aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorChris Mason <chris.mason@oracle.com>2008-09-05 16:13:11 -0400
committerChris Mason <chris.mason@oracle.com>2008-09-25 11:04:07 -0400
commite02119d5a7b4396c5a872582fddc8bd6d305a70a (patch)
tree825efe2a79dbca8d61256183f3526a5b5dc40dc6
parenta1b32a5932cfac7c38b442582285f3da2a09dfd8 (diff)
Btrfs: Add a write ahead tree log to optimize synchronous operations
File syncs and directory syncs are optimized by copying their items into a special (copy-on-write) log tree. There is one log tree per subvolume and the btrfs super block points to a tree of log tree roots. After a crash, items are copied out of the log tree and back into the subvolume. See tree-log.c for all the details. Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat
-rw-r--r--fs/btrfs/Makefile3
-rw-r--r--fs/btrfs/btrfs_inode.h8
-rw-r--r--fs/btrfs/compat.h15
-rw-r--r--fs/btrfs/ctree.c73
-rw-r--r--fs/btrfs/ctree.h72
-rw-r--r--fs/btrfs/dir-item.c3
-rw-r--r--fs/btrfs/disk-io.c138
-rw-r--r--fs/btrfs/disk-io.h8
-rw-r--r--fs/btrfs/extent-tree.c93
-rw-r--r--fs/btrfs/file.c39
-rw-r--r--fs/btrfs/inode.c261
-rw-r--r--fs/btrfs/root-tree.c5
-rw-r--r--fs/btrfs/transaction.c45
-rw-r--r--fs/btrfs/transaction.h1
-rw-r--r--fs/btrfs/tree-defrag.c4
-rw-r--r--fs/btrfs/tree-log.c2804
-rw-r--r--fs/btrfs/tree-log.h41
17 files changed, 3408 insertions, 205 deletions
diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index 3a01065d4ef1..b7addbfd8c22 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -7,8 +7,7 @@ btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
7 transaction.o bit-radix.o inode.o file.o tree-defrag.o \ 7 transaction.o bit-radix.o inode.o file.o tree-defrag.o \
8 extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \ 8 extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
9 extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \ 9 extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
10 ref-cache.o acl.o export.o 10 ref-cache.o export.o tree-log.o acl.o
11
12else 11else
13 12
14# Normal Makefile 13# Normal Makefile
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index 111f90524ae1..fcc8cf27e906 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -33,6 +33,7 @@ struct btrfs_inode {
33 struct extent_io_tree io_failure_tree; 33 struct extent_io_tree io_failure_tree;
34 struct mutex csum_mutex; 34 struct mutex csum_mutex;
35 struct mutex extent_mutex; 35 struct mutex extent_mutex;
36 struct mutex log_mutex;
36 struct inode vfs_inode; 37 struct inode vfs_inode;
37 struct btrfs_ordered_inode_tree ordered_tree; 38 struct btrfs_ordered_inode_tree ordered_tree;
38 39
@@ -44,10 +45,17 @@ struct btrfs_inode {
44 45
45 struct list_head delalloc_inodes; 46 struct list_head delalloc_inodes;
46 47
48 /* full 64 bit generation number */
49 u64 generation;
50
47 /* 51 /*
48 * transid of the trans_handle that last modified this inode 52 * transid of the trans_handle that last modified this inode
49 */ 53 */
50 u64 last_trans; 54 u64 last_trans;
55 /*
56 * transid that last logged this inode
57 */
58 u64 logged_trans;
51 u64 delalloc_bytes; 59 u64 delalloc_bytes;
52 u64 disk_i_size; 60 u64 disk_i_size;
53 u32 flags; 61 u32 flags;
diff --git a/fs/btrfs/compat.h b/fs/btrfs/compat.h
index d45fb37887bc..b0ed1887d9b1 100644
--- a/fs/btrfs/compat.h
+++ b/fs/btrfs/compat.h
@@ -22,6 +22,21 @@ static inline struct dentry *d_obtain_alias(struct inode *inode)
22} 22}
23#endif 23#endif
24 24
25#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
26static inline void btrfs_drop_nlink(struct inode *inode)
27{
28 inode->i_nlink--;
29}
30
31static inline void btrfs_inc_nlink(struct inode *inode)
32{
33 inode->i_nlink++;
34}
35#else
36# define btrfs_drop_nlink(inode) drop_nlink(inode)
37# define btrfs_inc_nlink(inode) inc_nlink(inode)
38#endif
39
25/* 40/*
26 * Even if AppArmor isn't enabled, it still has different prototypes. 41 * Even if AppArmor isn't enabled, it still has different prototypes.
27 * Add more distro/version pairs here to declare which has AppArmor applied. 42 * Add more distro/version pairs here to declare which has AppArmor applied.
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 7114faafa9d4..579124043d9b 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -60,7 +60,7 @@ void btrfs_free_path(struct btrfs_path *p)
60 kmem_cache_free(btrfs_path_cachep, p); 60 kmem_cache_free(btrfs_path_cachep, p);
61} 61}
62 62
63void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p) 63void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
64{ 64{
65 int i; 65 int i;
66 66
@@ -176,7 +176,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
176 return 0; 176 return 0;
177} 177}
178 178
179int __btrfs_cow_block(struct btrfs_trans_handle *trans, 179int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
180 struct btrfs_root *root, 180 struct btrfs_root *root,
181 struct extent_buffer *buf, 181 struct extent_buffer *buf,
182 struct extent_buffer *parent, int parent_slot, 182 struct extent_buffer *parent, int parent_slot,
@@ -294,7 +294,7 @@ int __btrfs_cow_block(struct btrfs_trans_handle *trans,
294 return 0; 294 return 0;
295} 295}
296 296
297int btrfs_cow_block(struct btrfs_trans_handle *trans, 297int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
298 struct btrfs_root *root, struct extent_buffer *buf, 298 struct btrfs_root *root, struct extent_buffer *buf,
299 struct extent_buffer *parent, int parent_slot, 299 struct extent_buffer *parent, int parent_slot,
300 struct extent_buffer **cow_ret, u64 prealloc_dest) 300 struct extent_buffer **cow_ret, u64 prealloc_dest)
@@ -677,9 +677,10 @@ static int noinline check_block(struct btrfs_root *root,
677 * 677 *
678 * slot may point to max if the key is bigger than all of the keys 678 * slot may point to max if the key is bigger than all of the keys
679 */ 679 */
680static int generic_bin_search(struct extent_buffer *eb, unsigned long p, 680static noinline int generic_bin_search(struct extent_buffer *eb,
681 int item_size, struct btrfs_key *key, 681 unsigned long p,
682 int max, int *slot) 682 int item_size, struct btrfs_key *key,
683 int max, int *slot)
683{ 684{
684 int low = 0; 685 int low = 0;
685 int high = max; 686 int high = max;
@@ -765,7 +766,7 @@ static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
765 return -1; 766 return -1;
766} 767}
767 768
768static struct extent_buffer *read_node_slot(struct btrfs_root *root, 769static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
769 struct extent_buffer *parent, int slot) 770 struct extent_buffer *parent, int slot)
770{ 771{
771 int level = btrfs_header_level(parent); 772 int level = btrfs_header_level(parent);
@@ -781,7 +782,7 @@ static struct extent_buffer *read_node_slot(struct btrfs_root *root,
781 btrfs_node_ptr_generation(parent, slot)); 782 btrfs_node_ptr_generation(parent, slot));
782} 783}
783 784
784static int balance_level(struct btrfs_trans_handle *trans, 785static noinline int balance_level(struct btrfs_trans_handle *trans,
785 struct btrfs_root *root, 786 struct btrfs_root *root,
786 struct btrfs_path *path, int level) 787 struct btrfs_path *path, int level)
787{ 788{
@@ -1128,8 +1129,9 @@ static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
1128/* 1129/*
1129 * readahead one full node of leaves 1130 * readahead one full node of leaves
1130 */ 1131 */
1131static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path, 1132static noinline void reada_for_search(struct btrfs_root *root,
1132 int level, int slot, u64 objectid) 1133 struct btrfs_path *path,
1134 int level, int slot, u64 objectid)
1133{ 1135{
1134 struct extent_buffer *node; 1136 struct extent_buffer *node;
1135 struct btrfs_disk_key disk_key; 1137 struct btrfs_disk_key disk_key;
@@ -1201,7 +1203,8 @@ static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1201 } 1203 }
1202} 1204}
1203 1205
1204static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock) 1206static noinline void unlock_up(struct btrfs_path *path, int level,
1207 int lowest_unlock)
1205{ 1208{
1206 int i; 1209 int i;
1207 int skip_level = level; 1210 int skip_level = level;
@@ -1759,8 +1762,9 @@ static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1759 * 1762 *
1760 * returns 0 on success and < 0 on failure 1763 * returns 0 on success and < 0 on failure
1761 */ 1764 */
1762static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 1765static noinline int split_node(struct btrfs_trans_handle *trans,
1763 *root, struct btrfs_path *path, int level) 1766 struct btrfs_root *root,
1767 struct btrfs_path *path, int level)
1764{ 1768{
1765 u64 root_gen; 1769 u64 root_gen;
1766 struct extent_buffer *c; 1770 struct extent_buffer *c;
@@ -1874,7 +1878,8 @@ static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1874 * the start of the leaf data. IOW, how much room 1878 * the start of the leaf data. IOW, how much room
1875 * the leaf has left for both items and data 1879 * the leaf has left for both items and data
1876 */ 1880 */
1877int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf) 1881int noinline btrfs_leaf_free_space(struct btrfs_root *root,
1882 struct extent_buffer *leaf)
1878{ 1883{
1879 int nritems = btrfs_header_nritems(leaf); 1884 int nritems = btrfs_header_nritems(leaf);
1880 int ret; 1885 int ret;
@@ -2283,9 +2288,11 @@ out:
2283 * 2288 *
2284 * returns 0 if all went well and < 0 on failure. 2289 * returns 0 if all went well and < 0 on failure.
2285 */ 2290 */
2286static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root 2291static noinline int split_leaf(struct btrfs_trans_handle *trans,
2287 *root, struct btrfs_key *ins_key, 2292 struct btrfs_root *root,
2288 struct btrfs_path *path, int data_size, int extend) 2293 struct btrfs_key *ins_key,
2294 struct btrfs_path *path, int data_size,
2295 int extend)
2289{ 2296{
2290 u64 root_gen; 2297 u64 root_gen;
2291 struct extent_buffer *l; 2298 struct extent_buffer *l;
@@ -3079,6 +3086,7 @@ int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3079 * was nothing in the tree that matched the search criteria. 3086 * was nothing in the tree that matched the search criteria.
3080 */ 3087 */
3081int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 3088int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3089 struct btrfs_key *max_key,
3082 struct btrfs_path *path, int cache_only, 3090 struct btrfs_path *path, int cache_only,
3083 u64 min_trans) 3091 u64 min_trans)
3084{ 3092{
@@ -3093,6 +3101,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3093again: 3101again:
3094 cur = btrfs_lock_root_node(root); 3102 cur = btrfs_lock_root_node(root);
3095 level = btrfs_header_level(cur); 3103 level = btrfs_header_level(cur);
3104 WARN_ON(path->nodes[level]);
3096 path->nodes[level] = cur; 3105 path->nodes[level] = cur;
3097 path->locks[level] = 1; 3106 path->locks[level] = 1;
3098 3107
@@ -3107,6 +3116,8 @@ again:
3107 3116
3108 /* at level = 0, we're done, setup the path and exit */ 3117 /* at level = 0, we're done, setup the path and exit */
3109 if (level == 0) { 3118 if (level == 0) {
3119 if (slot >= nritems)
3120 goto find_next_key;
3110 ret = 0; 3121 ret = 0;
3111 path->slots[level] = slot; 3122 path->slots[level] = slot;
3112 btrfs_item_key_to_cpu(cur, &found_key, slot); 3123 btrfs_item_key_to_cpu(cur, &found_key, slot);
@@ -3123,6 +3134,8 @@ again:
3123 u64 blockptr; 3134 u64 blockptr;
3124 u64 gen; 3135 u64 gen;
3125 struct extent_buffer *tmp; 3136 struct extent_buffer *tmp;
3137 struct btrfs_disk_key disk_key;
3138
3126 blockptr = btrfs_node_blockptr(cur, slot); 3139 blockptr = btrfs_node_blockptr(cur, slot);
3127 gen = btrfs_node_ptr_generation(cur, slot); 3140 gen = btrfs_node_ptr_generation(cur, slot);
3128 if (gen < min_trans) { 3141 if (gen < min_trans) {
@@ -3132,6 +3145,14 @@ again:
3132 if (!cache_only) 3145 if (!cache_only)
3133 break; 3146 break;
3134 3147
3148 if (max_key) {
3149 btrfs_node_key(cur, &disk_key, slot);
3150 if (comp_keys(&disk_key, max_key) >= 0) {
3151 ret = 1;
3152 goto out;
3153 }
3154 }
3155
3135 tmp = btrfs_find_tree_block(root, blockptr, 3156 tmp = btrfs_find_tree_block(root, blockptr,
3136 btrfs_level_size(root, level - 1)); 3157 btrfs_level_size(root, level - 1));
3137 3158
@@ -3143,14 +3164,16 @@ again:
3143 free_extent_buffer(tmp); 3164 free_extent_buffer(tmp);
3144 slot++; 3165 slot++;
3145 } 3166 }
3167find_next_key:
3146 /* 3168 /*
3147 * we didn't find a candidate key in this node, walk forward 3169 * we didn't find a candidate key in this node, walk forward
3148 * and find another one 3170 * and find another one
3149 */ 3171 */
3150 if (slot >= nritems) { 3172 if (slot >= nritems) {
3151 ret = btrfs_find_next_key(root, path, min_key, level, 3173 path->slots[level] = slot;
3174 sret = btrfs_find_next_key(root, path, min_key, level,
3152 cache_only, min_trans); 3175 cache_only, min_trans);
3153 if (ret == 0) { 3176 if (sret == 0) {
3154 btrfs_release_path(root, path); 3177 btrfs_release_path(root, path);
3155 goto again; 3178 goto again;
3156 } else { 3179 } else {
@@ -3351,6 +3374,7 @@ int btrfs_previous_item(struct btrfs_root *root,
3351{ 3374{
3352 struct btrfs_key found_key; 3375 struct btrfs_key found_key;
3353 struct extent_buffer *leaf; 3376 struct extent_buffer *leaf;
3377 u32 nritems;
3354 int ret; 3378 int ret;
3355 3379
3356 while(1) { 3380 while(1) {
@@ -3362,9 +3386,20 @@ int btrfs_previous_item(struct btrfs_root *root,
3362 path->slots[0]--; 3386 path->slots[0]--;
3363 } 3387 }
3364 leaf = path->nodes[0]; 3388 leaf = path->nodes[0];
3389 nritems = btrfs_header_nritems(leaf);
3390 if (nritems == 0)
3391 return 1;
3392 if (path->slots[0] == nritems)
3393 path->slots[0]--;
3394
3365 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 3395 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3366 if (found_key.type == type) 3396 if (found_key.type == type)
3367 return 0; 3397 return 0;
3398 if (found_key.objectid < min_objectid)
3399 break;
3400 if (found_key.objectid == min_objectid &&
3401 found_key.type < type)
3402 break;
3368 } 3403 }
3369 return 1; 3404 return 1;
3370} 3405}
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index b305ae7e10b0..6532b60683ef 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -77,6 +77,10 @@ struct btrfs_ordered_sum;
77/* orhpan objectid for tracking unlinked/truncated files */ 77/* orhpan objectid for tracking unlinked/truncated files */
78#define BTRFS_ORPHAN_OBJECTID -5ULL 78#define BTRFS_ORPHAN_OBJECTID -5ULL
79 79
80/* does write ahead logging to speed up fsyncs */
81#define BTRFS_TREE_LOG_OBJECTID -6ULL
82#define BTRFS_TREE_LOG_FIXUP_OBJECTID -7ULL
83
80/* 84/*
81 * All files have objectids higher than this. 85 * All files have objectids higher than this.
82 */ 86 */
@@ -276,6 +280,7 @@ struct btrfs_super_block {
276 __le64 generation; 280 __le64 generation;
277 __le64 root; 281 __le64 root;
278 __le64 chunk_root; 282 __le64 chunk_root;
283 __le64 log_root;
279 __le64 total_bytes; 284 __le64 total_bytes;
280 __le64 bytes_used; 285 __le64 bytes_used;
281 __le64 root_dir_objectid; 286 __le64 root_dir_objectid;
@@ -287,6 +292,7 @@ struct btrfs_super_block {
287 __le32 sys_chunk_array_size; 292 __le32 sys_chunk_array_size;
288 u8 root_level; 293 u8 root_level;
289 u8 chunk_root_level; 294 u8 chunk_root_level;
295 u8 log_root_level;
290 struct btrfs_dev_item dev_item; 296 struct btrfs_dev_item dev_item;
291 char label[BTRFS_LABEL_SIZE]; 297 char label[BTRFS_LABEL_SIZE];
292 u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE]; 298 u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE];
@@ -392,7 +398,10 @@ struct btrfs_timespec {
392 * make a new item type 398 * make a new item type
393 */ 399 */
394struct btrfs_inode_item { 400struct btrfs_inode_item {
401 /* nfs style generation number */
395 __le64 generation; 402 __le64 generation;
403 /* transid that last touched this inode */
404 __le64 transid;
396 __le64 size; 405 __le64 size;
397 __le64 nblocks; 406 __le64 nblocks;
398 __le64 block_group; 407 __le64 block_group;
@@ -409,8 +418,13 @@ struct btrfs_inode_item {
409 struct btrfs_timespec otime; 418 struct btrfs_timespec otime;
410} __attribute__ ((__packed__)); 419} __attribute__ ((__packed__));
411 420
421struct btrfs_dir_log_item {
422 __le64 end;
423} __attribute__ ((__packed__));
424
412struct btrfs_dir_item { 425struct btrfs_dir_item {
413 struct btrfs_disk_key location; 426 struct btrfs_disk_key location;
427 __le64 transid;
414 __le16 data_len; 428 __le16 data_len;
415 __le16 name_len; 429 __le16 name_len;
416 u8 type; 430 u8 type;
@@ -505,6 +519,9 @@ struct btrfs_fs_info {
505 struct btrfs_root *tree_root; 519 struct btrfs_root *tree_root;
506 struct btrfs_root *chunk_root; 520 struct btrfs_root *chunk_root;
507 struct btrfs_root *dev_root; 521 struct btrfs_root *dev_root;
522
523 /* the log root tree is a directory of all the other log roots */
524 struct btrfs_root *log_root_tree;
508 struct radix_tree_root fs_roots_radix; 525 struct radix_tree_root fs_roots_radix;
509 526
510 struct extent_io_tree free_space_cache; 527 struct extent_io_tree free_space_cache;
@@ -518,6 +535,7 @@ struct btrfs_fs_info {
518 535
519 u64 generation; 536 u64 generation;
520 u64 last_trans_committed; 537 u64 last_trans_committed;
538 u64 last_trans_new_blockgroup;
521 u64 open_ioctl_trans; 539 u64 open_ioctl_trans;
522 unsigned long mount_opt; 540 unsigned long mount_opt;
523 u64 max_extent; 541 u64 max_extent;
@@ -527,6 +545,9 @@ struct btrfs_fs_info {
527 wait_queue_head_t transaction_throttle; 545 wait_queue_head_t transaction_throttle;
528 wait_queue_head_t transaction_wait; 546 wait_queue_head_t transaction_wait;
529 wait_queue_head_t async_submit_wait; 547 wait_queue_head_t async_submit_wait;
548
549 wait_queue_head_t tree_log_wait;
550
530 struct btrfs_super_block super_copy; 551 struct btrfs_super_block super_copy;
531 struct btrfs_super_block super_for_commit; 552 struct btrfs_super_block super_for_commit;
532 struct block_device *__bdev; 553 struct block_device *__bdev;
@@ -535,6 +556,7 @@ struct btrfs_fs_info {
535 struct backing_dev_info bdi; 556 struct backing_dev_info bdi;
536 spinlock_t hash_lock; 557 spinlock_t hash_lock;
537 struct mutex trans_mutex; 558 struct mutex trans_mutex;
559 struct mutex tree_log_mutex;
538 struct mutex transaction_kthread_mutex; 560 struct mutex transaction_kthread_mutex;
539 struct mutex cleaner_mutex; 561 struct mutex cleaner_mutex;
540 struct mutex alloc_mutex; 562 struct mutex alloc_mutex;
@@ -544,8 +566,13 @@ struct btrfs_fs_info {
544 struct list_head trans_list; 566 struct list_head trans_list;
545 struct list_head hashers; 567 struct list_head hashers;
546 struct list_head dead_roots; 568 struct list_head dead_roots;
569
547 atomic_t nr_async_submits; 570 atomic_t nr_async_submits;
548 atomic_t nr_async_bios; 571 atomic_t nr_async_bios;
572 atomic_t tree_log_writers;
573 atomic_t tree_log_commit;
574 unsigned long tree_log_batch;
575 u64 tree_log_transid;
549 576
550 /* 577 /*
551 * this is used by the balancing code to wait for all the pending 578 * this is used by the balancing code to wait for all the pending
@@ -583,6 +610,7 @@ struct btrfs_fs_info {
583 struct completion kobj_unregister; 610 struct completion kobj_unregister;
584 int do_barriers; 611 int do_barriers;
585 int closing; 612 int closing;
613 int log_root_recovering;
586 atomic_t throttles; 614 atomic_t throttles;
587 atomic_t throttle_gen; 615 atomic_t throttle_gen;
588 616
@@ -596,6 +624,7 @@ struct btrfs_fs_info {
596 u64 delalloc_bytes; 624 u64 delalloc_bytes;
597 u64 last_alloc; 625 u64 last_alloc;
598 u64 last_data_alloc; 626 u64 last_data_alloc;
627 u64 last_log_alloc;
599 628
600 spinlock_t ref_cache_lock; 629 spinlock_t ref_cache_lock;
601 u64 total_ref_cache_size; 630 u64 total_ref_cache_size;
@@ -632,6 +661,7 @@ struct btrfs_root {
632 struct btrfs_leaf_ref_tree *ref_tree; 661 struct btrfs_leaf_ref_tree *ref_tree;
633 struct btrfs_leaf_ref_tree ref_tree_struct; 662 struct btrfs_leaf_ref_tree ref_tree_struct;
634 struct btrfs_dirty_root *dirty_root; 663 struct btrfs_dirty_root *dirty_root;
664 struct btrfs_root *log_root;
635 665
636 struct btrfs_root_item root_item; 666 struct btrfs_root_item root_item;
637 struct btrfs_key root_key; 667 struct btrfs_key root_key;
@@ -640,6 +670,7 @@ struct btrfs_root {
640 struct kobject root_kobj; 670 struct kobject root_kobj;
641 struct completion kobj_unregister; 671 struct completion kobj_unregister;
642 struct mutex objectid_mutex; 672 struct mutex objectid_mutex;
673 struct mutex log_mutex;
643 674
644 u64 objectid; 675 u64 objectid;
645 u64 last_trans; 676 u64 last_trans;
@@ -692,6 +723,8 @@ struct btrfs_root {
692 * dir items are the name -> inode pointers in a directory. There is one 723 * dir items are the name -> inode pointers in a directory. There is one
693 * for every name in a directory. 724 * for every name in a directory.
