aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-04-01 13:20:44 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-04-01 13:20:44 -0400
commitc226fd659fa7b6a7b038df5ae6856a68514bacde (patch)
treea695689d4d8cdd927400747b94250521eb863686
parentc09bca786ff941ed17c5f381c4eca5b106808c51 (diff)
parentd57e62b89796f751c9422801cbcd407a9f8dcdc4 (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable: Btrfs: try to free metadata pages when we free btree blocks Btrfs: add extra flushing for renames and truncates Btrfs: make sure btrfs_update_delayed_ref doesn't increase ref_mod Btrfs: optimize fsyncs on old files Btrfs: tree logging unlink/rename fixes Btrfs: Make sure i_nlink doesn't hit zero too soon during log replay Btrfs: limit balancing work while flushing delayed refs Btrfs: readahead checksums during btrfs_finish_ordered_io Btrfs: leave btree locks spinning more often Btrfs: Only let very young transactions grow during commit Btrfs: Check for a blocking lock before taking the spin Btrfs: reduce stack in cow_file_range Btrfs: reduce stalls during transaction commit Btrfs: process the delayed reference queue in clusters Btrfs: try to cleanup delayed refs while freeing extents Btrfs: reduce stack usage in some crucial tree balancing functions Btrfs: do extent allocation and reference count updates in the background Btrfs: don't preallocate metadata blocks during btrfs_search_slot
Diffstat
-rw-r--r--fs/btrfs/Makefile2
-rw-r--r--fs/btrfs/btrfs_inode.h31
-rw-r--r--fs/btrfs/ctree.c588
-rw-r--r--fs/btrfs/ctree.h69
-rw-r--r--fs/btrfs/delayed-ref.c669
-rw-r--r--fs/btrfs/delayed-ref.h193
-rw-r--r--fs/btrfs/dir-item.c3
-rw-r--r--fs/btrfs/disk-io.c81
-rw-r--r--fs/btrfs/disk-io.h1
-rw-r--r--fs/btrfs/extent-tree.c1674
-rw-r--r--fs/btrfs/extent_io.c51
-rw-r--r--fs/btrfs/extent_io.h3
-rw-r--r--fs/btrfs/file-item.c7
-rw-r--r--fs/btrfs/file.c50
-rw-r--r--fs/btrfs/inode-item.c3
-rw-r--r--fs/btrfs/inode.c194
-rw-r--r--fs/btrfs/locking.c21
-rw-r--r--fs/btrfs/ordered-data.c118
-rw-r--r--fs/btrfs/ordered-data.h4
-rw-r--r--fs/btrfs/transaction.c151
-rw-r--r--fs/btrfs/transaction.h8
-rw-r--r--fs/btrfs/tree-defrag.c2
-rw-r--r--fs/btrfs/tree-log.c444
-rw-r--r--fs/btrfs/tree-log.h17
24 files changed, 2762 insertions, 1622 deletions
diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index d2cf5a54a4b8..9adf5e4f7e96 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -8,7 +8,7 @@ btrfs-y := super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.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 export.o tree-log.o acl.o free-space-cache.o zlib.o \ 10 ref-cache.o export.o tree-log.o acl.o free-space-cache.o zlib.o \
11 compression.o 11 compression.o delayed-ref.o
12else 12else
13 13
14# Normal Makefile 14# Normal Makefile
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index 72677ce2b74f..b30986f00b9d 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -66,6 +66,12 @@ struct btrfs_inode {
66 */ 66 */
67 struct list_head delalloc_inodes; 67 struct list_head delalloc_inodes;
68 68
69 /*
70 * list for tracking inodes that must be sent to disk before a
71 * rename or truncate commit
72 */
73 struct list_head ordered_operations;
74
69 /* the space_info for where this inode's data allocations are done */ 75 /* the space_info for where this inode's data allocations are done */
70 struct btrfs_space_info *space_info; 76 struct btrfs_space_info *space_info;
71 77
@@ -86,12 +92,6 @@ struct btrfs_inode {
86 */ 92 */
87 u64 logged_trans; 93 u64 logged_trans;
88 94
89 /*
90 * trans that last made a change that should be fully fsync'd. This
91 * gets reset to zero each time the inode is logged
92 */
93 u64 log_dirty_trans;
94
95 /* total number of bytes pending delalloc, used by stat to calc the 95 /* total number of bytes pending delalloc, used by stat to calc the
96 * real block usage of the file 96 * real block usage of the file
97 */ 97 */
@@ -121,6 +121,25 @@ struct btrfs_inode {
121 /* the start of block group preferred for allocations. */ 121 /* the start of block group preferred for allocations. */
122 u64 block_group; 122 u64 block_group;
123 123
124 /* the fsync log has some corner cases that mean we have to check
125 * directories to see if any unlinks have been done before
126 * the directory was logged. See tree-log.c for all the
127 * details
128 */
129 u64 last_unlink_trans;
130
131 /*
132 * ordered_data_close is set by truncate when a file that used
133 * to have good data has been truncated to zero. When it is set
134 * the btrfs file release call will add this inode to the
135 * ordered operations list so that we make sure to flush out any
136 * new data the application may have written before commit.
