||
1956 | TASK_UNINTERRUPTIBLE); | ||
1957 | mutex_unlock(&log->fs_info->tree_log_mutex); | ||
1958 | if (atomic_read(&log->fs_info->tree_log_writers)) | ||
1959 | schedule(); | ||
1960 | mutex_lock(&log->fs_info->tree_log_mutex); | ||
1961 | finish_wait(&log->fs_info->tree_log_wait, &wait); | ||
1962 | } | ||
1963 | if (batch == log->fs_info->tree_log_batch) | ||
1964 | break; | 1912 | break; |
1965 | } | 1913 | } |
1966 | 1914 | ||
1967 | ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); | 1915 | ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); |
1968 | BUG_ON(ret); | 1916 | BUG_ON(ret); |
1969 | ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree, | 1917 | |
1970 | &root->fs_info->log_root_tree->dirty_log_pages); | 1918 | btrfs_set_root_bytenr(&log->root_item, log->node->start); |
1919 | btrfs_set_root_generation(&log->root_item, trans->transid); | ||
1920 | btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node)); | ||
1921 | |||
1922 | root->log_batch = 0; | ||
1923 | root->log_transid++; | ||
1924 | log->log_transid = root->log_transid; | ||
1925 | smp_mb(); | ||
1926 | /* | ||
1927 | * log tree has been flushed to disk, new modifications of | ||
1928 | * the log will be written to new positions. so it's safe to | ||
1929 | * allow log writers to go in. | ||
1930 | */ | ||
1931 | mutex_unlock(&root->log_mutex); | ||
1932 | |||
1933 | mutex_lock(&log_root_tree->log_mutex); | ||
1934 | log_root_tree->log_batch++; | ||
1935 | atomic_inc(&log_root_tree->log_writers); | ||
1936 | mutex_unlock(&log_root_tree->log_mutex); | ||
1937 | |||
1938 | ret = update_log_root(trans, log); | ||
1939 | BUG_ON(ret); | ||
1940 | |||
1941 | mutex_lock(&log_root_tree->log_mutex); | ||
1942 | if (atomic_dec_and_test(&log_root_tree->log_writers)) { | ||
1943 | smp_mb(); | ||
1944 | if (waitqueue_active(&log_root_tree->log_writer_wait)) | ||
1945 | wake_up(&log_root_tree->log_writer_wait); | ||
1946 | } | ||
1947 | |||
1948 | index2 = log_root_tree->log_transid % 2; | ||
1949 | if (atomic_read(&log_root_tree->log_commit[index2])) { | ||
1950 | wait_log_commit(log_root_tree, log_root_tree->log_transid); | ||
1951 | mutex_unlock(&log_root_tree->log_mutex); | ||
1952 | goto out; | ||
1953 | } | ||
1954 | atomic_set(&log_root_tree->log_commit[index2], 1); | ||
1955 | |||
1956 | if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) | ||
1957 | wait_log_commit(log_root_tree, log_root_tree->log_transid - 1); | ||
1958 | |||
1959 | wait_for_writer(log_root_tree); | ||
1960 | |||
1961 | ret = btrfs_write_and_wait_marked_extents(log_root_tree, | ||
1962 | &log_root_tree->dirty_log_pages); | ||
1971 | BUG_ON(ret); | 1963 | BUG_ON(ret); |
1972 | 1964 | ||
1973 | btrfs_set_super_log_root(&root->fs_info->super_for_commit, | 1965 | btrfs_set_super_log_root(&root->fs_info->super_for_commit, |
1974 | log->fs_info->log_root_tree->node->start); | 1966 | log_root_tree->node->start); |
1975 | btrfs_set_super_log_root_level(&root->fs_info->super_for_commit, | 1967 | btrfs_set_super_log_root_level(&root->fs_info->super_for_commit, |
1976 | btrfs_header_level(log->fs_info->log_root_tree->node)); | 1968 | btrfs_header_level(log_root_tree->node)); |
1969 | |||
1970 | log_root_tree->log_batch = 0; | ||
1971 | log_root_tree->log_transid++; | ||
1972 | smp_mb(); | ||
1973 | |||
1974 | mutex_unlock(&log_root_tree->log_mutex); | ||
1975 | |||
1976 | /* | ||
1977 | * nobody else is going to jump in and write the the ctree | ||
1978 | * super here because the log_commit atomic below is protecting | ||