694 */ 725 */
726#define BTRFS_DIR_LOG_ITEM_KEY 14
727#define BTRFS_DIR_LOG_INDEX_KEY 15
695#define BTRFS_DIR_ITEM_KEY 16 728#define BTRFS_DIR_ITEM_KEY 16
696#define BTRFS_DIR_INDEX_KEY 17 729#define BTRFS_DIR_INDEX_KEY 17
697/* 730/*
@@ -703,7 +736,8 @@ struct btrfs_root {
703 */ 736 */
704#define BTRFS_CSUM_ITEM_KEY 19 737#define BTRFS_CSUM_ITEM_KEY 19
705 738
706/* reserve 20-31 for other file stuff */ 739
740/* reserve 21-31 for other file/dir stuff */
707 741
708/* 742/*
709 * root items point to tree roots. There are typically in the root 743 * root items point to tree roots. There are typically in the root
@@ -938,6 +972,7 @@ BTRFS_SETGET_FUNCS(inode_ref_index, struct btrfs_inode_ref, index, 64);
938 972
939/* struct btrfs_inode_item */ 973/* struct btrfs_inode_item */
940BTRFS_SETGET_FUNCS(inode_generation, struct btrfs_inode_item, generation, 64); 974BTRFS_SETGET_FUNCS(inode_generation, struct btrfs_inode_item, generation, 64);
975BTRFS_SETGET_FUNCS(inode_transid, struct btrfs_inode_item, transid, 64);
941BTRFS_SETGET_FUNCS(inode_size, struct btrfs_inode_item, size, 64); 976BTRFS_SETGET_FUNCS(inode_size, struct btrfs_inode_item, size, 64);
942BTRFS_SETGET_FUNCS(inode_nblocks, struct btrfs_inode_item, nblocks, 64); 977BTRFS_SETGET_FUNCS(inode_nblocks, struct btrfs_inode_item, nblocks, 64);
943BTRFS_SETGET_FUNCS(inode_block_group, struct btrfs_inode_item, block_group, 64); 978BTRFS_SETGET_FUNCS(inode_block_group, struct btrfs_inode_item, block_group, 64);
@@ -1126,10 +1161,13 @@ static inline void btrfs_set_item_key(struct extent_buffer *eb,
1126 write_eb_member(eb, item, struct btrfs_item, key, disk_key); 1161 write_eb_member(eb, item, struct btrfs_item, key, disk_key);
1127} 1162}
1128 1163
1164BTRFS_SETGET_FUNCS(dir_log_end, struct btrfs_dir_log_item, end, 64);
1165
1129/* struct btrfs_dir_item */ 1166/* struct btrfs_dir_item */
1130BTRFS_SETGET_FUNCS(dir_data_len, struct btrfs_dir_item, data_len, 16); 1167BTRFS_SETGET_FUNCS(dir_data_len, struct btrfs_dir_item, data_len, 16);
1131BTRFS_SETGET_FUNCS(dir_type, struct btrfs_dir_item, type, 8); 1168BTRFS_SETGET_FUNCS(dir_type, struct btrfs_dir_item, type, 8);
1132BTRFS_SETGET_FUNCS(dir_name_len, struct btrfs_dir_item, name_len, 16); 1169BTRFS_SETGET_FUNCS(dir_name_len, struct btrfs_dir_item, name_len, 16);
1170BTRFS_SETGET_FUNCS(dir_transid, struct btrfs_dir_item, transid, 64);
1133 1171
1134static inline void btrfs_dir_item_key(struct extent_buffer *eb, 1172static inline void btrfs_dir_item_key(struct extent_buffer *eb,
1135 struct btrfs_dir_item *item, 1173 struct btrfs_dir_item *item,
@@ -1301,7 +1339,11 @@ BTRFS_SETGET_STACK_FUNCS(super_root_level, struct btrfs_super_block,
1301BTRFS_SETGET_STACK_FUNCS(super_chunk_root, struct btrfs_super_block, 1339BTRFS_SETGET_STACK_FUNCS(super_chunk_root, struct btrfs_super_block,
1302 chunk_root, 64); 1340 chunk_root, 64);
1303BTRFS_SETGET_STACK_FUNCS(super_chunk_root_level, struct btrfs_super_block, 1341BTRFS_SETGET_STACK_FUNCS(super_chunk_root_level, struct btrfs_super_block,
1304 chunk_root_level, 64); 1342 chunk_root_level, 8);
1343BTRFS_SETGET_STACK_FUNCS(super_log_root, struct btrfs_super_block,
1344 log_root, 64);
1345BTRFS_SETGET_STACK_FUNCS(super_log_root_level, struct btrfs_super_block,
1346 log_root_level, 8);
1305BTRFS_SETGET_STACK_FUNCS(super_total_bytes, struct btrfs_super_block, 1347BTRFS_SETGET_STACK_FUNCS(super_total_bytes, struct btrfs_super_block,
1306 total_bytes, 64); 1348 total_bytes, 64);
1307BTRFS_SETGET_STACK_FUNCS(super_bytes_used, struct btrfs_super_block, 1349BTRFS_SETGET_STACK_FUNCS(super_bytes_used, struct btrfs_super_block,
@@ -1405,6 +1447,12 @@ static inline struct dentry *fdentry(struct file *file) {
1405} 1447}
1406 1448
1407/* extent-tree.c */ 1449/* extent-tree.c */
1450int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
1451 u64 start, u64 len);
1452int btrfs_update_pinned_extents(struct btrfs_root *root,
1453 u64 bytenr, u64 num, int pin);
1454int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
1455 struct btrfs_root *root, struct extent_buffer *leaf);
1408int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans, 1456int btrfs_cross_ref_exists(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root, 1457 struct btrfs_root *root,
1410 struct btrfs_key *key, u64 bytenr); 1458 struct btrfs_key *key, u64 bytenr);
@@ -1448,6 +1496,11 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
1448 u64 root_objectid, u64 ref_generation, 1496 u64 root_objectid, u64 ref_generation,
1449 u64 owner, u64 owner_offset, 1497 u64 owner, u64 owner_offset,
1450 struct btrfs_key *ins); 1498 struct btrfs_key *ins);
1499int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
1500 struct btrfs_root *root,
1501 u64 root_objectid, u64 ref_generation,
1502 u64 owner, u64 owner_offset,
1503 struct btrfs_key *ins);
1451int btrfs_reserve_extent(struct btrfs_trans_handle *trans, 1504int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
1452 struct btrfs_root *root, 1505 struct btrfs_root *root,
1453 u64 num_bytes, u64 min_alloc_size, 1506 u64 num_bytes, u64 min_alloc_size,
@@ -1488,9 +1541,9 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
1488 struct btrfs_key *key, int lowest_level, 1541 struct btrfs_key *key, int lowest_level,
1489 int cache_only, u64 min_trans); 1542 int cache_only, u64 min_trans);
1490int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 1543int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
1544 struct btrfs_key *max_key,
1491 struct btrfs_path *path, int cache_only, 1545 struct btrfs_path *path, int cache_only,
1492 u64 min_trans); 1546 u64 min_trans);
1493
1494int btrfs_cow_block(struct btrfs_trans_handle *trans, 1547int btrfs_cow_block(struct btrfs_trans_handle *trans,
1495 struct btrfs_root *root, struct extent_buffer *buf, 1548 struct btrfs_root *root, struct extent_buffer *buf,
1496 struct extent_buffer *parent, int parent_slot, 1549 struct extent_buffer *parent, int parent_slot,
@@ -1656,6 +1709,18 @@ int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
1656#define PageChecked PageFsMisc 1709#define PageChecked PageFsMisc
1657#endif 1710#endif
1658 1711
1712int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1713 struct btrfs_root *root,
1714 struct inode *dir, struct inode *inode,
1715 const char *name, int name_len);
1716int btrfs_add_link(struct btrfs_trans_handle *trans,
1717 struct inode *parent_inode, struct inode *inode,
1718 const char *name, int name_len, int add_backref, u64 index);
1719int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1720 struct btrfs_root *root,
1721 struct inode *inode, u64 new_size,
1722 u32 min_type);
1723
1659int btrfs_start_delalloc_inodes(struct btrfs_root *root); 1724int btrfs_start_delalloc_inodes(struct btrfs_root *root);
1660int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end); 1725int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
1661int btrfs_writepages(struct address_space *mapping, 1726int btrfs_writepages(struct address_space *mapping,
@@ -1715,6 +1780,7 @@ int btrfs_update_inode(struct btrfs_trans_handle *trans,
1715long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 1780long btrfs_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1716 1781
1717/* file.c */ 1782/* file.c */
1783int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync);
1718int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end); 1784int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end);
1719int btrfs_check_file(struct btrfs_root *root, struct inode *inode); 1785int btrfs_check_file(struct btrfs_root *root, struct inode *inode);
1720extern struct file_operations btrfs_file_operations; 1786extern struct file_operations btrfs_file_operations;
diff --git a/fs/btrfs/dir-item.c b/fs/btrfs/dir-item.c
index 125094617fe8..e4f30090d640 100644
--- a/fs/btrfs/dir-item.c
+++ b/fs/btrfs/dir-item.c
@@ -96,6 +96,7 @@ int btrfs_insert_xattr_item(struct btrfs_trans_handle *trans,
96 btrfs_set_dir_item_key(leaf, dir_item, &disk_key); 96 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
97 btrfs_set_dir_type(leaf, dir_item, BTRFS_FT_XATTR); 97 btrfs_set_dir_type(leaf, dir_item, BTRFS_FT_XATTR);
98 btrfs_set_dir_name_len(leaf, dir_item, name_len); 98 btrfs_set_dir_name_len(leaf, dir_item, name_len);
99 btrfs_set_dir_transid(leaf, dir_item, trans->transid);
99 btrfs_set_dir_data_len(leaf, dir_item, data_len); 100 btrfs_set_dir_data_len(leaf, dir_item, data_len);
100 name_ptr = (unsigned long)(dir_item + 1); 101 name_ptr = (unsigned long)(dir_item + 1);
101 data_ptr = (unsigned long)((char *)name_ptr + name_len); 102 data_ptr = (unsigned long)((char *)name_ptr + name_len);
@@ -142,6 +143,7 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
142 btrfs_set_dir_type(leaf, dir_item, type); 143 btrfs_set_dir_type(leaf, dir_item, type);
143 btrfs_set_dir_data_len(leaf, dir_item, 0); 144 btrfs_set_dir_data_len(leaf, dir_item, 0);
144 btrfs_set_dir_name_len(leaf, dir_item, name_len); 145 btrfs_set_dir_name_len(leaf, dir_item, name_len);
146 btrfs_set_dir_transid(leaf, dir_item, trans->transid);
145 name_ptr = (unsigned long)(dir_item + 1); 147 name_ptr = (unsigned long)(dir_item + 1);
146 148
147 write_extent_buffer(leaf, name, name_ptr, name_len); 149 write_extent_buffer(leaf, name, name_ptr, name_len);
@@ -169,6 +171,7 @@ second_insert:
169 btrfs_set_dir_type(leaf, dir_item, type); 171 btrfs_set_dir_type(leaf, dir_item, type);
170 btrfs_set_dir_data_len(leaf, dir_item, 0); 172 btrfs_set_dir_data_len(leaf, dir_item, 0);
171 btrfs_set_dir_name_len(leaf, dir_item, name_len); 173 btrfs_set_dir_name_len(leaf, dir_item, name_len);
174 btrfs_set_dir_transid(leaf, dir_item, trans->transid);
172 name_ptr = (unsigned long)(dir_item + 1); 175 name_ptr = (unsigned long)(dir_item + 1);
173 write_extent_buffer(leaf, name, name_ptr, name_len); 176 write_extent_buffer(leaf, name, name_ptr, name_len);
174 btrfs_mark_buffer_dirty(leaf); 177 btrfs_mark_buffer_dirty(leaf);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 8e7a938bfbc7..a4373db5967a 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -41,6 +41,7 @@
41#include "async-thread.h" 41#include "async-thread.h"
42#include "locking.h" 42#include "locking.h"
43#include "ref-cache.h" 43#include "ref-cache.h"
44#include "tree-log.h"
44 45
45#if 0 46#if 0
46static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf) 47static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
@@ -694,6 +695,18 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
694} 695}
695 696
696 697
698int btrfs_write_tree_block(struct extent_buffer *buf)
699{
700 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
701 buf->start + buf->len - 1, WB_SYNC_NONE);
702}
703
704int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
705{
706 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
707 buf->start, buf->start + buf->len -1);
708}
709
697struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 710struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
698 u32 blocksize, u64 parent_transid) 711 u32 blocksize, u64 parent_transid)
699{ 712{
@@ -732,15 +745,6 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
732 return 0; 745 return 0;
733} 746}
734 747
735int wait_on_tree_block_writeback(struct btrfs_root *root,
736 struct extent_buffer *buf)
737{
738 struct inode *btree_inode = root->fs_info->btree_inode;
739 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
740 buf);
741 return 0;
742}
743
744static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 748static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
745 u32 stripesize, struct btrfs_root *root, 749 u32 stripesize, struct btrfs_root *root,
746 struct btrfs_fs_info *fs_info, 750 struct btrfs_fs_info *fs_info,
@@ -771,6 +775,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
771 spin_lock_init(&root->node_lock); 775 spin_lock_init(&root->node_lock);
772 spin_lock_init(&root->list_lock); 776 spin_lock_init(&root->list_lock);
773 mutex_init(&root->objectid_mutex); 777 mutex_init(&root->objectid_mutex);
778 mutex_init(&root->log_mutex);
774 779
775 btrfs_leaf_ref_tree_init(&root->ref_tree_struct); 780 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
776 root->ref_tree = &root->ref_tree_struct; 781 root->ref_tree = &root->ref_tree_struct;
@@ -809,11 +814,74 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
809 return 0; 814 return 0;
810} 815}
811 816
812struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info, 817int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
813 struct btrfs_key *location) 818 struct btrfs_fs_info *fs_info)
819{
820 struct extent_buffer *eb;
821 int ret;
822
823 if (!fs_info->log_root_tree)
824 return 0;
825
826 eb = fs_info->log_root_tree->node;
827
828 WARN_ON(btrfs_header_level(eb) != 0);
829 WARN_ON(btrfs_header_nritems(eb) != 0);
830
831 ret = btrfs_free_extent(trans, fs_info->tree_root,
832 eb->start, eb->len,
833 BTRFS_TREE_LOG_OBJECTID, 0, 0, 0, 1);
834 BUG_ON(ret);
835
836 free_extent_buffer(eb);
837 kfree(fs_info->log_root_tree);
838 fs_info->log_root_tree = NULL;
839 return 0;
840}
841
842int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
843 struct btrfs_fs_info *fs_info)
814{ 844{
815 struct btrfs_root *root; 845 struct btrfs_root *root;
816 struct btrfs_root *tree_root = fs_info->tree_root; 846 struct btrfs_root *tree_root = fs_info->tree_root;
847
848 root = kzalloc(sizeof(*root), GFP_NOFS);
849 if (!root)
850 return -ENOMEM;
851
852 __setup_root(tree_root->nodesize, tree_root->leafsize,
853 tree_root->sectorsize, tree_root->stripesize,
854 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
855
856 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
857 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
858 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
859 root->ref_cows = 0;
860
861 root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
862 BTRFS_TREE_LOG_OBJECTID,
863 0, 0, 0, 0, 0);
864
865 btrfs_set_header_nritems(root->node, 0);
866 btrfs_set_header_level(root->node, 0);
867 btrfs_set_header_bytenr(root->node, root->node->start);
868 btrfs_set_header_generation(root->node, trans->transid);
869 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
870
871 write_extent_buffer(root->node, root->fs_info->fsid,
872 (unsigned long)btrfs_header_fsid(root->node),
873 BTRFS_FSID_SIZE);
874 btrfs_mark_buffer_dirty(root->node);
875 btrfs_tree_unlock(root->node);
876 fs_info->log_root_tree = root;
877 return 0;
878}
879
880struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
881 struct btrfs_key *location)
882{
883 struct btrfs_root *root;
884 struct btrfs_fs_info *fs_info = tree_root->fs_info;
817 struct btrfs_path *path; 885 struct btrfs_path *path;
818 struct extent_buffer *l; 886 struct extent_buffer *l;
819 u64 highest_inode; 887 u64 highest_inode;
@@ -863,11 +931,13 @@ out:
863 blocksize, 0); 931 blocksize, 0);
864 BUG_ON(!root->node); 932 BUG_ON(!root->node);
865insert: 933insert:
866 root->ref_cows = 1; 934 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
867 ret = btrfs_find_highest_inode(root, &highest_inode); 935 root->ref_cows = 1;
868 if (ret == 0) { 936 ret = btrfs_find_highest_inode(root, &highest_inode);
869 root->highest_inode = highest_inode; 937 if (ret == 0) {
870 root->last_inode_alloc = highest_inode; 938 root->highest_inode = highest_inode;
939 root->last_inode_alloc = highest_inode;
940 }
871 } 941 }
872 return root; 942 return root;
873} 943}
@@ -907,7 +977,7 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
907 if (root) 977 if (root)
908 return root; 978 return root;
909 979
910 root = btrfs_read_fs_root_no_radix(fs_info, location); 980 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
911 if (IS_ERR(root)) 981 if (IS_ERR(root))
912 return root; 982 return root;
913 ret = radix_tree_insert(&fs_info->fs_roots_radix, 983 ret = radix_tree_insert(&fs_info->fs_roots_radix,
@@ -1250,16 +1320,18 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1250 u32 blocksize; 1320 u32 blocksize;
1251 u32 stripesize; 1321 u32 stripesize;
1252 struct buffer_head *bh; 1322 struct buffer_head *bh;
1253 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root), 1323 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1254 GFP_NOFS); 1324 GFP_NOFS);
1255 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root), 1325 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1256 GFP_NOFS); 1326 GFP_NOFS);
1257 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info), 1327 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1258 GFP_NOFS); 1328 GFP_NOFS);
1259 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root), 1329 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1260 GFP_NOFS); 1330 GFP_NOFS);
1261 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root), 1331 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1262 GFP_NOFS); 1332 GFP_NOFS);
1333 struct btrfs_root *log_tree_root;
1334
1263 int ret; 1335 int ret;
1264 int err = -EINVAL; 1336 int err = -EINVAL;
1265 1337
@@ -1343,6 +1415,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1343 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 1415 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1344 1416
1345 mutex_init(&fs_info->trans_mutex); 1417 mutex_init(&fs_info->trans_mutex);
1418 mutex_init(&fs_info->tree_log_mutex);
1346 mutex_init(&fs_info->drop_mutex); 1419 mutex_init(&fs_info->drop_mutex);
1347 mutex_init(&fs_info->alloc_mutex); 1420 mutex_init(&fs_info->alloc_mutex);
1348 mutex_init(&fs_info->chunk_mutex); 1421 mutex_init(&fs_info->chunk_mutex);
@@ -1352,6 +1425,10 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1352 init_waitqueue_head(&fs_info->transaction_throttle); 1425 init_waitqueue_head(&fs_info->transaction_throttle);
1353 init_waitqueue_head(&fs_info->transaction_wait); 1426 init_waitqueue_head(&fs_info->transaction_wait);
1354 init_waitqueue_head(&fs_info->async_submit_wait); 1427 init_waitqueue_head(&fs_info->async_submit_wait);
1428 init_waitqueue_head(&fs_info->tree_log_wait);
1429 atomic_set(&fs_info->tree_log_commit, 0);
1430 atomic_set(&fs_info->tree_log_writers, 0);
1431 fs_info->tree_log_transid = 0;
1355 1432
1356#if 0 1433#if 0
1357 ret = add_hasher(fs_info, "crc32c"); 1434 ret = add_hasher(fs_info, "crc32c");
@@ -1532,7 +1609,26 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1532 if (!fs_info->transaction_kthread) 1609 if (!fs_info->transaction_kthread)
1533 goto fail_cleaner; 1610 goto fail_cleaner;
1534 1611
1612 if (btrfs_super_log_root(disk_super) != 0) {
1613 u32 blocksize;
1614 u64 bytenr = btrfs_super_log_root(disk_super);
1615
1616 blocksize =
1617 btrfs_level_size(tree_root,
1618 btrfs_super_log_root_level(disk_super));
1535 1619
1620 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1621 GFP_NOFS);
1622
1623 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1624 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1625
1626 log_tree_root->node = read_tree_block(tree_root, bytenr,
1627 blocksize, 0);
1628 ret = btrfs_recover_log_trees(log_tree_root);
1629 BUG_ON(ret);
1630 }
1631 fs_info->last_trans_committed = btrfs_super_generation(disk_super);
1536 return tree_root; 1632 return tree_root;
1537 1633
1538fail_cleaner: 1634fail_cleaner:
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 2562a273ae18..6b6fdc697f31 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -45,7 +45,7 @@ struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
45struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info, 45struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
46 struct btrfs_key *location, 46 struct btrfs_key *location,
47 const char *name, int namelen); 47 const char *name, int namelen);
48struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info, 48struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
49 struct btrfs_key *location); 49 struct btrfs_key *location);
50struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info, 50struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
51 struct btrfs_key *location); 51 struct btrfs_key *location);
@@ -74,4 +74,10 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
74 extent_submit_bio_hook_t *submit_bio_hook); 74 extent_submit_bio_hook_t *submit_bio_hook);
75int btrfs_congested_async(struct btrfs_fs_info *info, int iodone); 75int btrfs_congested_async(struct btrfs_fs_info *info, int iodone);
76unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info); 76unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info);
77int btrfs_write_tree_block(struct extent_buffer *buf);
78int btrfs_wait_tree_block_writeback(struct extent_buffer *buf);
79int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
80 struct btrfs_fs_info *fs_info);
81int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
82 struct btrfs_fs_info *fs_info);
77#endif 83#endif
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index e63b3b4bed7c..646b9148ca21 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -496,6 +496,23 @@ static int match_extent_ref(struct extent_buffer *leaf,
496 return ret == 0; 496 return ret == 0;
497} 497}
498 498
499/* simple helper to search for an existing extent at a given offset */
500int btrfs_lookup_extent(struct btrfs_root *root, struct btrfs_path *path,
501 u64 start, u64 len)
502{
503 int ret;
504 struct btrfs_key key;
505
506 maybe_lock_mutex(root);
507 key.objectid = start;
508 key.offset = len;
509 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
510 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
511 0, 0);
512 maybe_unlock_mutex(root);
513 return ret;
514}
515
499static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans, 516static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
500 struct btrfs_root *root, 517 struct btrfs_root *root,
501 struct btrfs_path *path, u64 bytenr, 518 struct btrfs_path *path, u64 bytenr,
@@ -1409,7 +1426,7 @@ static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1409} 1426}
1410 1427
1411 1428
1412static int update_pinned_extents(struct btrfs_root *root, 1429int btrfs_update_pinned_extents(struct btrfs_root *root,
1413 u64 bytenr, u64 num, int pin) 1430 u64 bytenr, u64 num, int pin)
1414{ 1431{
1415 u64 len; 1432 u64 len;
@@ -1492,7 +1509,7 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1492 EXTENT_DIRTY); 1509 EXTENT_DIRTY);
1493 if (ret) 1510 if (ret)
1494 break; 1511 break;
1495 update_pinned_extents(root, start, end + 1 - start, 0); 1512 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1496 clear_extent_dirty(unpin, start, end, GFP_NOFS); 1513 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1497 set_extent_dirty(free_space_cache, start, end, GFP_NOFS); 1514 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1498 if (need_resched()) { 1515 if (need_resched()) {
@@ -1538,14 +1555,11 @@ static int finish_current_insert(struct btrfs_trans_handle *trans,
1538 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, 1555 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1539 GFP_NOFS); 1556 GFP_NOFS);
1540 1557
1541 eb = btrfs_find_tree_block(extent_root, ins.objectid, 1558 eb = btrfs_find_create_tree_block(extent_root, ins.objectid,
1542 ins.offset); 1559 ins.offset);
1543 1560
1544 if (!btrfs_buffer_uptodate(eb, trans->transid)) { 1561 if (!btrfs_buffer_uptodate(eb, trans->transid))
1545 mutex_unlock(&extent_root->fs_info->alloc_mutex);
1546 btrfs_read_buffer(eb, trans->transid); 1562 btrfs_read_buffer(eb, trans->transid);
1547 mutex_lock(&extent_root->fs_info->alloc_mutex);
1548 }
1549 1563
1550 btrfs_tree_lock(eb); 1564 btrfs_tree_lock(eb);
1551 level = btrfs_header_level(eb); 1565 level = btrfs_header_level(eb);
@@ -1585,13 +1599,20 @@ static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1585 struct extent_buffer *buf; 1599 struct extent_buffer *buf;
1586 buf = btrfs_find_tree_block(root, bytenr, num_bytes); 1600 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1587 if (buf) { 1601 if (buf) {
1602 /* we can reuse a block if it hasn't been written
1603 * and it is from this transaction. We can't
1604 * reuse anything from the tree log root because
1605 * it has tiny sub-transactions.