137 *
138 * yes, its silly to have a single bitflag, but we might grow more
139 * of these.
140 */
141 unsigned ordered_data_close:1;
142
124 struct inode vfs_inode; 143 struct inode vfs_inode;
125}; 144};
126 145
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 37f31b5529aa..dbb724124633 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -254,18 +254,13 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
254 * empty_size -- a hint that you plan on doing more cow. This is the size in 254 * empty_size -- a hint that you plan on doing more cow. This is the size in
255 * bytes the allocator should try to find free next to the block it returns. 255 * bytes the allocator should try to find free next to the block it returns.
256 * This is just a hint and may be ignored by the allocator. 256 * This is just a hint and may be ignored by the allocator.
257 *
258 * prealloc_dest -- if you have already reserved a destination for the cow,
259 * this uses that block instead of allocating a new one.
260 * btrfs_alloc_reserved_extent is used to finish the allocation.
261 */ 257 */
262static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 258static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
263 struct btrfs_root *root, 259 struct btrfs_root *root,
264 struct extent_buffer *buf, 260 struct extent_buffer *buf,
265 struct extent_buffer *parent, int parent_slot, 261 struct extent_buffer *parent, int parent_slot,
266 struct extent_buffer **cow_ret, 262 struct extent_buffer **cow_ret,
267 u64 search_start, u64 empty_size, 263 u64 search_start, u64 empty_size)
268 u64 prealloc_dest)
269{ 264{
270 u64 parent_start; 265 u64 parent_start;
271 struct extent_buffer *cow; 266 struct extent_buffer *cow;
@@ -291,26 +286,10 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
291 level = btrfs_header_level(buf); 286 level = btrfs_header_level(buf);
292 nritems = btrfs_header_nritems(buf); 287 nritems = btrfs_header_nritems(buf);
293 288
294 if (prealloc_dest) { 289 cow = btrfs_alloc_free_block(trans, root, buf->len,
295 struct btrfs_key ins; 290 parent_start, root->root_key.objectid,
296 291 trans->transid, level,
297 ins.objectid = prealloc_dest; 292 search_start, empty_size);
298 ins.offset = buf->len;
299 ins.type = BTRFS_EXTENT_ITEM_KEY;
300
301 ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
302 root->root_key.objectid,
303 trans->transid, level, &ins);
304 BUG_ON(ret);
305 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
306 buf->len, level);
307 } else {
308 cow = btrfs_alloc_free_block(trans, root, buf->len,
309 parent_start,
310 root->root_key.objectid,
311 trans->transid, level,
312 search_start, empty_size);
313 }
314 if (IS_ERR(cow)) 293 if (IS_ERR(cow))
315 return PTR_ERR(cow); 294 return PTR_ERR(cow);
316 295
@@ -413,7 +392,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
413noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 392noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root, struct extent_buffer *buf, 393 struct btrfs_root *root, struct extent_buffer *buf,
415 struct extent_buffer *parent, int parent_slot, 394 struct extent_buffer *parent, int parent_slot,
416 struct extent_buffer **cow_ret, u64 prealloc_dest) 395 struct extent_buffer **cow_ret)
417{ 396{
418 u64 search_start; 397 u64 search_start;
419 int ret; 398 int ret;
@@ -436,7 +415,6 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
436 btrfs_header_owner(buf) == root->root_key.objectid && 415 btrfs_header_owner(buf) == root->root_key.objectid &&
437 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { 416 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
438 *cow_ret = buf; 417 *cow_ret = buf;
439 WARN_ON(prealloc_dest);
440 return 0; 418 return 0;
441 } 419 }
442 420
@@ -447,8 +425,7 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
447 btrfs_set_lock_blocking(buf); 425 btrfs_set_lock_blocking(buf);
448 426
449 ret = __btrfs_cow_block(trans, root, buf, parent, 427 ret = __btrfs_cow_block(trans, root, buf, parent,
450 parent_slot, cow_ret, search_start, 0, 428 parent_slot, cow_ret, search_start, 0);
451 prealloc_dest);
452 return ret; 429 return ret;
453} 430}
454 431
@@ -617,7 +594,7 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
617 err = __btrfs_cow_block(trans, root, cur, parent, i, 594 err = __btrfs_cow_block(trans, root, cur, parent, i,