1979 | * us. We must be called with a transaction handle pinning | ||
1980 | * the running transaction open, so a full commit can't hop | ||
1981 | * in and cause problems either. | ||
1982 | */ | ||
1983 | write_ctree_super(trans, root->fs_info->tree_root, 2); | ||
1977 | 1984 | ||
1978 | write_ctree_super(trans, log->fs_info->tree_root, 2); | 1985 | atomic_set(&log_root_tree->log_commit[index2], 0); |
1979 | log->fs_info->tree_log_transid++; | ||
1980 | log->fs_info->tree_log_batch = 0; | ||
1981 | atomic_set(&log->fs_info->tree_log_commit, 0); | ||
1982 | smp_mb(); | 1986 | smp_mb(); |
1983 | if (waitqueue_active(&log->fs_info->tree_log_wait)) | 1987 | if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) |
1984 | wake_up(&log->fs_info->tree_log_wait); | 1988 | wake_up(&log_root_tree->log_commit_wait[index2]); |
1985 | out: | 1989 | out: |
1986 | mutex_unlock(&log->fs_info->tree_log_mutex); | 1990 | atomic_set(&root->log_commit[index1], 0); |
1991 | smp_mb(); | ||
1992 | if (waitqueue_active(&root->log_commit_wait[index1])) | ||
1993 | wake_up(&root->log_commit_wait[index1]); | ||
1987 | return 0; | 1994 | return 0; |
1988 | } | 1995 | } |
1989 | 1996 | ||
@@ -2019,38 +2026,18 @@ int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) | |||
2019 | start, end, GFP_NOFS); | 2026 | start, end, GFP_NOFS); |
2020 | } | 2027 | } |
2021 | 2028 | ||
2022 | log = root->log_root; | 2029 | if (log->log_transid > 0) { |
2023 | ret = btrfs_del_root(trans, root->fs_info->log_root_tree, | 2030 | ret = btrfs_del_root(trans, root->fs_info->log_root_tree, |
2024 | &log->root_key); | 2031 | &log->root_key); |
2025 | BUG_ON(ret); | 2032 | BUG_ON(ret); |
2033 | } | ||
2026 | root->log_root = NULL; | 2034 | root->log_root = NULL; |
2027 | kfree(root->log_root); | 2035 | free_extent_buffer(log->node); |
2036 | kfree(log); | ||
2028 | return 0; | 2037 | return 0; |
2029 | } | 2038 | } |
2030 | 2039 | ||
2031 | /* | 2040 | /* |
2032 | * helper function to update the item for a given subvolumes log root | ||
2033 | * in the tree of log roots | ||
2034 | */ | ||
2035 | static int update_log_root(struct btrfs_trans_handle *trans, | ||
2036 | struct btrfs_root *log) | ||
2037 | { | ||
2038 | u64 bytenr = btrfs_root_bytenr(&log->root_item); | ||
2039 | int ret; | ||
2040 | |||
2041 | if (log->node->start == bytenr) | ||
2042 | return 0; | ||
2043 | |||
2044 | btrfs_set_root_bytenr(&log->root_item, log->node->start); | ||
2045 | btrfs_set_root_generation(&log->root_item, trans->transid); | ||
2046 | btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node)); | ||
2047 | ret = btrfs_update_root(trans, log->fs_info->log_root_tree, | ||
2048 | &log->root_key, &log->root_item); | ||
2049 | BUG_ON(ret); | ||
2050 | return ret; | ||
2051 | } | ||
2052 | |||
2053 | /* | ||
2054 | * If both a file and directory are logged, and unlinks or renames are | 2041 | * If both a file and directory are logged, and unlinks or renames are |
2055 | * mixed in, we have a few interesting corners: | 2042 | * mixed in, we have a few interesting corners: |
2056 | * | 2043 | * |
@@ -2711,11 +2698,6 @@ next_slot: | |||
2711 | 2698 | ||
2712 | btrfs_free_path(path); | 2699 | btrfs_free_path(path); |
2713 | btrfs_free_path(dst_path); | 2700 | btrfs_free_path(dst_path); |
2714 | |||
2715 | mutex_lock(&root->fs_info->tree_log_mutex); | ||
2716 | ret = update_log_root(trans, log); | ||
2717 | BUG_ON(ret); | ||
2718 | mutex_unlock(&root->fs_info->tree_log_mutex); | ||
2719 | out: | 2701 | out: |
2720 | return 0; | 2702 | return 0; |
2721 | } | 2703 | } |