1606 */
1588 if (btrfs_buffer_uptodate(buf, 0) && 1607 if (btrfs_buffer_uptodate(buf, 0) &&
1589 btrfs_try_tree_lock(buf)) { 1608 btrfs_try_tree_lock(buf)) {
1590 u64 transid = 1609 u64 transid =
1591 root->fs_info->running_transaction->transid; 1610 root->fs_info->running_transaction->transid;
1592 u64 header_transid = 1611 u64 header_transid =
1593 btrfs_header_generation(buf); 1612 btrfs_header_generation(buf);
1594 if (header_transid == transid && 1613 if (btrfs_header_owner(buf) !=
1614 BTRFS_TREE_LOG_OBJECTID &&
1615 header_transid == transid &&
1595 !btrfs_header_flag(buf, 1616 !btrfs_header_flag(buf,
1596 BTRFS_HEADER_FLAG_WRITTEN)) { 1617 BTRFS_HEADER_FLAG_WRITTEN)) {
1597 clean_tree_block(NULL, root, buf); 1618 clean_tree_block(NULL, root, buf);
@@ -1603,7 +1624,7 @@ static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1603 } 1624 }
1604 free_extent_buffer(buf); 1625 free_extent_buffer(buf);
1605 } 1626 }
1606 update_pinned_extents(root, bytenr, num_bytes, 1); 1627 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1607 } else { 1628 } else {
1608 set_extent_bits(&root->fs_info->pending_del, 1629 set_extent_bits(&root->fs_info->pending_del,
1609 bytenr, bytenr + num_bytes - 1, 1630 bytenr, bytenr + num_bytes - 1,
@@ -1801,7 +1822,7 @@ static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1801 GFP_NOFS); 1822 GFP_NOFS);
1802 if (!test_range_bit(&extent_root->fs_info->extent_ins, 1823 if (!test_range_bit(&extent_root->fs_info->extent_ins,
1803 start, end, EXTENT_LOCKED, 0)) { 1824 start, end, EXTENT_LOCKED, 0)) {
1804 update_pinned_extents(extent_root, start, 1825 btrfs_update_pinned_extents(extent_root, start,
1805 end + 1 - start, 1); 1826 end + 1 - start, 1);
1806 ret = __free_extent(trans, extent_root, 1827 ret = __free_extent(trans, extent_root,
1807 start, end + 1 - start, 1828 start, end + 1 - start,
@@ -1919,6 +1940,12 @@ static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1919 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) { 1940 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1920 last_ptr = &root->fs_info->last_data_alloc; 1941 last_ptr = &root->fs_info->last_data_alloc;
1921 } 1942 }
1943 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
1944 last_ptr = &root->fs_info->last_log_alloc;
1945 if (!last_ptr == 0 && root->fs_info->last_alloc) {
1946 *last_ptr = root->fs_info->last_alloc + empty_cluster;
1947 }
1948 }
1922 1949
1923 if (last_ptr) { 1950 if (last_ptr) {
1924 if (*last_ptr) 1951 if (*last_ptr)
@@ -2268,6 +2295,35 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2268 maybe_unlock_mutex(root); 2295 maybe_unlock_mutex(root);
2269 return ret; 2296 return ret;
2270} 2297}
2298
2299/*
2300 * this is used by the tree logging recovery code. It records that
2301 * an extent has been allocated and makes sure to clear the free
2302 * space cache bits as well
2303 */
2304int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2305 struct btrfs_root *root,
2306 u64 root_objectid, u64 ref_generation,
2307 u64 owner, u64 owner_offset,
2308 struct btrfs_key *ins)
2309{
2310 int ret;
2311 struct btrfs_block_group_cache *block_group;
2312
2313 maybe_lock_mutex(root);
2314 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2315 cache_block_group(root, block_group);
2316
2317 clear_extent_dirty(&root->fs_info->free_space_cache,
2318 ins->objectid, ins->objectid + ins->offset - 1,
2319 GFP_NOFS);
2320 ret = __btrfs_alloc_reserved_extent(trans, root, root_objectid,
2321 ref_generation, owner,
2322 owner_offset, ins);
2323 maybe_unlock_mutex(root);
2324 return ret;
2325}
2326
2271/* 2327/*
2272 * finds a free extent and does all the dirty work required for allocation 2328 * finds a free extent and does all the dirty work required for allocation
2273 * returns the key for the extent through ins, and a tree buffer for 2329 * returns the key for the extent through ins, and a tree buffer for
@@ -2350,9 +2406,8 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2350 return buf; 2406 return buf;
2351} 2407}
2352 2408
2353static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans, 2409int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2354 struct btrfs_root *root, 2410 struct btrfs_root *root, struct extent_buffer *leaf)
2355 struct extent_buffer *leaf)
2356{ 2411{
2357 u64 leaf_owner; 2412 u64 leaf_owner;
2358 u64 leaf_generation; 2413 u64 leaf_generation;
@@ -2402,9 +2457,9 @@ static int noinline drop_leaf_ref_no_cache(struct btrfs_trans_handle *trans,
2402 return 0; 2457 return 0;
2403} 2458}
2404 2459
2405static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans, 2460static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2406 struct btrfs_root *root, 2461 struct btrfs_root *root,
2407 struct btrfs_leaf_ref *ref) 2462 struct btrfs_leaf_ref *ref)
2408{ 2463{
2409 int i; 2464 int i;
2410 int ret; 2465 int ret;
@@ -2512,7 +2567,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2512 btrfs_header_nritems(cur)) 2567 btrfs_header_nritems(cur))
2513 break; 2568 break;
2514 if (*level == 0) { 2569 if (*level == 0) {
2515 ret = drop_leaf_ref_no_cache(trans, root, cur); 2570 ret = btrfs_drop_leaf_ref(trans, root, cur);
2516 BUG_ON(ret); 2571 BUG_ON(ret);
2517 break; 2572 break;
2518 } 2573 }
@@ -2552,7 +2607,7 @@ static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2552 btrfs_node_key_to_cpu(cur, &key, path->slots[*level]); 2607 btrfs_node_key_to_cpu(cur, &key, path->slots[*level]);
2553 ref = btrfs_lookup_leaf_ref(root, bytenr); 2608 ref = btrfs_lookup_leaf_ref(root, bytenr);
2554 if (ref) { 2609 if (ref) {
2555 ret = drop_leaf_ref(trans, root, ref); 2610 ret = cache_drop_leaf_ref(trans, root, ref);
2556 BUG_ON(ret); 2611 BUG_ON(ret);
2557 btrfs_remove_leaf_ref(root, ref); 2612 btrfs_remove_leaf_ref(root, ref);
2558 btrfs_free_leaf_ref(root, ref); 2613 btrfs_free_leaf_ref(root, ref);
@@ -3628,6 +3683,8 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3628 extent_root = root->fs_info->extent_root; 3683 extent_root = root->fs_info->extent_root;
3629 block_group_cache = &root->fs_info->block_group_cache; 3684 block_group_cache = &root->fs_info->block_group_cache;
3630 3685
3686 root->fs_info->last_trans_new_blockgroup = trans->transid;
3687
3631 cache = kzalloc(sizeof(*cache), GFP_NOFS); 3688 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3632 BUG_ON(!cache); 3689 BUG_ON(!cache);
3633 cache->key.objectid = chunk_offset; 3690 cache->key.objectid = chunk_offset;
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index e9e86fbaa243..84ecf3ab8511 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -36,6 +36,8 @@
36#include "btrfs_inode.h" 36#include "btrfs_inode.h"
37#include "ioctl.h" 37#include "ioctl.h"
38#include "print-tree.h" 38#include "print-tree.h"
39#include "tree-log.h"
40#include "locking.h"
39#include "compat.h" 41#include "compat.h"
40 42
41 43
@@ -988,10 +990,27 @@ out_nolock:
988 *ppos = pos; 990 *ppos = pos;
989 991
990 if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 992 if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
991 err = sync_page_range(inode, inode->i_mapping, 993 struct btrfs_trans_handle *trans;
992 start_pos, num_written); 994
995 err = btrfs_fdatawrite_range(inode->i_mapping, start_pos,
996 start_pos + num_written -1,
997 WB_SYNC_NONE);
998 if (err < 0)
999 num_written = err;
1000
1001 err = btrfs_wait_on_page_writeback_range(inode->i_mapping,
1002 start_pos, start_pos + num_written - 1);
993 if (err < 0) 1003 if (err < 0)
994 num_written = err; 1004 num_written = err;
1005
1006 trans = btrfs_start_transaction(root, 1);
1007 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1008 if (ret == 0) {
1009 btrfs_sync_log(trans, root);
1010 btrfs_end_transaction(trans, root);
1011 } else {
1012 btrfs_commit_transaction(trans, root);
1013 }
995 } else if (num_written > 0 && (file->f_flags & O_DIRECT)) { 1014 } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
996#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22) 1015#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,22)
997 do_sync_file_range(file, start_pos, 1016 do_sync_file_range(file, start_pos,
@@ -1019,8 +1038,7 @@ int btrfs_release_file(struct inode * inode, struct file * filp)
1019 return 0; 1038 return 0;
1020} 1039}
1021 1040
1022static int btrfs_sync_file(struct file *file, 1041int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1023 struct dentry *dentry, int datasync)
1024{ 1042{
1025 struct inode *inode = dentry->d_inode; 1043 struct inode *inode = dentry->d_inode;
1026 struct btrfs_root *root = BTRFS_I(inode)->root; 1044 struct btrfs_root *root = BTRFS_I(inode)->root;
@@ -1043,6 +1061,8 @@ static int btrfs_sync_file(struct file *file,
1043 } 1061 }
1044 mutex_unlock(&root->fs_info->trans_mutex); 1062 mutex_unlock(&root->fs_info->trans_mutex);
1045 1063
1064 filemap_fdatawait(inode->i_mapping);
1065
1046 /* 1066 /*
1047 * ok we haven't committed the transaction yet, lets do a commit 1067 * ok we haven't committed the transaction yet, lets do a commit
1048 */ 1068 */
@@ -1054,7 +1074,16 @@ static int btrfs_sync_file(struct file *file,
1054 ret = -ENOMEM; 1074 ret = -ENOMEM;
1055 goto out; 1075 goto out;
1056 } 1076 }
1057 ret = btrfs_commit_transaction(trans, root); 1077
1078 ret = btrfs_log_dentry_safe(trans, root, file->f_dentry);
1079 if (ret < 0)
1080 goto out;
1081 if (ret > 0) {
1082 ret = btrfs_commit_transaction(trans, root);
1083 } else {
1084 btrfs_sync_log(trans, root);
1085 ret = btrfs_end_transaction(trans, root);
1086 }
1058out: 1087out:
1059 return ret > 0 ? EIO : ret; 1088 return ret > 0 ? EIO : ret;
1060} 1089}
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 43d3f2649ca3..65df9d830230 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -46,6 +46,8 @@
46#include "volumes.h" 46#include "volumes.h"
47#include "ordered-data.h" 47#include "ordered-data.h"
48#include "xattr.h" 48#include "xattr.h"
49#include "compat.h"
50#include "tree-log.h"
49 51
50struct btrfs_iget_args { 52struct btrfs_iget_args {
51 u64 ino; 53 u64 ino;
@@ -586,6 +588,7 @@ nocow:
586 &ordered_extent->list); 588 &ordered_extent->list);
587 589
588 btrfs_ordered_update_i_size(inode, ordered_extent); 590 btrfs_ordered_update_i_size(inode, ordered_extent);
591 btrfs_update_inode(trans, root, inode);
589 btrfs_remove_ordered_extent(inode, ordered_extent); 592 btrfs_remove_ordered_extent(inode, ordered_extent);
590 593
591 /* once for us */ 594 /* once for us */
@@ -593,7 +596,6 @@ nocow:
593 /* once for the tree */ 596 /* once for the tree */
594 btrfs_put_ordered_extent(ordered_extent); 597 btrfs_put_ordered_extent(ordered_extent);
595 598
596 btrfs_update_inode(trans, root, inode);
597 btrfs_end_transaction(trans, root); 599 btrfs_end_transaction(trans, root);
598 return 0; 600 return 0;
599} 601}
@@ -1007,7 +1009,8 @@ void btrfs_read_locked_inode(struct inode *inode)
1007 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec); 1009 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1008 1010
1009 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item); 1011 inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1010 inode->i_generation = btrfs_inode_generation(leaf, inode_item); 1012 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1013 inode->i_generation = BTRFS_I(inode)->generation;
1011 inode->i_rdev = 0; 1014 inode->i_rdev = 0;
1012 rdev = btrfs_inode_rdev(leaf, inode_item); 1015 rdev = btrfs_inode_rdev(leaf, inode_item);
1013 1016
@@ -1056,7 +1059,8 @@ make_bad:
1056 make_bad_inode(inode); 1059 make_bad_inode(inode);
1057} 1060}
1058 1061
1059static void fill_inode_item(struct extent_buffer *leaf, 1062static void fill_inode_item(struct btrfs_trans_handle *trans,
1063 struct extent_buffer *leaf,
1060 struct btrfs_inode_item *item, 1064 struct btrfs_inode_item *item,
1061 struct inode *inode) 1065 struct inode *inode)
1062{ 1066{
@@ -1082,7 +1086,8 @@ static void fill_inode_item(struct extent_buffer *leaf,
1082 inode->i_ctime.tv_nsec); 1086 inode->i_ctime.tv_nsec);
1083 1087
1084 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks); 1088 btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1085 btrfs_set_inode_generation(leaf, item, inode->i_generation); 1089 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1090 btrfs_set_inode_transid(leaf, item, trans->transid);
1086 btrfs_set_inode_rdev(leaf, item, inode->i_rdev); 1091 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1087 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags); 1092 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1088 btrfs_set_inode_block_group(leaf, item, 1093 btrfs_set_inode_block_group(leaf, item,
@@ -1112,7 +1117,7 @@ int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1112 inode_item = btrfs_item_ptr(leaf, path->slots[0], 1117 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1113 struct btrfs_inode_item); 1118 struct btrfs_inode_item);
1114 1119
1115 fill_inode_item(leaf, inode_item, inode); 1120 fill_inode_item(trans, leaf, inode_item, inode);
1116 btrfs_mark_buffer_dirty(leaf); 1121 btrfs_mark_buffer_dirty(leaf);
1117 btrfs_set_inode_last_trans(trans, inode); 1122 btrfs_set_inode_last_trans(trans, inode);
1118 ret = 0; 1123 ret = 0;
@@ -1122,14 +1127,12 @@ failed:
1122} 1127}
1123 1128
1124 1129
1125static int btrfs_unlink_trans(struct btrfs_trans_handle *trans, 1130int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1126 struct btrfs_root *root, 1131 struct btrfs_root *root,
1127 struct inode *dir, 1132 struct inode *dir, struct inode *inode,
1128 struct dentry *dentry) 1133 const char *name, int name_len)
1129{ 1134{
1130 struct btrfs_path *path; 1135 struct btrfs_path *path;
1131 const char *name = dentry->d_name.name;
1132 int name_len = dentry->d_name.len;
1133 int ret = 0; 1136 int ret = 0;
1134 struct extent_buffer *leaf; 1137 struct extent_buffer *leaf;
1135 struct btrfs_dir_item *di; 1138 struct btrfs_dir_item *di;
@@ -1160,13 +1163,12 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1160 btrfs_release_path(root, path); 1163 btrfs_release_path(root, path);
1161 1164
1162 ret = btrfs_del_inode_ref(trans, root, name, name_len, 1165 ret = btrfs_del_inode_ref(trans, root, name, name_len,
1163 dentry->d_inode->i_ino, 1166 inode->i_ino,
1164 dentry->d_parent->d_inode->i_ino, &index); 1167 dir->i_ino, &index);
1165 if (ret) { 1168 if (ret) {
1166 printk("failed to delete reference to %.*s, " 1169 printk("failed to delete reference to %.*s, "
1167 "inode %lu parent %lu\n", name_len, name, 1170 "inode %lu parent %lu\n", name_len, name,
1168 dentry->d_inode->i_ino, 1171 inode->i_ino, dir->i_ino);
1169 dentry->d_parent->d_inode->i_ino);
1170 goto err; 1172 goto err;
1171 } 1173 }
1172 1174
@@ -1183,21 +1185,25 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1183 ret = btrfs_delete_one_dir_name(trans, root, path, di); 1185 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1184 btrfs_release_path(root, path); 1186 btrfs_release_path(root, path);
1185 1187
1186 dentry->d_inode->i_ctime = dir->i_ctime; 1188 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1189 inode, dir->i_ino);
1190 BUG_ON(ret);
1191
1192 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1193 dir, index);
1194 BUG_ON(ret);
1187err: 1195err:
1188 btrfs_free_path(path); 1196 btrfs_free_path(path);
1189 if (!ret) { 1197 if (ret)
1190 btrfs_i_size_write(dir, dir->i_size - name_len * 2); 1198 goto out;
1191 dir->i_mtime = dir->i_ctime = CURRENT_TIME; 1199
1192 btrfs_update_inode(trans, root, dir); 1200 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1193#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) 1201 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1194 dentry->d_inode->i_nlink--; 1202 btrfs_update_inode(trans, root, dir);
1195#else 1203 btrfs_drop_nlink(inode);
1196 drop_nlink(dentry->d_inode); 1204 ret = btrfs_update_inode(trans, root, inode);
1197#endif 1205 dir->i_sb->s_dirt = 1;
1198 ret = btrfs_update_inode(trans, root, dentry->d_inode); 1206out:
1199 dir->i_sb->s_dirt = 1;
1200 }
1201 return ret; 1207 return ret;
1202} 1208}
1203 1209
@@ -1218,7 +1224,8 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1218 trans = btrfs_start_transaction(root, 1); 1224 trans = btrfs_start_transaction(root, 1);
1219 1225
1220 btrfs_set_trans_block_group(trans, dir); 1226 btrfs_set_trans_block_group(trans, dir);
1221 ret = btrfs_unlink_trans(trans, root, dir, dentry); 1227 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1228 dentry->d_name.name, dentry->d_name.len);
1222 1229
1223 if (inode->i_nlink == 0) 1230 if (inode->i_nlink == 0)
1224 ret = btrfs_orphan_add(trans, inode); 1231 ret = btrfs_orphan_add(trans, inode);
@@ -1256,7 +1263,8 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1256 goto fail_trans; 1263 goto fail_trans;
1257 1264
1258 /* now the directory is empty */ 1265 /* now the directory is empty */
1259 err = btrfs_unlink_trans(trans, root, dir, dentry); 1266 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1267 dentry->d_name.name, dentry->d_name.len);
1260 if (!err) { 1268 if (!err) {
1261 btrfs_i_size_write(inode, 0); 1269 btrfs_i_size_write(inode, 0);
1262 } 1270 }
@@ -1283,10 +1291,10 @@ fail:
1283 * min_type is the minimum key type to truncate down to. If set to 0, this 1291 * min_type is the minimum key type to truncate down to. If set to 0, this
1284 * will kill all the items on this inode, including the INODE_ITEM_KEY. 1292 * will kill all the items on this inode, including the INODE_ITEM_KEY.