618 &cur, search_start, 595 &cur, search_start,
619 min(16 * blocksize, 596 min(16 * blocksize,
620 (end_slot - i) * blocksize), 0); 597 (end_slot - i) * blocksize));
621 if (err) { 598 if (err) {
622 btrfs_tree_unlock(cur); 599 btrfs_tree_unlock(cur);
623 free_extent_buffer(cur); 600 free_extent_buffer(cur);
@@ -937,7 +914,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
937 BUG_ON(!child); 914 BUG_ON(!child);
938 btrfs_tree_lock(child); 915 btrfs_tree_lock(child);
939 btrfs_set_lock_blocking(child); 916 btrfs_set_lock_blocking(child);
940 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0); 917 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
941 BUG_ON(ret); 918 BUG_ON(ret);
942 919
943 spin_lock(&root->node_lock); 920 spin_lock(&root->node_lock);
@@ -945,6 +922,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
945 spin_unlock(&root->node_lock); 922 spin_unlock(&root->node_lock);
946 923
947 ret = btrfs_update_extent_ref(trans, root, child->start, 924 ret = btrfs_update_extent_ref(trans, root, child->start,
925 child->len,
948 mid->start, child->start, 926 mid->start, child->start,
949 root->root_key.objectid, 927 root->root_key.objectid,
950 trans->transid, level - 1); 928 trans->transid, level - 1);
@@ -971,6 +949,10 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
971 BTRFS_NODEPTRS_PER_BLOCK(root) / 4) 949 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
972 return 0; 950 return 0;
973 951
952 if (trans->transaction->delayed_refs.flushing &&
953 btrfs_header_nritems(mid) > 2)
954 return 0;
955
974 if (btrfs_header_nritems(mid) < 2) 956 if (btrfs_header_nritems(mid) < 2)
975 err_on_enospc = 1; 957 err_on_enospc = 1;
976 958
@@ -979,7 +961,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
979 btrfs_tree_lock(left); 961 btrfs_tree_lock(left);
980 btrfs_set_lock_blocking(left); 962 btrfs_set_lock_blocking(left);
981 wret = btrfs_cow_block(trans, root, left, 963 wret = btrfs_cow_block(trans, root, left,
982 parent, pslot - 1, &left, 0); 964 parent, pslot - 1, &left);
983 if (wret) { 965 if (wret) {
984 ret = wret; 966 ret = wret;
985 goto enospc; 967 goto enospc;
@@ -990,7 +972,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
990 btrfs_tree_lock(right); 972 btrfs_tree_lock(right);
991 btrfs_set_lock_blocking(right); 973 btrfs_set_lock_blocking(right);
992 wret = btrfs_cow_block(trans, root, right, 974 wret = btrfs_cow_block(trans, root, right,
993 parent, pslot + 1, &right, 0); 975 parent, pslot + 1, &right);
994 if (wret) { 976 if (wret) {
995 ret = wret; 977 ret = wret;
996 goto enospc; 978 goto enospc;
@@ -1171,7 +1153,7 @@ static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1171 wret = 1; 1153 wret = 1;
1172 } else { 1154 } else {
1173 ret = btrfs_cow_block(trans, root, left, parent, 1155 ret = btrfs_cow_block(trans, root, left, parent,
1174 pslot - 1, &left, 0); 1156 pslot - 1, &left);
1175 if (ret) 1157 if (ret)
1176 wret = 1; 1158 wret = 1;
1177 else { 1159 else {
@@ -1222,7 +1204,7 @@ static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1222 } else { 1204 } else {
1223 ret = btrfs_cow_block(trans, root, right, 1205 ret = btrfs_cow_block(trans, root, right,
1224 parent, pslot + 1, 1206 parent, pslot + 1,
1225 &right, 0); 1207 &right);
1226 if (ret) 1208 if (ret)
1227 wret = 1; 1209 wret = 1;
1228 else { 1210 else {
@@ -1492,7 +1474,6 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1492 u8 lowest_level = 0; 1474 u8 lowest_level = 0;
1493 u64 blocknr; 1475 u64 blocknr;
1494 u64 gen; 1476 u64 gen;
1495 struct btrfs_key prealloc_block;
1496 1477
1497 lowest_level = p->lowest_level; 1478 lowest_level = p->lowest_level;
1498 WARN_ON(lowest_level && ins_len > 0); 1479 WARN_ON(lowest_level && ins_len > 0);
@@ -1501,8 +1482,6 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1501 if (ins_len < 0) 1482 if (ins_len < 0)
1502 lowest_unlock = 2; 1483 lowest_unlock = 2;
1503 1484
1504 prealloc_block.objectid = 0;
1505
1506again: 1485again:
1507 if (p->skip_locking) 1486 if (p->skip_locking)
1508 b = btrfs_root_node(root); 1487 b = btrfs_root_node(root);
@@ -1529,44 +1508,11 @@ again:
1529 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) { 1508 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1530 goto cow_done; 1509 goto cow_done;
1531 } 1510 }
1532
1533 /* ok, we have to cow, is our old prealloc the right
1534 * size?