1285 */ 1293 */
1286static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans, 1294noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1287 struct btrfs_root *root, 1295 struct btrfs_root *root,
1288 struct inode *inode, 1296 struct inode *inode,
1289 u32 min_type) 1297 u64 new_size, u32 min_type)
1290{ 1298{
1291 int ret; 1299 int ret;
1292 struct btrfs_path *path; 1300 struct btrfs_path *path;
@@ -1307,7 +1315,9 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1307 int extent_type = -1; 1315 int extent_type = -1;
1308 u64 mask = root->sectorsize - 1; 1316 u64 mask = root->sectorsize - 1;
1309 1317
1310 btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1); 1318 if (root->ref_cows)
1319 btrfs_drop_extent_cache(inode,
1320 new_size & (~mask), (u64)-1);
1311 path = btrfs_alloc_path(); 1321 path = btrfs_alloc_path();
1312 path->reada = -1; 1322 path->reada = -1;
1313 BUG_ON(!path); 1323 BUG_ON(!path);
@@ -1324,7 +1334,13 @@ search_again:
1324 goto error; 1334 goto error;
1325 } 1335 }
1326 if (ret > 0) { 1336 if (ret > 0) {
1327 BUG_ON(path->slots[0] == 0); 1337 /* there are no items in the tree for us to truncate, we're
1338 * done
1339 */
1340 if (path->slots[0] == 0) {
1341 ret = 0;
1342 goto error;
1343 }
1328 path->slots[0]--; 1344 path->slots[0]--;
1329 } 1345 }
1330 1346
@@ -1358,10 +1374,10 @@ search_again:
1358 } 1374 }
1359 if (found_type == BTRFS_CSUM_ITEM_KEY) { 1375 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1360 ret = btrfs_csum_truncate(trans, root, path, 1376 ret = btrfs_csum_truncate(trans, root, path,
1361 inode->i_size); 1377 new_size);
1362 BUG_ON(ret); 1378 BUG_ON(ret);
1363 } 1379 }
1364 if (item_end < inode->i_size) { 1380 if (item_end < new_size) {
1365 if (found_type == BTRFS_DIR_ITEM_KEY) { 1381 if (found_type == BTRFS_DIR_ITEM_KEY) {
1366 found_type = BTRFS_INODE_ITEM_KEY; 1382 found_type = BTRFS_INODE_ITEM_KEY;
1367 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) { 1383 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
@@ -1378,7 +1394,7 @@ search_again:
1378 btrfs_set_key_type(&key, found_type); 1394 btrfs_set_key_type(&key, found_type);
1379 goto next; 1395 goto next;
1380 } 1396 }
1381 if (found_key.offset >= inode->i_size) 1397 if (found_key.offset >= new_size)
1382 del_item = 1; 1398 del_item = 1;
1383 else 1399 else
1384 del_item = 0; 1400 del_item = 0;
@@ -1394,7 +1410,7 @@ search_again:
1394 if (!del_item) { 1410 if (!del_item) {
1395 u64 orig_num_bytes = 1411 u64 orig_num_bytes =
1396 btrfs_file_extent_num_bytes(leaf, fi); 1412 btrfs_file_extent_num_bytes(leaf, fi);
1397 extent_num_bytes = inode->i_size - 1413 extent_num_bytes = new_size -
1398 found_key.offset + root->sectorsize - 1; 1414 found_key.offset + root->sectorsize - 1;
1399 extent_num_bytes = extent_num_bytes & 1415 extent_num_bytes = extent_num_bytes &
1400 ~((u64)root->sectorsize - 1); 1416 ~((u64)root->sectorsize - 1);
@@ -1402,7 +1418,7 @@ search_again:
1402 extent_num_bytes); 1418 extent_num_bytes);
1403 num_dec = (orig_num_bytes - 1419 num_dec = (orig_num_bytes -
1404 extent_num_bytes); 1420 extent_num_bytes);
1405 if (extent_start != 0) 1421 if (root->ref_cows && extent_start != 0)
1406 dec_i_blocks(inode, num_dec); 1422 dec_i_blocks(inode, num_dec);
1407 btrfs_mark_buffer_dirty(leaf); 1423 btrfs_mark_buffer_dirty(leaf);
1408 } else { 1424 } else {
@@ -1413,22 +1429,29 @@ search_again:
1413 num_dec = btrfs_file_extent_num_bytes(leaf, fi); 1429 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1414 if (extent_start != 0) { 1430 if (extent_start != 0) {
1415 found_extent = 1; 1431 found_extent = 1;
1416 dec_i_blocks(inode, num_dec); 1432 if (root->ref_cows)
1433 dec_i_blocks(inode, num_dec);
1434 }
1435 if (root->ref_cows) {
1436 root_gen =
1437 btrfs_header_generation(leaf);
1417 } 1438 }
1418 root_gen = btrfs_header_generation(leaf);
1419 root_owner = btrfs_header_owner(leaf); 1439 root_owner = btrfs_header_owner(leaf);
1420 } 1440 }
1421 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { 1441 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1422 if (!del_item) { 1442 if (!del_item) {
1423 u32 newsize = inode->i_size - found_key.offset; 1443 u32 size = new_size - found_key.offset;
1424 dec_i_blocks(inode, item_end + 1 - 1444
1425 found_key.offset - newsize); 1445 if (root->ref_cows) {
1426 newsize = 1446 dec_i_blocks(inode, item_end + 1 -
1427 btrfs_file_extent_calc_inline_size(newsize); 1447 found_key.offset - size);
1448 }
1449 size =
1450 btrfs_file_extent_calc_inline_size(size);
1428 ret = btrfs_truncate_item(trans, root, path, 1451 ret = btrfs_truncate_item(trans, root, path,
1429 newsize, 1); 1452 size, 1);
1430 BUG_ON(ret); 1453 BUG_ON(ret);
1431 } else { 1454 } else if (root->ref_cows) {
1432 dec_i_blocks(inode, item_end + 1 - 1455 dec_i_blocks(inode, item_end + 1 -
1433 found_key.offset); 1456 found_key.offset);
1434 } 1457 }
@@ -1666,7 +1689,7 @@ void btrfs_delete_inode(struct inode *inode)
1666 trans = btrfs_start_transaction(root, 1); 1689 trans = btrfs_start_transaction(root, 1);
1667 1690
1668 btrfs_set_trans_block_group(trans, inode); 1691 btrfs_set_trans_block_group(trans, inode);
1669 ret = btrfs_truncate_in_trans(trans, root, inode, 0); 1692 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
1670 if (ret) { 1693 if (ret) {
1671 btrfs_orphan_del(NULL, inode); 1694 btrfs_orphan_del(NULL, inode);
1672 goto no_delete_lock; 1695 goto no_delete_lock;
@@ -1753,15 +1776,20 @@ static int fixup_tree_root_location(struct btrfs_root *root,
1753 return 0; 1776 return 0;
1754} 1777}
1755 1778
1756static int btrfs_init_locked_inode(struct inode *inode, void *p) 1779static noinline void init_btrfs_i(struct inode *inode)
1757{ 1780{
1758 struct btrfs_iget_args *args = p; 1781 struct btrfs_inode *bi = BTRFS_I(inode);
1759 inode->i_ino = args->ino; 1782
1760 BTRFS_I(inode)->root = args->root; 1783 bi->i_acl = NULL;
1761 BTRFS_I(inode)->delalloc_bytes = 0; 1784 bi->i_default_acl = NULL;
1762 inode->i_mapping->writeback_index = 0; 1785
1763 BTRFS_I(inode)->disk_i_size = 0; 1786 bi->generation = 0;
1764 BTRFS_I(inode)->index_cnt = (u64)-1; 1787 bi->last_trans = 0;
1788 bi->logged_trans = 0;
1789 bi->delalloc_bytes = 0;
1790 bi->disk_i_size = 0;
1791 bi->flags = 0;
1792 bi->index_cnt = (u64)-1;
1765 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); 1793 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1766 extent_io_tree_init(&BTRFS_I(inode)->io_tree, 1794 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1767 inode->i_mapping, GFP_NOFS); 1795 inode->i_mapping, GFP_NOFS);
@@ -1771,6 +1799,15 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p)
1771 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); 1799 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1772 mutex_init(&BTRFS_I(inode)->csum_mutex); 1800 mutex_init(&BTRFS_I(inode)->csum_mutex);
1773 mutex_init(&BTRFS_I(inode)->extent_mutex); 1801 mutex_init(&BTRFS_I(inode)->extent_mutex);
1802 mutex_init(&BTRFS_I(inode)->log_mutex);
1803}
1804
1805static int btrfs_init_locked_inode(struct inode *inode, void *p)
1806{
1807 struct btrfs_iget_args *args = p;
1808 inode->i_ino = args->ino;
1809 init_btrfs_i(inode);
1810 BTRFS_I(inode)->root = args->root;
1774 return 0; 1811 return 0;
1775} 1812}
1776 1813
@@ -2263,21 +2300,10 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2263 * btrfs_get_inode_index_count has an explanation for the magic 2300 * btrfs_get_inode_index_count has an explanation for the magic
2264 * number 2301 * number
2265 */ 2302 */
2303 init_btrfs_i(inode);
2266 BTRFS_I(inode)->index_cnt = 2; 2304 BTRFS_I(inode)->index_cnt = 2;
2267
2268 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2269 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2270 inode->i_mapping, GFP_NOFS);
2271 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2272 inode->i_mapping, GFP_NOFS);
2273 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2274 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2275 mutex_init(&BTRFS_I(inode)->csum_mutex);
2276 mutex_init(&BTRFS_I(inode)->extent_mutex);
2277 BTRFS_I(inode)->delalloc_bytes = 0;
2278 inode->i_mapping->writeback_index = 0;
2279 BTRFS_I(inode)->disk_i_size = 0;
2280 BTRFS_I(inode)->root = root; 2305 BTRFS_I(inode)->root = root;
2306 BTRFS_I(inode)->generation = trans->transid;
2281 2307
2282 if (mode & S_IFDIR) 2308 if (mode & S_IFDIR)
2283 owner = 0; 2309 owner = 0;
@@ -2290,7 +2316,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2290 new_inode_group = group; 2316 new_inode_group = group;
2291 } 2317 }
2292 BTRFS_I(inode)->block_group = new_inode_group; 2318 BTRFS_I(inode)->block_group = new_inode_group;
2293 BTRFS_I(inode)->flags = 0;
2294 2319
2295 key[0].objectid = objectid; 2320 key[0].objectid = objectid;
2296 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY); 2321 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
@@ -2318,7 +2343,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2318 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 2343 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2319 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0], 2344 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2320 struct btrfs_inode_item); 2345 struct btrfs_inode_item);
2321 fill_inode_item(path->nodes[0], inode_item, inode); 2346 fill_inode_item(trans, path->nodes[0], inode_item, inode);
2322 2347
2323 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, 2348 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2324 struct btrfs_inode_ref); 2349 struct btrfs_inode_ref);
@@ -2349,38 +2374,34 @@ static inline u8 btrfs_inode_type(struct inode *inode)
2349 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT]; 2374 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2350} 2375}
2351 2376
2352static int btrfs_add_link(struct btrfs_trans_handle *trans, 2377int btrfs_add_link(struct btrfs_trans_handle *trans,
2353 struct dentry *dentry, struct inode *inode, 2378 struct inode *parent_inode, struct inode *inode,
2354 int add_backref, u64 index) 2379 const char *name, int name_len, int add_backref, u64 index)
2355{ 2380{
2356 int ret; 2381 int ret;
2357 struct btrfs_key key; 2382 struct btrfs_key key;
2358 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root; 2383 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
2359 struct inode *parent_inode = dentry->d_parent->d_inode;
2360 2384
2361 key.objectid = inode->i_ino; 2385 key.objectid = inode->i_ino;
2362 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY); 2386 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2363 key.offset = 0; 2387 key.offset = 0;
2364 2388
2365 ret = btrfs_insert_dir_item(trans, root, 2389 ret = btrfs_insert_dir_item(trans, root, name, name_len,
2366 dentry->d_name.name, dentry->d_name.len, 2390 parent_inode->i_ino,
2367 dentry->d_parent->d_inode->i_ino,
2368 &key, btrfs_inode_type(inode), 2391 &key, btrfs_inode_type(inode),
2369 index); 2392 index);
2370 if (ret == 0) { 2393 if (ret == 0) {
2371 if (add_backref) { 2394 if (add_backref) {
2372 ret = btrfs_insert_inode_ref(trans, root, 2395 ret = btrfs_insert_inode_ref(trans, root,
2373 dentry->d_name.name, 2396 name, name_len,
2374 dentry->d_name.len, 2397 inode->i_ino,
2375 inode->i_ino, 2398 parent_inode->i_ino,
2376 parent_inode->i_ino, 2399 index);
2377 index);
2378 } 2400 }
2379 btrfs_i_size_write(parent_inode, parent_inode->i_size + 2401 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2380 dentry->d_name.len * 2); 2402 name_len * 2);
2381 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME; 2403 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2382 ret = btrfs_update_inode(trans, root, 2404 ret = btrfs_update_inode(trans, root, parent_inode);
2383 dentry->d_parent->d_inode);
2384 } 2405 }
2385 return ret; 2406 return ret;
2386} 2407}
@@ -2389,7 +2410,9 @@ static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2389 struct dentry *dentry, struct inode *inode, 2410 struct dentry *dentry, struct inode *inode,
2390 int backref, u64 index) 2411 int backref, u64 index)
2391{ 2412{
2392 int err = btrfs_add_link(trans, dentry, inode, backref, index); 2413 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2414 inode, dentry->d_name.name,
2415 dentry->d_name.len, backref, index);
2393 if (!err) { 2416 if (!err) {
2394 d_instantiate(dentry, inode); 2417 d_instantiate(dentry, inode);
2395 return 0; 2418 return 0;
@@ -2513,19 +2536,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
2513 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 2536 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2514 inode->i_fop = &btrfs_file_operations; 2537 inode->i_fop = &btrfs_file_operations;
2515 inode->i_op = &btrfs_file_inode_operations; 2538 inode->i_op = &btrfs_file_inode_operations;
2516 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2517 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2518 inode->i_mapping, GFP_NOFS);
2519 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2520 inode->i_mapping, GFP_NOFS);
2521 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2522 mutex_init(&BTRFS_I(inode)->csum_mutex);
2523 mutex_init(&BTRFS_I(inode)->extent_mutex);
2524 BTRFS_I(inode)->delalloc_bytes = 0;
2525 BTRFS_I(inode)->disk_i_size = 0;
2526 inode->i_mapping->writeback_index = 0;
2527 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; 2539 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2528 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2529 } 2540 }
2530 dir->i_sb->s_dirt = 1; 2541 dir->i_sb->s_dirt = 1;
2531 btrfs_update_inode_block_group(trans, inode); 2542 btrfs_update_inode_block_group(trans, inode);
@@ -2556,11 +2567,7 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2556 if (inode->i_nlink == 0) 2567 if (inode->i_nlink == 0)
2557 return -ENOENT; 2568 return -ENOENT;
2558 2569
2559#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) 2570 btrfs_inc_nlink(inode);
2560 inode->i_nlink++;
2561#else
2562 inc_nlink(inode);
2563#endif
2564 err = btrfs_check_free_space(root, 1, 0); 2571 err = btrfs_check_free_space(root, 1, 0);
2565 if (err) 2572 if (err)
2566 goto fail; 2573 goto fail;
@@ -2650,7 +2657,9 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2650 if (err) 2657 if (err)
2651 goto out_fail; 2658 goto out_fail;
2652 2659
2653 err = btrfs_add_link(trans, dentry, inode, 0, index); 2660 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2661 inode, dentry->d_name.name,
2662 dentry->d_name.len, 0, index);
2654 if (err) 2663 if (err)
2655 goto out_fail; 2664 goto out_fail;
2656 2665
@@ -3221,7 +3230,7 @@ static void btrfs_truncate(struct inode *inode)
3221 if (ret) 3230 if (ret)
3222 goto out; 3231 goto out;
3223 /* FIXME, add redo link to tree so we don't leak on crash */ 3232 /* FIXME, add redo link to tree so we don't leak on crash */
3224 ret = btrfs_truncate_in_trans(trans, root, inode, 3233 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3225 BTRFS_EXTENT_DATA_KEY); 3234 BTRFS_EXTENT_DATA_KEY);
3226 btrfs_update_inode(trans, root, inode); 3235 btrfs_update_inode(trans, root, inode);
3227 3236
@@ -3304,6 +3313,7 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
3304 if (!ei) 3313 if (!ei)
3305 return NULL; 3314 return NULL;
3306 ei->last_trans = 0; 3315 ei->last_trans = 0;
3316 ei->logged_trans = 0;
3307 btrfs_ordered_inode_tree_init(&ei->ordered_tree); 3317 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3308 ei->i_acl = BTRFS_ACL_NOT_CACHED; 3318 ei->i_acl = BTRFS_ACL_NOT_CACHED;
3309 ei->i_default_acl = BTRFS_ACL_NOT_CACHED; 3319 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
@@ -3463,31 +3473,39 @@ static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3463 3473
3464 btrfs_set_trans_block_group(trans, new_dir); 3474 btrfs_set_trans_block_group(trans, new_dir);
3465 3475
3466 old_dentry->d_inode->i_nlink++; 3476 btrfs_inc_nlink(old_dentry->d_inode);
3467 old_dir->i_ctime = old_dir->i_mtime = ctime; 3477 old_dir->i_ctime = old_dir->i_mtime = ctime;
3468 new_dir->i_ctime = new_dir->i_mtime = ctime; 3478 new_dir->i_ctime = new_dir->i_mtime = ctime;
3469 old_inode->i_ctime = ctime; 3479 old_inode->i_ctime = ctime;
3470 3480
3471 ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry); 3481 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
3482 old_dentry->d_name.name,
3483 old_dentry->d_name.len);
3472 if (ret) 3484 if (ret)
3473 goto out_fail; 3485 goto out_fail;
3474 3486
3475 if (new_inode) { 3487 if (new_inode) {
3476 new_inode->i_ctime = CURRENT_TIME; 3488 new_inode->i_ctime = CURRENT_TIME;
3477 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry); 3489 ret = btrfs_unlink_inode(trans, root, new_dir,
3490 new_dentry->d_inode,
3491 new_dentry->d_name.name,
3492 new_dentry->d_name.len);
3478 if (ret) 3493 if (ret)
3479 goto out_fail; 3494 goto out_fail;
3480 if (new_inode->i_nlink == 0) { 3495 if (new_inode->i_nlink == 0) {
3481 ret = btrfs_orphan_add(trans, new_inode); 3496 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
3482 if (ret) 3497 if (ret)
3483 goto out_fail; 3498 goto out_fail;
3484 } 3499 }
3500
3485 } 3501 }
3486 ret = btrfs_set_inode_index(new_dir, old_inode, &index); 3502 ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3487 if (ret) 3503 if (ret)
3488 goto out_fail; 3504 goto out_fail;
3489 3505
3490 ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index); 3506 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
3507 old_inode, new_dentry->d_name.name,
3508 new_dentry->d_name.len, 1, index);
3491 if (ret) 3509 if (ret)
3492 goto out_fail; 3510 goto out_fail;
3493 3511
@@ -3577,19 +3595,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3577 inode->i_mapping->backing_dev_info = &root->fs_info->bdi; 3595 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3578 inode->i_fop = &btrfs_file_operations; 3596 inode->i_fop = &btrfs_file_operations;
3579 inode->i_op = &btrfs_file_inode_operations; 3597 inode->i_op = &btrfs_file_inode_operations;
3580 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3581 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3582 inode->i_mapping, GFP_NOFS);
3583 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3584 inode->i_mapping, GFP_NOFS);
3585 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3586 mutex_init(&BTRFS_I(inode)->csum_mutex);
3587 mutex_init(&BTRFS_I(inode)->extent_mutex);
3588 BTRFS_I(inode)->delalloc_bytes = 0;
3589 BTRFS_I(inode)->disk_i_size = 0;
3590 inode->i_mapping->writeback_index = 0;
3591 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops; 3598 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3592 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3593 } 3599 }
3594 dir->i_sb->s_dirt = 1; 3600 dir->i_sb->s_dirt = 1;
3595 btrfs_update_inode_block_group(trans, inode); 3601 btrfs_update_inode_block_group(trans, inode);
@@ -3691,6 +3697,7 @@ static struct file_operations btrfs_dir_file_operations = {
3691 .compat_ioctl = btrfs_ioctl, 3697 .compat_ioctl = btrfs_ioctl,
3692#endif 3698#endif
3693 .release = btrfs_release_file, 3699 .release = btrfs_release_file,
3700 .fsync = btrfs_sync_file,
3694}; 3701};
3695 3702
3696static struct extent_io_ops btrfs_extent_io_ops = { 3703static struct extent_io_ops btrfs_extent_io_ops = {
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 36726696e58b..e3984f902e71 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -202,8 +202,9 @@ again:
202 memcpy(&found_key, &key, sizeof(key)); 202 memcpy(&found_key, &key, sizeof(key));
203 key.offset++; 203 key.offset++;
204 btrfs_release_path(root, path); 204 btrfs_release_path(root, path);
205 dead_root = btrfs_read_fs_root_no_radix(root->fs_info, 205 dead_root =
206 &found_key); 206 btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
207 &found_key);
207 if (IS_ERR(dead_root)) { 208 if (IS_ERR(dead_root)) {
208 ret = PTR_ERR(dead_root); 209 ret = PTR_ERR(dead_root);
209 goto err; 210 goto err;
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index eff3ad72991b..49c4f5b40ed6 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -25,6 +25,7 @@
25#include "transaction.h" 25#include "transaction.h"
26#include "locking.h" 26#include "locking.h"
27#include "ref-cache.h" 27#include "ref-cache.h"
28#include "tree-log.h"
28 29
29static int total_trans = 0; 30static int total_trans = 0;
30extern struct kmem_cache *btrfs_trans_handle_cachep; 31extern struct kmem_cache *btrfs_trans_handle_cachep;
@@ -57,6 +58,7 @@ static noinline int join_transaction(struct btrfs_root *root)
57 root->fs_info->generation++; 58 root->fs_info->generation++;
58 root->fs_info->last_alloc = 0; 59 root->fs_info->last_alloc = 0;
59 root->fs_info->last_data_alloc = 0; 60 root->fs_info->last_data_alloc = 0;
61 root->fs_info->last_log_alloc = 0;
60 cur_trans->num_writers = 1; 62 cur_trans->num_writers = 1;
61 cur_trans->num_joined = 0; 63 cur_trans->num_joined = 0;
62 cur_trans->transid = root->fs_info->generation; 64 cur_trans->transid = root->fs_info->generation;
@@ -83,7 +85,7 @@ static noinline int join_transaction(struct btrfs_root *root)
83 return 0; 85 return 0;
84} 86}
85 87
86static noinline int record_root_in_trans(struct btrfs_root *root) 88noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
87{ 89{
88 struct btrfs_dirty_root *dirty; 90 struct btrfs_dirty_root *dirty;
89 u64 running_trans_id = root->fs_info->running_transaction->transid; 91 u64 running_trans_id = root->fs_info->running_transaction->transid;
@@ -151,7 +153,7 @@ static void wait_current_trans(struct btrfs_root *root)
151 } 153 }
152} 154}
153 155
154struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 156static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
155 int num_blocks, int wait) 157 int num_blocks, int wait)
156{ 158{
157 struct btrfs_trans_handle *h = 159 struct btrfs_trans_handle *h =
@@ -164,7 +166,7 @@ struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
164 ret = join_transaction(root); 166 ret = join_transaction(root);
165 BUG_ON(ret); 167 BUG_ON(ret);
166 168
167 record_root_in_trans(root); 169 btrfs_record_root_in_trans(root);
168 h->transid = root->fs_info->running_transaction->transid; 170 h->transid = root->fs_info->running_transaction->transid;
169 h->transaction = root->fs_info->running_transaction; 171 h->transaction = root->fs_info->running_transaction;
170 h->blocks_reserved = num_blocks; 172 h->blocks_reserved = num_blocks;
@@ -456,6 +458,8 @@ static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
456 BUG_ON(!root->ref_tree); 458 BUG_ON(!root->ref_tree);
457 dirty = root->dirty_root; 459 dirty = root->dirty_root;
458 460
461 btrfs_free_log(trans, root);
462
459 if (root->commit_root == root->node) { 463 if (root->commit_root == root->node) {
460 WARN_ON(root->node->start != 464 WARN_ON(root->node->start !=
461 btrfs_root_bytenr(&root->root_item)); 465 btrfs_root_bytenr(&root->root_item));
@@ -600,7 +604,7 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
600 num_bytes -= btrfs_root_used(&dirty->root->root_item); 604 num_bytes -= btrfs_root_used(&dirty->root->root_item);
601 bytes_used = btrfs_root_used(&root->root_item); 605 bytes_used = btrfs_root_used(&root->root_item);
602 if (num_bytes) { 606 if (num_bytes) {
603 record_root_in_trans(root); 607 btrfs_record_root_in_trans(root);
604 btrfs_set_root_used(&root->root_item, 608 btrfs_set_root_used(&root->root_item,
605 bytes_used - num_bytes); 609 bytes_used - num_bytes);
606 } 610 }
@@ -745,7 +749,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
745 int ret; 749 int ret;
746 750
747 INIT_LIST_HEAD(&dirty_fs_roots); 751 INIT_LIST_HEAD(&dirty_fs_roots);
748
749 mutex_lock(&root->fs_info->trans_mutex); 752 mutex_lock(&root->fs_info->trans_mutex);
750 if (trans->transaction->in_commit) { 753 if (trans->transaction->in_commit) {
751 cur_trans = trans->transaction; 754 cur_trans = trans->transaction;
@@ -821,10 +824,30 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
821 824
822 WARN_ON(cur_trans != trans->transaction); 825 WARN_ON(cur_trans != trans->transaction);
823 826
827 /* btrfs_commit_tree_roots is responsible for getting the
828 * various roots consistent with each other. Every pointer
829 * in the tree of tree roots has to point to the most up to date
830 * root for every subvolume and other tree. So, we have to keep
831 * the tree logging code from jumping in and changing any
832 * of the trees.
833 *
834 * At this point in the commit, there can't be any tree-log
835 * writers, but a little lower down we drop the trans mutex
836 * and let new people in. By holding the tree_log_mutex
837 * from now until after the super is written, we avoid races
838 * with the tree-log code.
839 */
840 mutex_lock(&root->fs_info->tree_log_mutex);
841
824 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix, 842 ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
825 &dirty_fs_roots); 843 &dirty_fs_roots);
826 BUG_ON(ret); 844 BUG_ON(ret);
827 845
846 /* add_dirty_roots gets rid of all the tree log roots, it is now
847 * safe to free the root of tree log roots
848 */
849 btrfs_free_log_root_tree(trans, root->fs_info);
850
828 ret = btrfs_commit_tree_roots(trans, root); 851 ret = btrfs_commit_tree_roots(trans, root);
829 BUG_ON(ret); 852 BUG_ON(ret);
830 853
@@ -843,6 +866,12 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
843 chunk_root->node->start); 866 chunk_root->node->start);
844 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy, 867 btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
845 btrfs_header_level(chunk_root->node)); 868 btrfs_header_level(chunk_root->node));
869
870 if (!root->fs_info->log_root_recovering) {
871 btrfs_set_super_log_root(&root->fs_info->super_copy, 0);
872 btrfs_set_super_log_root_level(&root->fs_info->super_copy, 0);
873 }
874
846 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy, 875 memcpy(&root->fs_info->super_for_commit, &root->fs_info->super_copy,
847 sizeof(root->fs_info->super_copy)); 876 sizeof(root->fs_info->super_copy));
848 877
@@ -857,6 +886,12 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
857 BUG_ON(ret); 886 BUG_ON(ret);
858 write_ctree_super(trans, root); 887 write_ctree_super(trans, root);
859 888
889 /*
890 * the super is written, we can safely allow the tree-loggers
891 * to go about their business
892 */
893 mutex_unlock(&root->fs_info->tree_log_mutex);
894
860 btrfs_finish_extent_commit(trans, root, pinned_copy); 895 btrfs_finish_extent_commit(trans, root, pinned_copy);
861 mutex_lock(&root->fs_info->trans_mutex); 896 mutex_lock(&root->fs_info->trans_mutex);
862 897
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index 598baa312417..cc63650d60d6 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -98,4 +98,5 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
98int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 98int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
99 struct btrfs_root *root); 99 struct btrfs_root *root);
100void btrfs_throttle(struct btrfs_root *root); 100void btrfs_throttle(struct btrfs_root *root);
101int btrfs_record_root_in_trans(struct btrfs_root *root);
101#endif 102#endif
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index cc2650b06952..b3bb5bbad76e 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -81,12 +81,12 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
81 memcpy(&key, &root->defrag_progress, sizeof(key)); 81 memcpy(&key, &root->defrag_progress, sizeof(key));
82 } 82 }
83 83
84 path->lowest_level = 1;
85 path->keep_locks = 1; 84 path->keep_locks = 1;
86 if (cache_only) 85 if (cache_only)
87 min_trans = root->defrag_trans_start; 86 min_trans = root->defrag_trans_start;
88 87
89 ret = btrfs_search_forward(root, &key, path, cache_only, min_trans); 88 ret = btrfs_search_forward(root, &key, NULL, path,
89 cache_only, min_trans);
90 if (ret < 0) 90 if (ret < 0)
91 goto out; 91 goto out;
92 if (ret > 0) { 92 if (ret > 0) {
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
new file mode 100644
index 000000000000..d1ce8314b948
--- /dev/null
+++ b/fs/btrfs/tree-log.c
@@ -0,0 +1,2804 @@
1/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/sched.h>
20#include "ctree.h"
21#include "transaction.h"
22#include "disk-io.h"
23#include "locking.h"
24#include "print-tree.h"
25#include "compat.h"
26
27/* magic values for the inode_only field in btrfs_log_inode:
28 *
29 * LOG_INODE_ALL means to log everything
30 * LOG_INODE_EXISTS means to log just enough to recreate the inode
31 * during log replay
32 */
33#define LOG_INODE_ALL 0
34#define LOG_INODE_EXISTS 1
35
36/*
37 * stages for the tree walking. The first
38 * stage (0) is to only pin down the blocks we find
39 * the second stage (1) is to make sure that all the inodes
40 * we find in the log are created in the subvolume.
41 *
42 * The last stage is to deal with directories and links and extents
43 * and all the other fun semantics
44 */
45#define LOG_WALK_PIN_ONLY 0
46#define LOG_WALK_REPLAY_INODES 1
47#define LOG_WALK_REPLAY_ALL 2
48
49static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
50 struct btrfs_root *root, struct inode *inode,
51 int inode_only);
52
53/*
54 * tree logging is a special write ahead log used to make sure that
55 * fsyncs and O_SYNCs can happen without doing full tree commits.
56 *
57 * Full tree commits are expensive because they require commonly
58 * modified blocks to be recowed, creating many dirty pages in the
59 * extent tree an 4x-6x higher write load than ext3.
60 *
61 * Instead of doing a tree commit on every fsync, we use the
62 * key ranges and transaction ids to find items for a given file or directory
63 * that have changed in this transaction. Those items are copied into
64 * a special tree (one per subvolume root), that tree is written to disk
65 * and then the fsync is considered complete.
66 *
67 * After a crash, items are copied out of the log-tree back into the
68 * subvolume tree. Any file data extents found are recorded in the extent
69 * allocation tree, and the log-tree freed.
70 *
71 * The log tree is read three times, once to pin down all the extents it is
72 * using in ram and once, once to create all the inodes logged in the tree
73 * and once to do all the other items.
74 */
75
76/*
77 * btrfs_add_log_tree adds a new per-subvolume log tree into the
78 * tree of log tree roots. This must be called with a tree log transaction
79 * running (see start_log_trans).
80 */
81int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root)
83{
84 struct btrfs_key key;
85 struct btrfs_root_item root_item;
86 struct btrfs_inode_item *inode_item;
87 struct extent_buffer *leaf;
88 struct btrfs_root *new_root = root;
89 int ret;
90 u64 objectid = root->root_key.objectid;
91
92 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
93 BTRFS_TREE_LOG_OBJECTID,
94 0, 0, 0, 0, 0);
95 if (IS_ERR(leaf)) {
96 ret = PTR_ERR(leaf);
97 return ret;
98 }
99
100 btrfs_set_header_nritems(leaf, 0);
101 btrfs_set_header_level(leaf, 0);
102 btrfs_set_header_bytenr(leaf, leaf->start);
103 btrfs_set_header_generation(leaf, trans->transid);
104 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
105
106 write_extent_buffer(leaf, root->fs_info->fsid,
107 (unsigned long)btrfs_header_fsid(leaf),
108 BTRFS_FSID_SIZE);
109 btrfs_mark_buffer_dirty(leaf);
110
111 inode_item = &root_item.inode;
112 memset(inode_item, 0, sizeof(*inode_item));
113 inode_item->generation = cpu_to_le64(1);
114 inode_item->size = cpu_to_le64(3);
115 inode_item->nlink = cpu_to_le32(1);
116 inode_item->nblocks = cpu_to_le64(1);
117 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
118
119 btrfs_set_root_bytenr(&root_item, leaf->start);
120 btrfs_set_root_level(&root_item, 0);
121 btrfs_set_root_refs(&root_item, 0);
122 btrfs_set_root_used(&root_item, 0);
123
124 memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
125 root_item.drop_level = 0;
126
127 btrfs_tree_unlock(leaf);
128 free_extent_buffer(leaf);
129 leaf = NULL;
130
131 btrfs_set_root_dirid(&root_item, 0);
132
133 key.objectid = BTRFS_TREE_LOG_OBJECTID;
134 key.offset = objectid;
135 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
136 ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
137 &root_item);
138 if (ret)
139 goto fail;
140
141 new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
142 &key);
143 BUG_ON(!new_root);
144
145 WARN_ON(root->log_root);
146 root->log_root = new_root;
147
148 /*
149 * log trees do not get reference counted because they go away
150 * before a real commit is actually done. They do store pointers
151 * to file data extents, and those reference counts still get
152 * updated (along with back refs to the log tree).
153 */
154 new_root->ref_cows = 0;
155 new_root->last_trans = trans->transid;
156fail:
157 return ret;
158}
159
160/*
161 * start a sub transaction and setup the log tree
162 * this increments the log tree writer count to make the people
163 * syncing the tree wait for us to finish
164 */
165static int start_log_trans(struct btrfs_trans_handle *trans,
166 struct btrfs_root *root)
167{
168 int ret;
169 mutex_lock(&root->fs_info->tree_log_mutex);
170 if (!root->fs_info->log_root_tree) {
171 ret = btrfs_init_log_root_tree(trans, root->fs_info);
172 BUG_ON(ret);
173 }
174 if (!root->log_root) {
175 ret = btrfs_add_log_tree(trans, root);
176 BUG_ON(ret);
177 }
178 atomic_inc(&root->fs_info->tree_log_writers);
179 root->fs_info->tree_log_batch++;
180 mutex_unlock(&root->fs_info->tree_log_mutex);
181 return 0;
182}
183
184/*
185 * returns 0 if there was a log transaction running and we were able
186 * to join, or returns -ENOENT if there were not transactions
187 * in progress
188 */
189static int join_running_log_trans(struct btrfs_root *root)
190{
191 int ret = -ENOENT;
192
193 smp_mb();
194 if (!root->log_root)
195 return -ENOENT;
196
197 mutex_lock(&root->fs_info->tree_log_mutex);
198 if (root->log_root) {
199 ret = 0;
200 atomic_inc(&root->fs_info->tree_log_writers);
201 root->fs_info->tree_log_batch++;
202 }
203 mutex_unlock(&root->fs_info->tree_log_mutex);
204 return ret;
205}
206
207/*
208 * indicate we're done making changes to the log tree
209 * and wake up anyone waiting to do a sync
210 */
211static int end_log_trans(struct btrfs_root *root)
212{
213 atomic_dec(&root->fs_info->tree_log_writers);
214 smp_mb();
215 if (waitqueue_active(&root->fs_info->tree_log_wait))
216 wake_up(&root->fs_info->tree_log_wait);
217 return 0;
218}
219
220
221/*
222 * the walk control struct is used to pass state down the chain when
223 * processing the log tree. The stage field tells us which part
224 * of the log tree processing we are currently doing. The others
225 * are state fields used for that specific part
226 */
227struct walk_control {
228 /* should we free the extent on disk when done? This is used
229 * at transaction commit time while freeing a log tree
230 */
231 int free;
232
233 /* should we write out the extent buffer? This is used
234 * while flushing the log tree to disk during a sync
235 */
236 int write;
237
238 /* should we wait for the extent buffer io to finish? Also used
239 * while flushing the log tree to disk for a sync
240 */
241 int wait;
242
243 /* pin only walk, we record which extents on disk belong to the
244 * log trees
245 */
246 int pin;
247
248 /* what stage of the replay code we're currently in */
249 int stage;
250
251 /* the root we are currently replaying */
252 struct btrfs_root *replay_dest;
253
254 /* the trans handle for the current replay */
255 struct btrfs_trans_handle *trans;
256
257 /* the function that gets used to process blocks we find in the
258 * tree. Note the extent_buffer might not be up to date when it is
259 * passed in, and it must be checked or read if you need the data
260 * inside it
261 */
262 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
263 struct walk_control *wc, u64 gen);
264};
265
266/*
267 * process_func used to pin down extents, write them or wait on them
268 */
269static int process_one_buffer(struct btrfs_root *log,
270 struct extent_buffer *eb,
271 struct walk_control *wc, u64 gen)
272{
273 if (wc->pin) {
274 mutex_lock(&log->fs_info->alloc_mutex);
275 btrfs_update_pinned_extents(log->fs_info->extent_root,
276 eb->start, eb->len, 1);
277 mutex_unlock(&log->fs_info->alloc_mutex);
278 }
279
280 if (btrfs_buffer_uptodate(eb, gen)) {
281 if (wc->write)
282 btrfs_write_tree_block(eb);
283 if (wc->wait)
284 btrfs_wait_tree_block_writeback(eb);
285 }
286 return 0;
287}
288
289/*
290 * Item overwrite used by replay and tree logging. eb, slot and key all refer
291 * to the src data we are copying out.
292 *
293 * root is the tree we are copying into, and path is a scratch
294 * path for use in this function (it should be released on entry and
295 * will be released on exit).
296 *
297 * If the key is already in the destination tree the existing item is
298 * overwritten. If the existing item isn't big enough, it is extended.
299 * If it is too large, it is truncated.
300 *
301 * If the key isn't in the destination yet, a new item is inserted.
302 */
303static noinline int overwrite_item(struct btrfs_trans_handle *trans,
304 struct btrfs_root *root,
305 struct btrfs_path *path,
306 struct extent_buffer *eb, int slot,
307 struct btrfs_key *key)
308{
309 int ret;
310 u32 item_size;
311 u64 saved_i_size = 0;
312 int save_old_i_size = 0;
313 unsigned long src_ptr;
314 unsigned long dst_ptr;
315 int overwrite_root = 0;
316
317 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
318 overwrite_root = 1;
319
320 item_size = btrfs_item_size_nr(eb, slot);
321 src_ptr = btrfs_item_ptr_offset(eb, slot);
322
323 /* look for the key in the destination tree */
324 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
325 if (ret == 0) {
326 char *src_copy;
327 char *dst_copy;
328 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
329 path->slots[0]);
330 if (dst_size != item_size)
331 goto insert;
332
333 if (item_size == 0) {
334 btrfs_release_path(root, path);
335 return 0;
336 }
337 dst_copy = kmalloc(item_size, GFP_NOFS);
338 src_copy = kmalloc(item_size, GFP_NOFS);
339
340 read_extent_buffer(eb, src_copy, src_ptr, item_size);
341
342 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
343 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
344 item_size);
345 ret = memcmp(dst_copy, src_copy, item_size);
346
347 kfree(dst_copy);
348 kfree(src_copy);
349 /*
350 * they have the same contents, just return, this saves
351 * us from cowing blocks in the destination tree and doing
352 * extra writes that may not have been done by a previous
353 * sync
354 */
355 if (ret == 0) {
356 btrfs_release_path(root, path);
357 return 0;
358 }
359
360 }
361insert:
362 btrfs_release_path(root, path);
363 /* try to insert the key into the destination tree */
364 ret = btrfs_insert_empty_item(trans, root, path,
365 key, item_size);
366
367 /* make sure any existing item is the correct size */
368 if (ret == -EEXIST) {
369 u32 found_size;
370 found_size = btrfs_item_size_nr(path->nodes[0],
371 path->slots[0]);
372 if (found_size > item_size) {
373 btrfs_truncate_item(trans, root, path, item_size, 1);
374 } else if (found_size < item_size) {
375 ret = btrfs_del_item(trans, root,
376 path);
377 BUG_ON(ret);
378
379 btrfs_release_path(root, path);
380 ret = btrfs_insert_empty_item(trans,
381 root, path, key, item_size);
382 BUG_ON(ret);
383 }
384 } else if (ret) {
385 BUG();
386 }
387 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
388 path->slots[0]);
389
390 /* don't overwrite an existing inode if the generation number
391 * was logged as zero. This is done when the tree logging code
392 * is just logging an inode to make sure it exists after recovery.
393 *
394 * Also, don't overwrite i_size on directories during replay.
395 * log replay inserts and removes directory items based on the
396 * state of the tree found in the subvolume, and i_size is modified
397 * as it goes
398 */
399 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
400 struct btrfs_inode_item *src_item;
401 struct btrfs_inode_item *dst_item;
402
403 src_item = (struct btrfs_inode_item *)src_ptr;
404 dst_item = (struct btrfs_inode_item *)dst_ptr;
405
406 if (btrfs_inode_generation(eb, src_item) == 0)
407 goto no_copy;
408
409 if (overwrite_root &&
410 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
411 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
412 save_old_i_size = 1;
413 saved_i_size = btrfs_inode_size(path->nodes[0],
414 dst_item);
415 }
416 }
417
418 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
419 src_ptr, item_size);
420
421 if (save_old_i_size) {
422 struct btrfs_inode_item *dst_item;
423 dst_item = (struct btrfs_inode_item *)dst_ptr;
424 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
425 }
426
427 /* make sure the generation is filled in */
428 if (key->type == BTRFS_INODE_ITEM_KEY) {
429 struct btrfs_inode_item *dst_item;
430 dst_item = (struct btrfs_inode_item *)dst_ptr;
431 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
432 btrfs_set_inode_generation(path->nodes[0], dst_item,
433 trans->transid);
434 }
435 }
436no_copy:
437 btrfs_mark_buffer_dirty(path->nodes[0]);
438 btrfs_release_path(root, path);
439 return 0;
440}
441
442/*
443 * simple helper to read an inode off the disk from a given root
444 * This can only be called for subvolume roots and not for the log
445 */
446static noinline struct inode *read_one_inode(struct btrfs_root *root,
447 u64 objectid)
448{
449 struct inode *inode;
450 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
451 if (inode->i_state & I_NEW) {
452 BTRFS_I(inode)->root = root;
453 BTRFS_I(inode)->location.objectid = objectid;
454 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
455 BTRFS_I(inode)->location.offset = 0;
456 btrfs_read_locked_inode(inode);
457 unlock_new_inode(inode);
458
459 }
460 if (is_bad_inode(inode)) {
461 iput(inode);
462 inode = NULL;
463 }
464 return inode;
465}
466
467/* replays a single extent in 'eb' at 'slot' with 'key' into the
468 * subvolume 'root'. path is released on entry and should be released
469 * on exit.
470 *
471 * extents in the log tree have not been allocated out of the extent
472 * tree yet. So, this completes the allocation, taking a reference
473 * as required if the extent already exists or creating a new extent
474 * if it isn't in the extent allocation tree yet.
475 *
476 * The extent is inserted into the file, dropping any existing extents
477 * from the file that overlap the new one.
478 */
479static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
480 struct btrfs_root *root,
481 struct btrfs_path *path,
482 struct extent_buffer *eb, int slot,
483 struct btrfs_key *key)
484{
485 int found_type;
486 u64 mask = root->sectorsize - 1;
487 u64 extent_end;
488 u64 alloc_hint;
489 u64 start = key->offset;
490 struct btrfs_file_extent_item *item;
491 struct inode *inode = NULL;
492 unsigned long size;
493 int ret = 0;
494
495 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
496 found_type = btrfs_file_extent_type(eb, item);
497
498 if (found_type == BTRFS_FILE_EXTENT_REG)
499 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
500 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
501 size = btrfs_file_extent_inline_len(eb,
502 btrfs_item_nr(eb, slot));
503 extent_end = (start + size + mask) & ~mask;
504 } else {
505 ret = 0;
506 goto out;
507 }
508
509 inode = read_one_inode(root, key->objectid);
510 if (!inode) {
511 ret = -EIO;
512 goto out;
513 }
514
515 /*
516 * first check to see if we already have this extent in the
517 * file. This must be done before the btrfs_drop_extents run
518 * so we don't try to drop this extent.
519 */
520 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
521 start, 0);
522
523 if (ret == 0 && found_type == BTRFS_FILE_EXTENT_REG) {
524 struct btrfs_file_extent_item cmp1;
525 struct btrfs_file_extent_item cmp2;
526 struct btrfs_file_extent_item *existing;
527 struct extent_buffer *leaf;
528
529 leaf = path->nodes[0];
530 existing = btrfs_item_ptr(leaf, path->slots[0],
531 struct btrfs_file_extent_item);
532
533 read_extent_buffer(eb, &cmp1, (unsigned long)item,
534 sizeof(cmp1));
535 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
536 sizeof(cmp2));
537
538 /*
539 * we already have a pointer to this exact extent,
540 * we don't have to do anything
541 */
542 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
543 btrfs_release_path(root, path);
544 goto out;
545 }
546 }
547 btrfs_release_path(root, path);
548
549 /* drop any overlapping extents */
550 ret = btrfs_drop_extents(trans, root, inode,
551 start, extent_end, start, &alloc_hint);
552 BUG_ON(ret);
553
554 BUG_ON(ret);
555 if (found_type == BTRFS_FILE_EXTENT_REG) {
556 struct btrfs_key ins;
557
558 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
559 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
560 ins.type = BTRFS_EXTENT_ITEM_KEY;
561
562 /* insert the extent pointer in the file */
563 ret = overwrite_item(trans, root, path, eb, slot, key);
564 BUG_ON(ret);
565
566 /*
567 * is this extent already allocated in the extent
568 * allocation tree? If so, just add a reference
569 */
570 ret = btrfs_lookup_extent(root, path, ins.objectid, ins.offset);
571 btrfs_release_path(root, path);
572 if (ret == 0) {
573 ret = btrfs_inc_extent_ref(trans, root,
574 ins.objectid, ins.offset,
575 root->root_key.objectid,
576 trans->transid, key->objectid, start);
577 } else {
578 /*
579 * insert the extent pointer in the extent
580 * allocation tree
581 */
582 ret = btrfs_alloc_logged_extent(trans, root,
583 root->root_key.objectid,
584 trans->transid, key->objectid,
585 start, &ins);
586 BUG_ON(ret);
587 }
588 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
589 /* inline extents are easy, we just overwrite them */
590 ret = overwrite_item(trans, root, path, eb, slot, key);
591 BUG_ON(ret);
592 }
593 /* btrfs_drop_extents changes i_blocks, update it here */
594 inode->i_blocks += (extent_end - start) >> 9;
595 btrfs_update_inode(trans, root, inode);
596out:
597 if (inode)
598 iput(inode);
599 return ret;
600}
601
602/*
603 * when cleaning up conflicts between the directory names in the
604 * subvolume, directory names in the log and directory names in the
605 * inode back references, we may have to unlink inodes from directories.
606 *
607 * This is a helper function to do the unlink of a specific directory
608 * item
609 */
610static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
611 struct btrfs_root *root,
612 struct btrfs_path *path,
613 struct inode *dir,
614 struct btrfs_dir_item *di)
615{
616 struct inode *inode;
617 char *name;
618 int name_len;
619 struct extent_buffer *leaf;
620 struct btrfs_key location;
621 int ret;
622
623 leaf = path->nodes[0];
624
625 btrfs_dir_item_key_to_cpu(leaf, di, &location);
626 name_len = btrfs_dir_name_len(leaf, di);
627 name = kmalloc(name_len, GFP_NOFS);
628 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
629 btrfs_release_path(root, path);
630
631 inode = read_one_inode(root, location.objectid);
632 BUG_ON(!inode);
633
634 btrfs_inc_nlink(inode);
635 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
636 kfree(name);
637
638 iput(inode);
639 return ret;
640}
641
642/*
643 * helper function to see if a given name and sequence number found
644 * in an inode back reference are already in a directory and correctly
645 * point to this inode
646 */
647static noinline int inode_in_dir(struct btrfs_root *root,
648 struct btrfs_path *path,
649 u64 dirid, u64 objectid, u64 index,
650 const char *name, int name_len)
651{
652 struct btrfs_dir_item *di;
653 struct btrfs_key location;
654 int match = 0;
655
656 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
657 index, name, name_len, 0);
658 if (di && !IS_ERR(di)) {
659 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
660 if (location.objectid != objectid)
661 goto out;
662 } else
663 goto out;
664 btrfs_release_path(root, path);
665
666 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
667 if (di && !IS_ERR(di)) {
668 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
669 if (location.objectid != objectid)
670 goto out;
671 } else
672 goto out;
673 match = 1;
674out:
675 btrfs_release_path(root, path);
676 return match;
677}
678
679/*
680 * helper function to check a log tree for a named back reference in
681 * an inode. This is used to decide if a back reference that is
682 * found in the subvolume conflicts with what we find in the log.
683 *
684 * inode backreferences may have multiple refs in a single item,
685 * during replay we process one reference at a time, and we don't
686 * want to delete valid links to a file from the subvolume if that
687 * link is also in the log.
688 */
689static noinline int backref_in_log(struct btrfs_root *log,
690 struct btrfs_key *key,
691 char *name, int namelen)
692{
693 struct btrfs_path *path;
694 struct btrfs_inode_ref *ref;
695 unsigned long ptr;
696 unsigned long ptr_end;
697 unsigned long name_ptr;
698 int found_name_len;
699 int item_size;
700 int ret;
701 int match = 0;
702
703 path = btrfs_alloc_path();
704 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
705 if (ret != 0)
706 goto out;
707
708 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
709 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
710 ptr_end = ptr + item_size;
711 while (ptr < ptr_end) {
712 ref = (struct btrfs_inode_ref *)ptr;
713 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
714 if (found_name_len == namelen) {
715 name_ptr = (unsigned long)(ref + 1);
716 ret = memcmp_extent_buffer(path->nodes[0], name,
717 name_ptr, namelen);
718 if (ret == 0) {
719 match = 1;
720 goto out;
721 }
722 }
723 ptr = (unsigned long)(ref + 1) + found_name_len;
724 }
725out:
726 btrfs_free_path(path);
727 return match;
728}
729
730
731/*
732 * replay one inode back reference item found in the log tree.
733 * eb, slot and key refer to the buffer and key found in the log tree.
734 * root is the destination we are replaying into, and path is for temp
735 * use by this function. (it should be released on return).
736 */
737static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
738 struct btrfs_root *root,
739 struct btrfs_root *log,
740 struct btrfs_path *path,
741 struct extent_buffer *eb, int slot,
742 struct btrfs_key *key)
743{
744 struct inode *dir;
745 int ret;
746 struct btrfs_key location;
747 struct btrfs_inode_ref *ref;
748 struct btrfs_dir_item *di;
749 struct inode *inode;
750 char *name;
751 int namelen;
752 unsigned long ref_ptr;
753 unsigned long ref_end;
754
755 location.objectid = key->objectid;
756 location.type = BTRFS_INODE_ITEM_KEY;
757 location.offset = 0;
758
759 /*
760 * it is possible that we didn't log all the parent directories
761 * for a given inode. If we don't find the dir, just don't
762 * copy the back ref in. The link count fixup code will take
763 * care of the rest
764 */
765 dir = read_one_inode(root, key->offset);
766 if (!dir)
767 return -ENOENT;
768
769 inode = read_one_inode(root, key->objectid);
770 BUG_ON(!dir);
771
772 ref_ptr = btrfs_item_ptr_offset(eb, slot);
773 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
774
775again:
776 ref = (struct btrfs_inode_ref *)ref_ptr;
777
778 namelen = btrfs_inode_ref_name_len(eb, ref);
779 name = kmalloc(namelen, GFP_NOFS);
780 BUG_ON(!name);
781
782 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
783
784 /* if we already have a perfect match, we're done */
785 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
786 btrfs_inode_ref_index(eb, ref),
787 name, namelen)) {
788 goto out;
789 }
790
791 /*
792 * look for a conflicting back reference in the metadata.
793 * if we find one we have to unlink that name of the file
794 * before we add our new link. Later on, we overwrite any
795 * existing back reference, and we don't want to create
796 * dangling pointers in the directory.
797 */
798conflict_again:
799 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
800 if (ret == 0) {
801 char *victim_name;
802 int victim_name_len;
803 struct btrfs_inode_ref *victim_ref;
804 unsigned long ptr;
805 unsigned long ptr_end;
806 struct extent_buffer *leaf = path->nodes[0];
807
808 /* are we trying to overwrite a back ref for the root directory
809 * if so, just jump out, we're done
810 */
811 if (key->objectid == key->offset)
812 goto out_nowrite;
813
814 /* check all the names in this back reference to see
815 * if they are in the log. if so, we allow them to stay
816 * otherwise they must be unlinked as a conflict
817 */
818 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
819 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
820 while(ptr < ptr_end) {
821 victim_ref = (struct btrfs_inode_ref *)ptr;
822 victim_name_len = btrfs_inode_ref_name_len(leaf,
823 victim_ref);
824 victim_name = kmalloc(victim_name_len, GFP_NOFS);
825 BUG_ON(!victim_name);
826
827 read_extent_buffer(leaf, victim_name,
828 (unsigned long)(victim_ref + 1),
829 victim_name_len);
830
831 if (!backref_in_log(log, key, victim_name,
832 victim_name_len)) {
833 btrfs_inc_nlink(inode);
834 btrfs_release_path(root, path);
835 ret = btrfs_unlink_inode(trans, root, dir,
836 inode, victim_name,
837 victim_name_len);
838 kfree(victim_name);
839 btrfs_release_path(root, path);
840 goto conflict_again;
841 }
842 kfree(victim_name);
843 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
844 }
845 BUG_ON(ret);
846 }
847 btrfs_release_path(root, path);
848
849 /* look for a conflicting sequence number */
850 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
851 btrfs_inode_ref_index(eb, ref),
852 name, namelen, 0);
853 if (di && !IS_ERR(di)) {
854 ret = drop_one_dir_item(trans, root, path, dir, di);
855 BUG_ON(ret);
856 }
857 btrfs_release_path(root, path);
858
859
860 /* look for a conflicting name */
861 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
862 name, namelen, 0);
863 if (di && !IS_ERR(di)) {
864 ret = drop_one_dir_item(trans, root, path, dir, di);
865 BUG_ON(ret);
866 }
867 btrfs_release_path(root, path);
868
869 /* insert our name */
870 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
871 btrfs_inode_ref_index(eb, ref));
872 BUG_ON(ret);
873
874 btrfs_update_inode(trans, root, inode);
875
876out:
877 ref_ptr = (unsigned long)(ref + 1) + namelen;
878 kfree(name);
879 if (ref_ptr < ref_end)
880 goto again;
881
882 /* finally write the back reference in the inode */
883 ret = overwrite_item(trans, root, path, eb, slot, key);
884 BUG_ON(ret);
885
886out_nowrite:
887 btrfs_release_path(root, path);
888 iput(dir);
889 iput(inode);
890 return 0;
891}
892
893/*
894 * replay one csum item from the log tree into the subvolume 'root'
895 * eb, slot and key all refer to the log tree
896 * path is for temp use by this function and should be released on return
897 *
898 * This copies the checksums out of the log tree and inserts them into
899 * the subvolume. Any existing checksums for this range in the file
900 * are overwritten, and new items are added where required.
901 *
902 * We keep this simple by reusing the btrfs_ordered_sum code from
903 * the data=ordered mode. This basically means making a copy
904 * of all the checksums in ram, which we have to do anyway for kmap
905 * rules.
906 *
907 * The copy is then sent down to btrfs_csum_file_blocks, which
908 * does all the hard work of finding existing items in the file
909 * or adding new ones.
910 */
911static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
912 struct btrfs_root *root,
913 struct btrfs_path *path,
914 struct extent_buffer *eb, int slot,
915 struct btrfs_key *key)
916{
917 int ret;
918 u32 item_size = btrfs_item_size_nr(eb, slot);
919 u64 cur_offset;
920 unsigned long file_bytes;
921 struct btrfs_ordered_sum *sums;
922 struct btrfs_sector_sum *sector_sum;
923 struct inode *inode;
924 unsigned long ptr;
925
926 file_bytes = (item_size / BTRFS_CRC32_SIZE) * root->sectorsize;
927 inode = read_one_inode(root, key->objectid);
928 if (!inode) {
929 return -EIO;
930 }
931
932 sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
933 if (!sums) {
934 iput(inode);
935 return -ENOMEM;
936 }
937
938 INIT_LIST_HEAD(&sums->list);
939 sums->len = file_bytes;
940 sums->file_offset = key->offset;
941
942 /*
943 * copy all the sums into the ordered sum struct
944 */
945 sector_sum = sums->sums;
946 cur_offset = key->offset;
947 ptr = btrfs_item_ptr_offset(eb, slot);
948 while(item_size > 0) {
949 sector_sum->offset = cur_offset;
950 read_extent_buffer(eb, &sector_sum->sum, ptr, BTRFS_CRC32_SIZE);
951 sector_sum++;
952 item_size -= BTRFS_CRC32_SIZE;
953 ptr += BTRFS_CRC32_SIZE;
954 cur_offset += root->sectorsize;
955 }
956
957 /* let btrfs_csum_file_blocks add them into the file */
958 ret = btrfs_csum_file_blocks(trans, root, inode, sums);
959 BUG_ON(ret);
960 kfree(sums);
961 iput(inode);
962
963 return 0;
964}
965/*
966 * There are a few corners where the link count of the file can't
967 * be properly maintained during replay. So, instead of adding
968 * lots of complexity to the log code, we just scan the backrefs
969 * for any file that has been through replay.
970 *
971 * The scan will update the link count on the inode to reflect the
972 * number of back refs found. If it goes down to zero, the iput
973 * will free the inode.
974 */
975static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root,
977 struct inode *inode)
978{
979 struct btrfs_path *path;
980 int ret;
981 struct btrfs_key key;
982 u64 nlink = 0;
983 unsigned long ptr;
984 unsigned long ptr_end;
985 int name_len;
986
987 key.objectid = inode->i_ino;
988 key.type = BTRFS_INODE_REF_KEY;
989 key.offset = (u64)-1;
990
991 path = btrfs_alloc_path();
992
993 while(1) {
994 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
995 if (ret < 0)
996 break;
997 if (ret > 0) {
998 if (path->slots[0] == 0)
999 break;
1000 path->slots[0]--;
1001 }
1002 btrfs_item_key_to_cpu(path->nodes[0], &key,
1003 path->slots[0]);
1004 if (key.objectid != inode->i_ino ||
1005 key.type != BTRFS_INODE_REF_KEY)
1006 break;
1007 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1008 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1009 path->slots[0]);
1010 while(ptr < ptr_end) {
1011 struct btrfs_inode_ref *ref;
1012
1013 ref = (struct btrfs_inode_ref *)ptr;
1014 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1015 ref);
1016 ptr = (unsigned long)(ref + 1) + name_len;
1017 nlink++;
1018 }
1019
1020 if (key.offset == 0)
1021 break;
1022 key.offset--;
1023 btrfs_release_path(root, path);
1024 }
1025 btrfs_free_path(path);
1026 if (nlink != inode->i_nlink) {
1027 inode->i_nlink = nlink;
1028 btrfs_update_inode(trans, root, inode);
1029 }
1030
1031 return 0;
1032}
1033
1034static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1035 struct btrfs_root *root,
1036 struct btrfs_path *path)
1037{
1038 int ret;
1039 struct btrfs_key key;
1040 struct inode *inode;
1041
1042 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1043 key.type = BTRFS_ORPHAN_ITEM_KEY;
1044 key.offset = (u64)-1;
1045 while(1) {
1046 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1047 if (ret < 0)
1048 break;
1049
1050 if (ret == 1) {
1051 if (path->slots[0] == 0)
1052 break;
1053 path->slots[0]--;
1054 }
1055
1056 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1057 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1058 key.type != BTRFS_ORPHAN_ITEM_KEY)
1059 break;
1060
1061 ret = btrfs_del_item(trans, root, path);
1062 BUG_ON(ret);
1063
1064 btrfs_release_path(root, path);
1065 inode = read_one_inode(root, key.offset);
1066 BUG_ON(!inode);
1067
1068 ret = fixup_inode_link_count(trans, root, inode);
1069 BUG_ON(ret);
1070
1071 iput(inode);
1072
1073 if (key.offset == 0)
1074 break;
1075 key.offset--;
1076 }
1077 btrfs_release_path(root, path);
1078 return 0;
1079}
1080
1081
1082/*
1083 * record a given inode in the fixup dir so we can check its link
1084 * count when replay is done. The link count is incremented here
1085 * so the inode won't go away until we check it
1086 */
1087static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1088 struct btrfs_root *root,
1089 struct btrfs_path *path,
1090 u64 objectid)
1091{
1092 struct btrfs_key key;
1093 int ret = 0;
1094 struct inode *inode;
1095
1096 inode = read_one_inode(root, objectid);
1097 BUG_ON(!inode);
1098
1099 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1100 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1101 key.offset = objectid;
1102
1103 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1104
1105 btrfs_release_path(root, path);
1106 if (ret == 0) {
1107 btrfs_inc_nlink(inode);
1108 btrfs_update_inode(trans, root, inode);
1109 } else if (ret == -EEXIST) {
1110 ret = 0;
1111 } else {
1112 BUG();
1113 }
1114 iput(inode);
1115
1116 return ret;
1117}
1118
1119/*
1120 * when replaying the log for a directory, we only insert names
1121 * for inodes that actually exist. This means an fsync on a directory
1122 * does not implicitly fsync all the new files in it
1123 */
1124static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1125 struct btrfs_root *root,
1126 struct btrfs_path *path,
1127 u64 dirid, u64 index,
1128 char *name, int name_len, u8 type,
1129 struct btrfs_key *location)
1130{
1131 struct inode *inode;
1132 struct inode *dir;
1133 int ret;
1134
1135 inode = read_one_inode(root, location->objectid);
1136 if (!inode)
1137 return -ENOENT;
1138
1139 dir = read_one_inode(root, dirid);
1140 if (!dir) {
1141 iput(inode);
1142 return -EIO;
1143 }
1144 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1145
1146 /* FIXME, put inode into FIXUP list */
1147
1148 iput(inode);
1149 iput(dir);
1150 return ret;
1151}
1152
1153/*
1154 * take a single entry in a log directory item and replay it into
1155 * the subvolume.
1156 *
1157 * if a conflicting item exists in the subdirectory already,
1158 * the inode it points to is unlinked and put into the link count
1159 * fix up tree.
1160 *
1161 * If a name from the log points to a file or directory that does
1162 * not exist in the FS, it is skipped. fsyncs on directories
1163 * do not force down inodes inside that directory, just changes to the
1164 * names or unlinks in a directory.
1165 */
1166static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1167 struct btrfs_root *root,
1168 struct btrfs_path *path,
1169 struct extent_buffer *eb,
1170 struct btrfs_dir_item *di,
1171 struct btrfs_key *key)
1172{
1173 char *name;
1174 int name_len;
1175 struct btrfs_dir_item *dst_di;
1176 struct btrfs_key found_key;
1177 struct btrfs_key log_key;
1178 struct inode *dir;
1179 struct inode *inode;
1180 u8 log_type;
1181 int ret;
1182
1183 dir = read_one_inode(root, key->objectid);
1184 BUG_ON(!dir);
1185
1186 name_len = btrfs_dir_name_len(eb, di);
1187 name = kmalloc(name_len, GFP_NOFS);
1188 log_type = btrfs_dir_type(eb, di);
1189 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1190 name_len);
1191
1192 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1193 if (key->type == BTRFS_DIR_ITEM_KEY) {
1194 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1195 name, name_len, 1);
1196 }
1197 else if (key->type == BTRFS_DIR_INDEX_KEY) {
1198 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1199 key->objectid,
1200 key->offset, name,
1201 name_len, 1);
1202 } else {
1203 BUG();
1204 }
1205 if (!dst_di || IS_ERR(dst_di)) {
1206 /* we need a sequence number to insert, so we only
1207 * do inserts for the BTRFS_DIR_INDEX_KEY types
1208 */
1209 if (key->type != BTRFS_DIR_INDEX_KEY)
1210 goto out;
1211 goto insert;
1212 }
1213
1214 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1215 /* the existing item matches the logged item */
1216 if (found_key.objectid == log_key.objectid &&
1217 found_key.type == log_key.type &&
1218 found_key.offset == log_key.offset &&
1219 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1220 goto out;
1221 }
1222
1223 /*
1224 * don't drop the conflicting directory entry if the inode
1225 * for the new entry doesn't exist
1226 */
1227 inode = read_one_inode(root, log_key.objectid);
1228 if (!inode)
1229 goto out;
1230
1231 iput(inode);
1232 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1233 BUG_ON(ret);
1234
1235 if (key->type == BTRFS_DIR_INDEX_KEY)
1236 goto insert;
1237out:
1238 btrfs_release_path(root, path);
1239 kfree(name);
1240 iput(dir);
1241 return 0;
1242
1243insert:
1244 btrfs_release_path(root, path);
1245 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1246 name, name_len, log_type, &log_key);
1247
1248 if (ret && ret != -ENOENT)
1249 BUG();
1250 goto out;
1251}
1252
1253/*
1254 * find all the names in a directory item and reconcile them into
1255 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1256 * one name in a directory item, but the same code gets used for
1257 * both directory index types
1258 */
1259static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1260 struct btrfs_root *root,
1261 struct btrfs_path *path,
1262 struct extent_buffer *eb, int slot,
1263 struct btrfs_key *key)
1264{
1265 int ret;
1266 u32 item_size = btrfs_item_size_nr(eb, slot);
1267 struct btrfs_dir_item *di;
1268 int name_len;
1269 unsigned long ptr;
1270 unsigned long ptr_end;
1271
1272 ptr = btrfs_item_ptr_offset(eb, slot);
1273 ptr_end = ptr + item_size;
1274 while(ptr < ptr_end) {
1275 di = (struct btrfs_dir_item *)ptr;
1276 name_len = btrfs_dir_name_len(eb, di);
1277 ret = replay_one_name(trans, root, path, eb, di, key);
1278 BUG_ON(ret);
1279 ptr = (unsigned long)(di + 1);
1280 ptr += name_len;
1281 }
1282 return 0;
1283}
1284
1285/*
1286 * directory replay has two parts. There are the standard directory
1287 * items in the log copied from the subvolume, and range items
1288 * created in the log while the subvolume was logged.
1289 *
1290 * The range items tell us which parts of the key space the log
1291 * is authoritative for. During replay, if a key in the subvolume
1292 * directory is in a logged range item, but not actually in the log
1293 * that means it was deleted from the directory before the fsync
1294 * and should be removed.
1295 */
1296static noinline int find_dir_range(struct btrfs_root *root,
1297 struct btrfs_path *path,
1298 u64 dirid, int key_type,
1299 u64 *start_ret, u64 *end_ret)
1300{
1301 struct btrfs_key key;
1302 u64 found_end;
1303 struct btrfs_dir_log_item *item;
1304 int ret;
1305 int nritems;
1306
1307 if (*start_ret == (u64)-1)
1308 return 1;
1309
1310 key.objectid = dirid;
1311 key.type = key_type;
1312 key.offset = *start_ret;
1313
1314 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1315 if (ret < 0)
1316 goto out;
1317 if (ret > 0) {
1318 if (path->slots[0] == 0)
1319 goto out;
1320 path->slots[0]--;
1321 }
1322 if (ret != 0)
1323 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1324
1325 if (key.type != key_type || key.objectid != dirid) {
1326 ret = 1;
1327 goto next;
1328 }
1329 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1330 struct btrfs_dir_log_item);
1331 found_end = btrfs_dir_log_end(path->nodes[0], item);
1332
1333 if (*start_ret >= key.offset && *start_ret <= found_end) {
1334 ret = 0;
1335 *start_ret = key.offset;
1336 *end_ret = found_end;
1337 goto out;
1338 }
1339 ret = 1;
1340next:
1341 /* check the next slot in the tree to see if it is a valid item */
1342 nritems = btrfs_header_nritems(path->nodes[0]);
1343 if (path->slots[0] >= nritems) {
1344 ret = btrfs_next_leaf(root, path);
1345 if (ret)
1346 goto out;
1347 } else {
1348 path->slots[0]++;
1349 }
1350
1351 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1352
1353 if (key.type != key_type || key.objectid != dirid) {
1354 ret = 1;
1355 goto out;
1356 }
1357 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1358 struct btrfs_dir_log_item);
1359 found_end = btrfs_dir_log_end(path->nodes[0], item);
1360 *start_ret = key.offset;
1361 *end_ret = found_end;
1362 ret = 0;
1363out:
1364 btrfs_release_path(root, path);
1365 return ret;
1366}
1367
1368/*
1369 * this looks for a given directory item in the log. If the directory
1370 * item is not in the log, the item is removed and the inode it points
1371 * to is unlinked
1372 */
1373static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1374 struct btrfs_root *root,
1375 struct btrfs_root *log,
1376 struct btrfs_path *path,
1377 struct btrfs_path *log_path,
1378 struct inode *dir,
1379 struct btrfs_key *dir_key)
1380{
1381 int ret;
1382 struct extent_buffer *eb;
1383 int slot;
1384 u32 item_size;
1385 struct btrfs_dir_item *di;
1386 struct btrfs_dir_item *log_di;
1387 int name_len;
1388 unsigned long ptr;
1389 unsigned long ptr_end;
1390 char *name;
1391 struct inode *inode;
1392 struct btrfs_key location;
1393
1394again:
1395 eb = path->nodes[0];
1396 slot = path->slots[0];
1397 item_size = btrfs_item_size_nr(eb, slot);
1398 ptr = btrfs_item_ptr_offset(eb, slot);
1399 ptr_end = ptr + item_size;
1400 while(ptr < ptr_end) {
1401 di = (struct btrfs_dir_item *)ptr;
1402 name_len = btrfs_dir_name_len(eb, di);
1403 name = kmalloc(name_len, GFP_NOFS);
1404 if (!name) {
1405 ret = -ENOMEM;
1406 goto out;
1407 }
1408 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1409 name_len);
1410 log_di = NULL;
1411 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1412 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1413 dir_key->objectid,
1414 name, name_len, 0);
1415 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1416 log_di = btrfs_lookup_dir_index_item(trans, log,
1417 log_path,
1418 dir_key->objectid,
1419 dir_key->offset,
1420 name, name_len, 0);
1421 }
1422 if (!log_di || IS_ERR(log_di)) {
1423 btrfs_dir_item_key_to_cpu(eb, di, &location);
1424 btrfs_release_path(root, path);
1425 btrfs_release_path(log, log_path);
1426 inode = read_one_inode(root, location.objectid);
1427 BUG_ON(!inode);
1428
1429 ret = link_to_fixup_dir(trans, root,
1430 path, location.objectid);
1431 BUG_ON(ret);
1432 btrfs_inc_nlink(inode);
1433 ret = btrfs_unlink_inode(trans, root, dir, inode,
1434 name, name_len);
1435 BUG_ON(ret);
1436 kfree(name);
1437 iput(inode);
1438
1439 /* there might still be more names under this key
1440 * check and repeat if required
1441 */
1442 ret = btrfs_search_slot(NULL, root, dir_key, path,
1443 0, 0);
1444 if (ret == 0)
1445 goto again;
1446 ret = 0;
1447 goto out;
1448 }
1449 btrfs_release_path(log, log_path);
1450 kfree(name);
1451
1452 ptr = (unsigned long)(di + 1);
1453 ptr += name_len;
1454 }
1455 ret = 0;
1456out:
1457 btrfs_release_path(root, path);
1458 btrfs_release_path(log, log_path);
1459 return ret;
1460}
1461
1462/*
1463 * deletion replay happens before we copy any new directory items
1464 * out of the log or out of backreferences from inodes. It
1465 * scans the log to find ranges of keys that log is authoritative for,
1466 * and then scans the directory to find items in those ranges that are
1467 * not present in the log.
1468 *
1469 * Anything we don't find in the log is unlinked and removed from the
1470 * directory.
1471 */
1472static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1473 struct btrfs_root *root,
1474 struct btrfs_root *log,
1475 struct btrfs_path *path,
1476 u64 dirid)
1477{
1478 u64 range_start;
1479 u64 range_end;
1480 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1481 int ret = 0;
1482 struct btrfs_key dir_key;
1483 struct btrfs_key found_key;
1484 struct btrfs_path *log_path;
1485 struct inode *dir;
1486
1487 dir_key.objectid = dirid;
1488 dir_key.type = BTRFS_DIR_ITEM_KEY;
1489 log_path = btrfs_alloc_path();
1490 if (!log_path)
1491 return -ENOMEM;
1492
1493 dir = read_one_inode(root, dirid);
1494 /* it isn't an error if the inode isn't there, that can happen
1495 * because we replay the deletes before we copy in the inode item
1496 * from the log
1497 */
1498 if (!dir) {
1499 btrfs_free_path(log_path);
1500 return 0;
1501 }
1502again:
1503 range_start = 0;
1504 range_end = 0;
1505 while(1) {
1506 ret = find_dir_range(log, path, dirid, key_type,
1507 &range_start, &range_end);
1508 if (ret != 0)
1509 break;
1510
1511 dir_key.offset = range_start;
1512 while(1) {
1513 int nritems;
1514 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1515 0, 0);
1516 if (ret < 0)
1517 goto out;
1518
1519 nritems = btrfs_header_nritems(path->nodes[0]);
1520 if (path->slots[0] >= nritems) {
1521 ret = btrfs_next_leaf(root, path);
1522 if (ret)
1523 break;
1524 }
1525 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1526 path->slots[0]);
1527 if (found_key.objectid != dirid ||
1528 found_key.type != dir_key.type)
1529 goto next_type;
1530
1531 if (found_key.offset > range_end)
1532 break;
1533
1534 ret = check_item_in_log(trans, root, log, path,
1535 log_path, dir, &found_key);
1536 BUG_ON(ret);
1537 if (found_key.offset == (u64)-1)
1538 break;
1539 dir_key.offset = found_key.offset + 1;
1540 }
1541 btrfs_release_path(root, path);
1542 if (range_end == (u64)-1)
1543 break;
1544 range_start = range_end + 1;
1545 }
1546
1547next_type:
1548 ret = 0;
1549 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1550 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1551 dir_key.type = BTRFS_DIR_INDEX_KEY;
1552 btrfs_release_path(root, path);
1553 goto again;
1554 }
1555out:
1556 btrfs_release_path(root, path);
1557 btrfs_free_path(log_path);
1558 iput(dir);
1559 return ret;
1560}
1561
1562/*
1563 * the process_func used to replay items from the log tree. This
1564 * gets called in two different stages. The first stage just looks
1565 * for inodes and makes sure they are all copied into the subvolume.
1566 *
1567 * The second stage copies all the other item types from the log into
1568 * the subvolume. The two stage approach is slower, but gets rid of
1569 * lots of complexity around inodes referencing other inodes that exist
1570 * only in the log (references come from either directory items or inode
1571 * back refs).
1572 */
1573static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1574 struct walk_control *wc, u64 gen)
1575{
1576 int nritems;
1577 struct btrfs_path *path;
1578 struct btrfs_root *root = wc->replay_dest;
1579 struct btrfs_key key;
1580 u32 item_size;
1581 int level;
1582 int i;
1583 int ret;
1584
1585 btrfs_read_buffer(eb, gen);
1586
1587 level = btrfs_header_level(eb);
1588
1589 if (level != 0)
1590 return 0;
1591
1592 path = btrfs_alloc_path();
1593 BUG_ON(!path);
1594
1595 nritems = btrfs_header_nritems(eb);
1596 for (i = 0; i < nritems; i++) {
1597 btrfs_item_key_to_cpu(eb, &key, i);
1598 item_size = btrfs_item_size_nr(eb, i);
1599
1600 /* inode keys are done during the first stage */
1601 if (key.type == BTRFS_INODE_ITEM_KEY &&
1602 wc->stage == LOG_WALK_REPLAY_INODES) {
1603 struct inode *inode;
1604 struct btrfs_inode_item *inode_item;
1605 u32 mode;
1606
1607 inode_item = btrfs_item_ptr(eb, i,
1608 struct btrfs_inode_item);
1609 mode = btrfs_inode_mode(eb, inode_item);
1610 if (S_ISDIR(mode)) {
1611 ret = replay_dir_deletes(wc->trans,
1612 root, log, path, key.objectid);
1613 BUG_ON(ret);
1614 }
1615 ret = overwrite_item(wc->trans, root, path,
1616 eb, i, &key);
1617 BUG_ON(ret);
1618
1619 /* for regular files, truncate away
1620 * extents past the new EOF
1621 */
1622 if (S_ISREG(mode)) {
1623 inode = read_one_inode(root,
1624 key.objectid);
1625 BUG_ON(!inode);
1626
1627 ret = btrfs_truncate_inode_items(wc->trans,
1628 root, inode, inode->i_size,
1629 BTRFS_EXTENT_DATA_KEY);
1630 BUG_ON(ret);
1631 iput(inode);
1632 }
1633 ret = link_to_fixup_dir(wc->trans, root,
1634 path, key.objectid);
1635 BUG_ON(ret);
1636 }
1637 if (wc->stage < LOG_WALK_REPLAY_ALL)
1638 continue;
1639
1640 /* these keys are simply copied */
1641 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1642 ret = overwrite_item(wc->trans, root, path,
1643 eb, i, &key);
1644 BUG_ON(ret);
1645 } else if (key.type == BTRFS_INODE_REF_KEY) {
1646 ret = add_inode_ref(wc->trans, root, log, path,
1647 eb, i, &key);
1648 BUG_ON(ret && ret != -ENOENT);
1649 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1650 ret = replay_one_extent(wc->trans, root, path,
1651 eb, i, &key);
1652 BUG_ON(ret);
1653 } else if (key.type == BTRFS_CSUM_ITEM_KEY) {
1654 ret = replay_one_csum(wc->trans, root, path,
1655 eb, i, &key);
1656 BUG_ON(ret);
1657 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1658 key.type == BTRFS_DIR_INDEX_KEY) {
1659 ret = replay_one_dir_item(wc->trans, root, path,
1660 eb, i, &key);
1661 BUG_ON(ret);
1662 }
1663 }
1664 btrfs_free_path(path);
1665 return 0;
1666}
1667
1668static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1669 struct btrfs_root *root,
1670 struct btrfs_path *path, int *level,
1671 struct walk_control *wc)
1672{
1673 u64 root_owner;
1674 u64 root_gen;
1675 u64 bytenr;
1676 u64 ptr_gen;
1677 struct extent_buffer *next;
1678 struct extent_buffer *cur;
1679 struct extent_buffer *parent;
1680 u32 blocksize;
1681 int ret = 0;
1682
1683 WARN_ON(*level < 0);
1684 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1685
1686 while(*level > 0) {
1687 WARN_ON(*level < 0);
1688 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1689 cur = path->nodes[*level];
1690
1691 if (btrfs_header_level(cur) != *level)
1692 WARN_ON(1);
1693
1694 if (path->slots[*level] >=
1695 btrfs_header_nritems(cur))
1696 break;
1697
1698 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1699 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1700 blocksize = btrfs_level_size(root, *level - 1);
1701
1702 parent = path->nodes[*level];
1703 root_owner = btrfs_header_owner(parent);
1704 root_gen = btrfs_header_generation(parent);
1705
1706 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1707
1708 wc->process_func(root, next, wc, ptr_gen);
1709
1710 if (*level == 1) {
1711 path->slots[*level]++;
1712 if (wc->free) {
1713 btrfs_read_buffer(next, ptr_gen);
1714
1715 btrfs_tree_lock(next);
1716 clean_tree_block(trans, root, next);
1717 btrfs_wait_tree_block_writeback(next);
1718 btrfs_tree_unlock(next);
1719
1720 ret = btrfs_drop_leaf_ref(trans, root, next);
1721 BUG_ON(ret);
1722
1723 WARN_ON(root_owner !=
1724 BTRFS_TREE_LOG_OBJECTID);
1725 ret = btrfs_free_extent(trans, root, bytenr,
1726 blocksize, root_owner,
1727 root_gen, 0, 0, 1);
1728 BUG_ON(ret);
1729 }
1730 free_extent_buffer(next);
1731 continue;
1732 }
1733 btrfs_read_buffer(next, ptr_gen);
1734
1735 WARN_ON(*level <= 0);
1736 if (path->nodes[*level-1])
1737 free_extent_buffer(path->nodes[*level-1]);
1738 path->nodes[*level-1] = next;
1739 *level = btrfs_header_level(next);
1740 path->slots[*level] = 0;
1741 cond_resched();
1742 }
1743 WARN_ON(*level < 0);
1744 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1745
1746 if (path->nodes[*level] == root->node) {
1747 parent = path->nodes[*level];
1748 } else {
1749 parent = path->nodes[*level + 1];
1750 }
1751 bytenr = path->nodes[*level]->start;
1752
1753 blocksize = btrfs_level_size(root, *level);
1754 root_owner = btrfs_header_owner(parent);
1755 root_gen = btrfs_header_generation(parent);
1756
1757 wc->process_func(root, path->nodes[*level], wc,
1758 btrfs_header_generation(path->nodes[*level]));
1759
1760 if (wc->free) {
1761 next = path->nodes[*level];
1762 btrfs_tree_lock(next);
1763 clean_tree_block(trans, root, next);
1764 btrfs_wait_tree_block_writeback(next);
1765 btrfs_tree_unlock(next);
1766
1767 if (*level == 0) {
1768 ret = btrfs_drop_leaf_ref(trans, root, next);
1769 BUG_ON(ret);
1770 }
1771 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1772 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
1773 root_owner, root_gen, 0, 0, 1);
1774 BUG_ON(ret);
1775 }
1776 free_extent_buffer(path->nodes[*level]);
1777 path->nodes[*level] = NULL;
1778 *level += 1;
1779
1780 cond_resched();
1781 return 0;
1782}
1783
1784static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 struct btrfs_path *path, int *level,
1787 struct walk_control *wc)
1788{
1789 u64 root_owner;
1790 u64 root_gen;
1791 int i;
1792 int slot;
1793 int ret;
1794
1795 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1796 slot = path->slots[i];
1797 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1798 struct extent_buffer *node;
1799 node = path->nodes[i];
1800 path->slots[i]++;
1801 *level = i;
1802 WARN_ON(*level == 0);
1803 return 0;
1804 } else {
1805 if (path->nodes[*level] == root->node) {
1806 root_owner = root->root_key.objectid;
1807 root_gen =
1808 btrfs_header_generation(path->nodes[*level]);
1809 } else {
1810 struct extent_buffer *node;
1811 node = path->nodes[*level + 1];
1812 root_owner = btrfs_header_owner(node);
1813 root_gen = btrfs_header_generation(node);
1814 }
1815 wc->process_func(root, path->nodes[*level], wc,
1816 btrfs_header_generation(path->nodes[*level]));
1817 if (wc->free) {
1818 struct extent_buffer *next;
1819
1820 next = path->nodes[*level];
1821
1822 btrfs_tree_lock(next);
1823 clean_tree_block(trans, root, next);
1824 btrfs_wait_tree_block_writeback(next);
1825 btrfs_tree_unlock(next);
1826
1827 if (*level == 0) {
1828 ret = btrfs_drop_leaf_ref(trans, root,
1829 next);
1830 BUG_ON(ret);
1831 }
1832
1833 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1834 ret = btrfs_free_extent(trans, root,
1835 path->nodes[*level]->start,
1836 path->nodes[*level]->len,
1837 root_owner, root_gen, 0, 0, 1);
1838 BUG_ON(ret);
1839 }
1840 free_extent_buffer(path->nodes[*level]);
1841 path->nodes[*level] = NULL;
1842 *level = i + 1;
1843 }
1844 }
1845 return 1;
1846}
1847
1848/*
1849 * drop the reference count on the tree rooted at 'snap'. This traverses
1850 * the tree freeing any blocks that have a ref count of zero after being
1851 * decremented.
1852 */
1853static int walk_log_tree(struct btrfs_trans_handle *trans,
1854 struct btrfs_root *log, struct walk_control *wc)
1855{
1856 int ret = 0;
1857 int wret;
1858 int level;
1859 struct btrfs_path *path;
1860 int i;
1861 int orig_level;
1862
1863 path = btrfs_alloc_path();
1864 BUG_ON(!path);
1865
1866 level = btrfs_header_level(log->node);
1867 orig_level = level;
1868 path->nodes[level] = log->node;
1869 extent_buffer_get(log->node);
1870 path->slots[level] = 0;
1871
1872 while(1) {
1873 wret = walk_down_log_tree(trans, log, path, &level, wc);
1874 if (wret > 0)
1875 break;
1876 if (wret < 0)
1877 ret = wret;
1878
1879 wret = walk_up_log_tree(trans, log, path, &level, wc);
1880 if (wret > 0)
1881 break;
1882 if (wret < 0)
1883 ret = wret;
1884 }
1885
1886 /* was the root node processed? if not, catch it here */
1887 if (path->nodes[orig_level]) {
1888 wc->process_func(log, path->nodes[orig_level], wc,
1889 btrfs_header_generation(path->nodes[orig_level]));
1890 if (wc->free) {
1891 struct extent_buffer *next;
1892
1893 next = path->nodes[orig_level];
1894
1895 btrfs_tree_lock(next);
1896 clean_tree_block(trans, log, next);
1897 btrfs_wait_tree_block_writeback(next);
1898 btrfs_tree_unlock(next);
1899
1900 if (orig_level == 0) {
1901 ret = btrfs_drop_leaf_ref(trans, log,
1902 next);
1903 BUG_ON(ret);
1904 }
1905 WARN_ON(log->root_key.objectid !=
1906 BTRFS_TREE_LOG_OBJECTID);
1907 ret = btrfs_free_extent(trans, log,
1908 next->start, next->len,
1909 log->root_key.objectid,
1910 btrfs_header_generation(next),
1911 0, 0, 1);
1912 BUG_ON(ret);
1913 }
1914 }
1915
1916 for (i = 0; i <= orig_level; i++) {
1917 if (path->nodes[i]) {
1918 free_extent_buffer(path->nodes[i]);
1919 path->nodes[i] = NULL;
1920 }
1921 }
1922 btrfs_free_path(path);
1923 if (wc->free)
1924 free_extent_buffer(log->node);
1925 return ret;
1926}
1927
1928int wait_log_commit(struct btrfs_root *log)
1929{
1930 DEFINE_WAIT(wait);
1931 u64 transid = log->fs_info->tree_log_transid;
1932
1933 do {
1934 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1935 TASK_UNINTERRUPTIBLE);
1936 mutex_unlock(&log->fs_info->tree_log_mutex);
1937 if (atomic_read(&log->fs_info->tree_log_commit))
1938 schedule();
1939 finish_wait(&log->fs_info->tree_log_wait, &wait);
1940 mutex_lock(&log->fs_info->tree_log_mutex);
1941 } while(transid == log->fs_info->tree_log_transid &&
1942 atomic_read(&log->fs_info->tree_log_commit));
1943 return 0;
1944}
1945
1946/*
1947 * btrfs_sync_log does sends a given tree log down to the disk and
1948 * updates the super blocks to record it. When this call is done,
1949 * you know that any inodes previously logged are safely on disk
1950 */
1951int btrfs_sync_log(struct btrfs_trans_handle *trans,
1952 struct btrfs_root *root)
1953{
1954 int ret;
1955 unsigned long batch;
1956 struct btrfs_root *log = root->log_root;
1957 struct walk_control wc = {
1958 .write = 1,
1959 .process_func = process_one_buffer
1960 };
1961
1962 mutex_lock(&log->fs_info->tree_log_mutex);
1963 if (atomic_read(&log->fs_info->tree_log_commit)) {
1964 wait_log_commit(log);
1965 goto out;
1966 }
1967 atomic_set(&log->fs_info->tree_log_commit, 1);
1968
1969 while(1) {
1970 mutex_unlock(&log->fs_info->tree_log_mutex);
1971 schedule_timeout_uninterruptible(1);
1972 mutex_lock(&log->fs_info->tree_log_mutex);
1973 batch = log->fs_info->tree_log_batch;
1974
1975 while(atomic_read(&log->fs_info->tree_log_writers)) {
1976 DEFINE_WAIT(wait);
1977 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1978 TASK_UNINTERRUPTIBLE);
1979 batch = log->fs_info->tree_log_batch;
1980 mutex_unlock(&log->fs_info->tree_log_mutex);
1981 if (atomic_read(&log->fs_info->tree_log_writers))
1982 schedule();
1983 mutex_lock(&log->fs_info->tree_log_mutex);
1984 finish_wait(&log->fs_info->tree_log_wait, &wait);
1985 }
1986 if (batch == log->fs_info->tree_log_batch)
1987 break;
1988 }
1989 ret = walk_log_tree(trans, log, &wc);
1990 BUG_ON(ret);
1991
1992 ret = walk_log_tree(trans, log->fs_info->log_root_tree, &wc);
1993 BUG_ON(ret);
1994
1995 wc.wait = 1;
1996
1997 ret = walk_log_tree(trans, log, &wc);
1998 BUG_ON(ret);
1999
2000 ret = walk_log_tree(trans, log->fs_info->log_root_tree, &wc);
2001 BUG_ON(ret);
2002
2003 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2004 log->fs_info->log_root_tree->node->start);
2005 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2006 btrfs_header_level(log->fs_info->log_root_tree->node));
2007
2008 write_ctree_super(trans, log->fs_info->tree_root);
2009 log->fs_info->tree_log_transid++;
2010 log->fs_info->tree_log_batch = 0;
2011 atomic_set(&log->fs_info->tree_log_commit, 0);
2012 smp_mb();
2013 if (waitqueue_active(&log->fs_info->tree_log_wait))
2014 wake_up(&log->fs_info->tree_log_wait);
2015out:
2016 mutex_unlock(&log->fs_info->tree_log_mutex);
2017 return 0;
2018
2019}
2020
2021/*
2022 * free all the extents used by the tree log. This should be called
2023 * at commit time of the full transaction
2024 */
2025int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2026{
2027 int ret;
2028 struct btrfs_root *log;
2029 struct key;
2030 struct walk_control wc = {
2031 .free = 1,
2032 .process_func = process_one_buffer
2033 };
2034
2035 if (!root->log_root)
2036 return 0;
2037
2038 log = root->log_root;
2039 ret = walk_log_tree(trans, log, &wc);
2040 BUG_ON(ret);
2041
2042 log = root->log_root;
2043 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2044 &log->root_key);
2045 BUG_ON(ret);
2046 root->log_root = NULL;
2047 kfree(root->log_root);
2048 return 0;
2049}
2050
2051/*
2052 * helper function to update the item for a given subvolumes log root
2053 * in the tree of log roots
2054 */
2055static int update_log_root(struct btrfs_trans_handle *trans,
2056 struct btrfs_root *log)
2057{
2058 u64 bytenr = btrfs_root_bytenr(&log->root_item);
2059 int ret;
2060
2061 if (log->node->start == bytenr)
2062 return 0;
2063
2064 btrfs_set_root_bytenr(&log->root_item, log->node->start);
2065 btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2066 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2067 &log->root_key, &log->root_item);
2068 BUG_ON(ret);
2069 return ret;
2070}
2071
2072/*
2073 * If both a file and directory are logged, and unlinks or renames are
2074 * mixed in, we have a few interesting corners:
2075 *
2076 * create file X in dir Y
2077 * link file X to X.link in dir Y
2078 * fsync file X
2079 * unlink file X but leave X.link
2080 * fsync dir Y
2081 *
2082 * After a crash we would expect only X.link to exist. But file X
2083 * didn't get fsync'd again so the log has back refs for X and X.link.
2084 *
2085 * We solve this by removing directory entries and inode backrefs from the
2086 * log when a file that was logged in the current transaction is
2087 * unlinked. Any later fsync will include the updated log entries, and
2088 * we'll be able to reconstruct the proper directory items from backrefs.
2089 *
2090 * This optimizations allows us to avoid relogging the entire inode
2091 * or the entire directory.
2092 */
2093int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2094 struct btrfs_root *root,
2095 const char *name, int name_len,
2096 struct inode *dir, u64 index)
2097{
2098 struct btrfs_root *log;
2099 struct btrfs_dir_item *di;
2100 struct btrfs_path *path;
2101 int ret;
2102 int bytes_del = 0;
2103
2104 ret = join_running_log_trans(root);
2105 if (ret)
2106 return 0;
2107
2108 mutex_lock(&BTRFS_I(dir)->log_mutex);
2109
2110 log = root->log_root;
2111 path = btrfs_alloc_path();
2112 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2113 name, name_len, -1);
2114 if (di && !IS_ERR(di)) {
2115 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2116 bytes_del += name_len;
2117 BUG_ON(ret);
2118 }
2119 btrfs_release_path(log, path);
2120 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2121 index, name, name_len, -1);
2122 if (di && !IS_ERR(di)) {
2123 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2124 bytes_del += name_len;
2125 BUG_ON(ret);
2126 }
2127
2128 /* update the directory size in the log to reflect the names
2129 * we have removed
2130 */
2131 if (bytes_del) {
2132 struct btrfs_key key;
2133
2134 key.objectid = dir->i_ino;
2135 key.offset = 0;
2136 key.type = BTRFS_INODE_ITEM_KEY;
2137 btrfs_release_path(log, path);
2138
2139 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2140 if (ret == 0) {
2141 struct btrfs_inode_item *item;
2142 u64 i_size;
2143
2144 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2145 struct btrfs_inode_item);
2146 i_size = btrfs_inode_size(path->nodes[0], item);
2147 if (i_size > bytes_del)
2148 i_size -= bytes_del;
2149 else
2150 i_size = 0;
2151 btrfs_set_inode_size(path->nodes[0], item, i_size);
2152 btrfs_mark_buffer_dirty(path->nodes[0]);
2153 } else
2154 ret = 0;
2155 btrfs_release_path(log, path);
2156 }
2157
2158 btrfs_free_path(path);
2159 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2160 end_log_trans(root);
2161
2162 return 0;
2163}
2164
2165/* see comments for btrfs_del_dir_entries_in_log */
2166int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2167 struct btrfs_root *root,
2168 const char *name, int name_len,
2169 struct inode *inode, u64 dirid)
2170{
2171 struct btrfs_root *log;
2172 u64 index;
2173 int ret;
2174
2175 ret = join_running_log_trans(root);
2176 if (ret)
2177 return 0;
2178 log = root->log_root;
2179 mutex_lock(&BTRFS_I(inode)->log_mutex);
2180
2181 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2182 dirid, &index);
2183 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2184 end_log_trans(root);
2185
2186 if (ret == 0 || ret == -ENOENT)
2187 return 0;
2188 return ret;
2189}
2190
2191/*
2192 * creates a range item in the log for 'dirid'. first_offset and
2193 * last_offset tell us which parts of the key space the log should
2194 * be considered authoritative for.
2195 */
2196static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *log,
2198 struct btrfs_path *path,
2199 int key_type, u64 dirid,
2200 u64 first_offset, u64 last_offset)
2201{
2202 int ret;
2203 struct btrfs_key key;
2204 struct btrfs_dir_log_item *item;
2205
2206 key.objectid = dirid;
2207 key.offset = first_offset;
2208 if (key_type == BTRFS_DIR_ITEM_KEY)
2209 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2210 else
2211 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2212 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2213 BUG_ON(ret);
2214
2215 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2216 struct btrfs_dir_log_item);
2217 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2218 btrfs_mark_buffer_dirty(path->nodes[0]);
2219 btrfs_release_path(log, path);
2220 return 0;
2221}
2222
2223/*
2224 * log all the items included in the current transaction for a given
2225 * directory. This also creates the range items in the log tree required
2226 * to replay anything deleted before the fsync
2227 */
2228static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2229 struct btrfs_root *root, struct inode *inode,
2230 struct btrfs_path *path,
2231 struct btrfs_path *dst_path, int key_type,
2232 u64 min_offset, u64 *last_offset_ret)
2233{
2234 struct btrfs_key min_key;
2235 struct btrfs_key max_key;
2236 struct btrfs_root *log = root->log_root;
2237 struct extent_buffer *src;
2238 int ret;
2239 int i;
2240 int nritems;
2241 u64 first_offset = min_offset;
2242 u64 last_offset = (u64)-1;
2243
2244 log = root->log_root;
2245 max_key.objectid = inode->i_ino;
2246 max_key.offset = (u64)-1;
2247 max_key.type = key_type;
2248
2249 min_key.objectid = inode->i_ino;
2250 min_key.type = key_type;
2251 min_key.offset = min_offset;
2252
2253 path->keep_locks = 1;
2254
2255 ret = btrfs_search_forward(root, &min_key, &max_key,
2256 path, 0, trans->transid);
2257
2258 /*
2259 * we didn't find anything from this transaction, see if there
2260 * is anything at all
2261 */
2262 if (ret != 0 || min_key.objectid != inode->i_ino ||
2263 min_key.type != key_type) {
2264 min_key.objectid = inode->i_ino;
2265 min_key.type = key_type;
2266 min_key.offset = (u64)-1;
2267 btrfs_release_path(root, path);
2268 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2269 if (ret < 0) {
2270 btrfs_release_path(root, path);
2271 return ret;
2272 }
2273 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2274
2275 /* if ret == 0 there are items for this type,
2276 * create a range to tell us the last key of this type.
2277 * otherwise, there are no items in this directory after
2278 * *min_offset, and we create a range to indicate that.
2279 */
2280 if (ret == 0) {
2281 struct btrfs_key tmp;
2282 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2283 path->slots[0]);
2284 if (key_type == tmp.type) {
2285 first_offset = max(min_offset, tmp.offset) + 1;
2286 }
2287 }
2288 goto done;
2289 }
2290
2291 /* go backward to find any previous key */
2292 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2293 if (ret == 0) {
2294 struct btrfs_key tmp;
2295 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2296 if (key_type == tmp.type) {
2297 first_offset = tmp.offset;
2298 ret = overwrite_item(trans, log, dst_path,
2299 path->nodes[0], path->slots[0],
2300 &tmp);
2301 }
2302 }
2303 btrfs_release_path(root, path);
2304
2305 /* find the first key from this transaction again */
2306 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2307 if (ret != 0) {
2308 WARN_ON(1);
2309 goto done;
2310 }
2311
2312 /*
2313 * we have a block from this transaction, log every item in it
2314 * from our directory
2315 */
2316 while(1) {
2317 struct btrfs_key tmp;
2318 src = path->nodes[0];
2319 nritems = btrfs_header_nritems(src);
2320 for (i = path->slots[0]; i < nritems; i++) {
2321 btrfs_item_key_to_cpu(src, &min_key, i);
2322
2323 if (min_key.objectid != inode->i_ino ||
2324 min_key.type != key_type)
2325 goto done;
2326 ret = overwrite_item(trans, log, dst_path, src, i,
2327 &min_key);
2328 BUG_ON(ret);
2329 }
2330 path->slots[0] = nritems;
2331
2332 /*
2333 * look ahead to the next item and see if it is also
2334 * from this directory and from this transaction
2335 */
2336 ret = btrfs_next_leaf(root, path);
2337 if (ret == 1) {
2338 last_offset = (u64)-1;
2339 goto done;
2340 }
2341 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2342 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2343 last_offset = (u64)-1;
2344 goto done;
2345 }
2346 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2347 ret = overwrite_item(trans, log, dst_path,
2348 path->nodes[0], path->slots[0],
2349 &tmp);
2350
2351 BUG_ON(ret);
2352 last_offset = tmp.offset;
2353 goto done;
2354 }
2355 }
2356done:
2357 *last_offset_ret = last_offset;
2358 btrfs_release_path(root, path);
2359 btrfs_release_path(log, dst_path);
2360
2361 /* insert the log range keys to indicate where the log is valid */
2362 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2363 first_offset, last_offset);
2364 BUG_ON(ret);
2365 return 0;
2366}
2367
2368/*
2369 * logging directories is very similar to logging inodes, We find all the items
2370 * from the current transaction and write them to the log.
2371 *
2372 * The recovery code scans the directory in the subvolume, and if it finds a
2373 * key in the range logged that is not present in the log tree, then it means
2374 * that dir entry was unlinked during the transaction.
2375 *
2376 * In order for that scan to work, we must include one key smaller than
2377 * the smallest logged by this transaction and one key larger than the largest
2378 * key logged by this transaction.
2379 */
2380static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2381 struct btrfs_root *root, struct inode *inode,
2382 struct btrfs_path *path,
2383 struct btrfs_path *dst_path)
2384{
2385 u64 min_key;
2386 u64 max_key;
2387 int ret;
2388 int key_type = BTRFS_DIR_ITEM_KEY;
2389
2390again:
2391 min_key = 0;
2392 max_key = 0;
2393 while(1) {
2394 ret = log_dir_items(trans, root, inode, path,
2395 dst_path, key_type, min_key,
2396 &max_key);
2397 BUG_ON(ret);
2398 if (max_key == (u64)-1)
2399 break;
2400 min_key = max_key + 1;
2401 }
2402
2403 if (key_type == BTRFS_DIR_ITEM_KEY) {
2404 key_type = BTRFS_DIR_INDEX_KEY;
2405 goto again;
2406 }
2407 return 0;
2408}
2409
2410/*
2411 * a helper function to drop items from the log before we relog an
2412 * inode. max_key_type indicates the highest item type to remove.
2413 * This cannot be run for file data extents because it does not
2414 * free the extents they point to.
2415 */
2416static int drop_objectid_items(struct btrfs_trans_handle *trans,
2417 struct btrfs_root *log,
2418 struct btrfs_path *path,
2419 u64 objectid, int max_key_type)
2420{
2421 int ret;
2422 struct btrfs_key key;
2423 struct btrfs_key found_key;
2424
2425 key.objectid = objectid;
2426 key.type = max_key_type;
2427 key.offset = (u64)-1;
2428
2429 while(1) {
2430 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2431
2432 if (ret != 1)
2433 break;
2434
2435 if (path->slots[0] == 0)
2436 break;
2437
2438 path->slots[0]--;
2439 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2440 path->slots[0]);
2441
2442 if (found_key.objectid != objectid)
2443 break;
2444
2445 ret = btrfs_del_item(trans, log, path);
2446 BUG_ON(ret);
2447 btrfs_release_path(log, path);
2448 }
2449 btrfs_release_path(log, path);
2450 return 0;
2451}
2452
2453/* log a single inode in the tree log.
2454 * At least one parent directory for this inode must exist in the tree
2455 * or be logged already.
2456 *
2457 * Any items from this inode changed by the current transaction are copied
2458 * to the log tree. An extra reference is taken on any extents in this
2459 * file, allowing us to avoid a whole pile of corner cases around logging
2460 * blocks that have been removed from the tree.
2461 *
2462 * See LOG_INODE_ALL and related defines for a description of what inode_only
2463 * does.
2464 *
2465 * This handles both files and directories.
2466 */
2467static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *root, struct inode *inode,
2469 int inode_only)
2470{
2471 struct btrfs_path *path;
2472 struct btrfs_path *dst_path;
2473 struct btrfs_key min_key;
2474 struct btrfs_key max_key;
2475 struct btrfs_root *log = root->log_root;
2476 unsigned long src_offset;
2477 unsigned long dst_offset;
2478 struct extent_buffer *src;
2479 struct btrfs_file_extent_item *extent;
2480 struct btrfs_inode_item *inode_item;
2481 u32 size;
2482 int ret;
2483
2484 log = root->log_root;
2485
2486 path = btrfs_alloc_path();
2487 dst_path = btrfs_alloc_path();
2488
2489 min_key.objectid = inode->i_ino;
2490 min_key.type = BTRFS_INODE_ITEM_KEY;
2491 min_key.offset = 0;
2492
2493 max_key.objectid = inode->i_ino;
2494 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2495 max_key.type = BTRFS_XATTR_ITEM_KEY;
2496 else
2497 max_key.type = (u8)-1;
2498 max_key.offset = (u64)-1;
2499
2500 /*
2501 * if this inode has already been logged and we're in inode_only
2502 * mode, we don't want to delete the things that have already
2503 * been written to the log.
2504 *
2505 * But, if the inode has been through an inode_only log,
2506 * the logged_trans field is not set. This allows us to catch
2507 * any new names for this inode in the backrefs by logging it
2508 * again
2509 */
2510 if (inode_only == LOG_INODE_EXISTS &&
2511 BTRFS_I(inode)->logged_trans == trans->transid) {
2512 btrfs_free_path(path);
2513 btrfs_free_path(dst_path);
2514 goto out;
2515 }
2516 mutex_lock(&BTRFS_I(inode)->log_mutex);
2517
2518 /*
2519 * a brute force approach to making sure we get the most uptodate
2520 * copies of everything.
2521 */
2522 if (S_ISDIR(inode->i_mode)) {
2523 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2524
2525 if (inode_only == LOG_INODE_EXISTS)
2526 max_key_type = BTRFS_XATTR_ITEM_KEY;
2527 ret = drop_objectid_items(trans, log, path,
2528 inode->i_ino, max_key_type);
2529 } else {
2530 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2531 }
2532 BUG_ON(ret);
2533 path->keep_locks = 1;
2534
2535 while(1) {
2536 ret = btrfs_search_forward(root, &min_key, &max_key,
2537 path, 0, trans->transid);
2538 if (ret != 0)
2539 break;
2540
2541 if (min_key.objectid != inode->i_ino)
2542 break;
2543 if (min_key.type > max_key.type)
2544 break;
2545
2546 src = path->nodes[0];
2547 size = btrfs_item_size_nr(src, path->slots[0]);
2548 ret = btrfs_insert_empty_item(trans, log, dst_path, &min_key,
2549 size);
2550 if (ret)
2551 BUG();
2552
2553 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2554 dst_path->slots[0]);
2555
2556 src_offset = btrfs_item_ptr_offset(src, path->slots[0]);
2557
2558 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2559 src_offset, size);
2560
2561 if (inode_only == LOG_INODE_EXISTS &&
2562 min_key.type == BTRFS_INODE_ITEM_KEY) {
2563 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2564 dst_path->slots[0],
2565 struct btrfs_inode_item);
2566 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2567
2568 /* set the generation to zero so the recover code
2569 * can tell the difference between an logging
2570 * just to say 'this inode exists' and a logging
2571 * to say 'update this inode with these values'
2572 */
2573 btrfs_set_inode_generation(dst_path->nodes[0],
2574 inode_item, 0);
2575 }
2576 /* take a reference on file data extents so that truncates
2577 * or deletes of this inode don't have to relog the inode
2578 * again
2579 */
2580 if (btrfs_key_type(&min_key) == BTRFS_EXTENT_DATA_KEY) {
2581 int found_type;
2582 extent = btrfs_item_ptr(src, path->slots[0],
2583 struct btrfs_file_extent_item);
2584
2585 found_type = btrfs_file_extent_type(src, extent);
2586 if (found_type == BTRFS_FILE_EXTENT_REG) {
2587 u64 ds = btrfs_file_extent_disk_bytenr(src,
2588 extent);
2589 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2590 extent);
2591 /* ds == 0 is a hole */
2592 if (ds != 0) {
2593 ret = btrfs_inc_extent_ref(trans, log,
2594 ds, dl,
2595 log->root_key.objectid,
2596 0,
2597 inode->i_ino,
2598 min_key.offset);
2599 BUG_ON(ret);
2600 }
2601 }
2602 }
2603
2604 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2605 btrfs_release_path(root, path);
2606 btrfs_release_path(log, dst_path);
2607
2608 if (min_key.offset < (u64)-1)
2609 min_key.offset++;
2610 else if (min_key.type < (u8)-1)
2611 min_key.type++;
2612 else if (min_key.objectid < (u64)-1)
2613 min_key.objectid++;
2614 else
2615 break;
2616 }
2617 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2618 btrfs_release_path(root, path);
2619 btrfs_release_path(log, dst_path);
2620 ret = log_directory_changes(trans, root, inode, path, dst_path);
2621 BUG_ON(ret);
2622 }
2623 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2624
2625 btrfs_free_path(path);
2626 btrfs_free_path(dst_path);
2627
2628 mutex_lock(&root->fs_info->tree_log_mutex);
2629 ret = update_log_root(trans, log);
2630 BUG_ON(ret);
2631 mutex_unlock(&root->fs_info->tree_log_mutex);
2632out:
2633 return 0;
2634}
2635
2636int btrfs_log_inode(struct btrfs_trans_handle *trans,
2637 struct btrfs_root *root, struct inode *inode,
2638 int inode_only)
2639{
2640 int ret;
2641
2642 start_log_trans(trans, root);
2643 ret = __btrfs_log_inode(trans, root, inode, inode_only);
2644 end_log_trans(root);
2645 return ret;
2646}
2647
2648/*
2649 * helper function around btrfs_log_inode to make sure newly created
2650 * parent directories also end up in the log. A minimal inode and backref
2651 * only logging is done of any parent directories that are older than
2652 * the last committed transaction
2653 */
2654int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2655 struct btrfs_root *root, struct dentry *dentry)
2656{
2657 int inode_only = LOG_INODE_ALL;
2658 struct super_block *sb;
2659 int ret;
2660
2661 start_log_trans(trans, root);
2662 sb = dentry->d_inode->i_sb;
2663 while(1) {
2664 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2665 inode_only);
2666 BUG_ON(ret);
2667 inode_only = LOG_INODE_EXISTS;
2668
2669 dentry = dentry->d_parent;
2670 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
2671 break;
2672
2673 if (BTRFS_I(dentry->d_inode)->generation <=
2674 root->fs_info->last_trans_committed)
2675 break;
2676 }
2677 end_log_trans(root);
2678 return 0;
2679}
2680
2681/*
2682 * it is not safe to log dentry if the chunk root has added new
2683 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2684 * If this returns 1, you must commit the transaction to safely get your
2685 * data on disk.
2686 */
2687int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2688 struct btrfs_root *root, struct dentry *dentry)
2689{
2690 u64 gen;
2691 gen = root->fs_info->last_trans_new_blockgroup;
2692 if (gen > root->fs_info->last_trans_committed)
2693 return 1;
2694 else
2695 return btrfs_log_dentry(trans, root, dentry);
2696}
2697
2698/*
2699 * should be called during mount to recover any replay any log trees
2700 * from the FS
2701 */
2702int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2703{
2704 int ret;
2705 struct btrfs_path *path;
2706 struct btrfs_trans_handle *trans;
2707 struct btrfs_key key;
2708 struct btrfs_key found_key;
2709 struct btrfs_key tmp_key;
2710 struct btrfs_root *log;
2711 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2712 struct walk_control wc = {
2713 .process_func = process_one_buffer,
2714 .stage = 0,
2715 };
2716
2717 fs_info->log_root_recovering = 1;
2718 path = btrfs_alloc_path();
2719 BUG_ON(!path);
2720
2721 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2722
2723 wc.trans = trans;
2724 wc.pin = 1;
2725
2726 walk_log_tree(trans, log_root_tree, &wc);
2727
2728again:
2729 key.objectid = BTRFS_TREE_LOG_OBJECTID;
2730 key.offset = (u64)-1;
2731 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2732
2733 while(1) {
2734 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2735 if (ret < 0)
2736 break;
2737 if (ret > 0) {
2738 if (path->slots[0] == 0)
2739 break;
2740 path->slots[0]--;
2741 }
2742 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2743 path->slots[0]);
2744 btrfs_release_path(log_root_tree, path);
2745 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2746 break;
2747
2748 log = btrfs_read_fs_root_no_radix(log_root_tree,
2749 &found_key);
2750 BUG_ON(!log);
2751
2752
2753 tmp_key.objectid = found_key.offset;
2754 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2755 tmp_key.offset = (u64)-1;
2756
2757 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
2758
2759 BUG_ON(!wc.replay_dest);
2760
2761 btrfs_record_root_in_trans(wc.replay_dest);
2762 ret = walk_log_tree(trans, log, &wc);
2763 BUG_ON(ret);
2764
2765 if (wc.stage == LOG_WALK_REPLAY_ALL) {
2766 ret = fixup_inode_link_counts(trans, wc.replay_dest,
2767 path);
2768 BUG_ON(ret);
2769 }
2770
2771 key.offset = found_key.offset - 1;
2772 free_extent_buffer(log->node);
2773 kfree(log);
2774
2775 if (found_key.offset == 0)
2776 break;
2777 }
2778 btrfs_release_path(log_root_tree, path);
2779
2780 /* step one is to pin it all, step two is to replay just inodes */
2781 if (wc.pin) {
2782 wc.pin = 0;
2783 wc.process_func = replay_one_buffer;
2784 wc.stage = LOG_WALK_REPLAY_INODES;
2785 goto again;
2786 }
2787 /* step three is to replay everything */
2788 if (wc.stage < LOG_WALK_REPLAY_ALL) {
2789 wc.stage++;
2790 goto again;
2791 }
2792
2793 btrfs_free_path(path);
2794
2795 free_extent_buffer(log_root_tree->node);
2796 log_root_tree->log_root = NULL;
2797 fs_info->log_root_recovering = 0;
2798
2799 /* step 4: commit the transaction, which also unpins the blocks */
2800 btrfs_commit_transaction(trans, fs_info->tree_root);
2801
2802 kfree(log_root_tree);
2803 return 0;
2804}
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
new file mode 100644
index 000000000000..b9409b32ed02
--- /dev/null
+++ b/fs/btrfs/tree-log.h
@@ -0,0 +1,41 @@
1/*
2 * Copyright (C) 2008 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#ifndef __TREE_LOG_
20#define __TREE_LOG_
21
22int btrfs_sync_log(struct btrfs_trans_handle *trans,
23 struct btrfs_root *root);
24int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root);
25int btrfs_log_dentry(struct btrfs_trans_handle *trans,
26 struct btrfs_root *root, struct dentry *dentry);
27int btrfs_recover_log_trees(struct btrfs_root *tree_root);
28int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
29 struct btrfs_root *root, struct dentry *dentry);
30int btrfs_log_inode(struct btrfs_trans_handle *trans,
31 struct btrfs_root *root, struct inode *inode,
32 int inode_only);
33int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root,
35 const char *name, int name_len,
36 struct inode *dir, u64 index);
37int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root,
39 const char *name, int name_len,
40 struct inode *inode, u64 dirid);
41#endif
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-21 09:16:25 -0500