1535 */
1536 if (prealloc_block.objectid &&
1537 prealloc_block.offset != b->len) {
1538 btrfs_release_path(root, p);
1539 btrfs_free_reserved_extent(root,
1540 prealloc_block.objectid,
1541 prealloc_block.offset);
1542 prealloc_block.objectid = 0;
1543 goto again;
1544 }
1545
1546 /*
1547 * for higher level blocks, try not to allocate blocks
1548 * with the block and the parent locks held.
1549 */
1550 if (level > 0 && !prealloc_block.objectid) {
1551 u32 size = b->len;
1552 u64 hint = b->start;
1553
1554 btrfs_release_path(root, p);
1555 ret = btrfs_reserve_extent(trans, root,
1556 size, size, 0,
1557 hint, (u64)-1,
1558 &prealloc_block, 0);
1559 BUG_ON(ret);
1560 goto again;
1561 }
1562
1563 btrfs_set_path_blocking(p); 1511 btrfs_set_path_blocking(p);
1564 1512
1565 wret = btrfs_cow_block(trans, root, b, 1513 wret = btrfs_cow_block(trans, root, b,
1566 p->nodes[level + 1], 1514 p->nodes[level + 1],
1567 p->slots[level + 1], 1515 p->slots[level + 1], &b);
1568 &b, prealloc_block.objectid);
1569 prealloc_block.objectid = 0;
1570 if (wret) { 1516 if (wret) {
1571 free_extent_buffer(b); 1517 free_extent_buffer(b);
1572 ret = wret; 1518 ret = wret;
@@ -1742,12 +1688,8 @@ done:
1742 * we don't really know what they plan on doing with the path 1688 * we don't really know what they plan on doing with the path
1743 * from here on, so for now just mark it as blocking 1689 * from here on, so for now just mark it as blocking
1744 */ 1690 */
1745 btrfs_set_path_blocking(p); 1691 if (!p->leave_spinning)
1746 if (prealloc_block.objectid) { 1692 btrfs_set_path_blocking(p);
1747 btrfs_free_reserved_extent(root,
1748 prealloc_block.objectid,
1749 prealloc_block.offset);
1750 }
1751 return ret; 1693 return ret;
1752} 1694}
1753 1695
@@ -1768,7 +1710,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans,
1768 int ret; 1710 int ret;
1769 1711
1770 eb = btrfs_lock_root_node(root); 1712 eb = btrfs_lock_root_node(root);
1771 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0); 1713 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
1772 BUG_ON(ret); 1714 BUG_ON(ret);
1773 1715
1774 btrfs_set_lock_blocking(eb); 1716 btrfs_set_lock_blocking(eb);
@@ -1826,7 +1768,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans,
1826 } 1768 }
1827 1769
1828 ret = btrfs_cow_block(trans, root, eb, parent, slot, 1770 ret = btrfs_cow_block(trans, root, eb, parent, slot,
1829 &eb, 0); 1771 &eb);
1830 BUG_ON(ret); 1772 BUG_ON(ret);
1831 1773
1832 if (root->root_key.objectid == 1774 if (root->root_key.objectid ==
@@ -2139,7 +2081,7 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2139 spin_unlock(&root->node_lock); 2081 spin_unlock(&root->node_lock);
2140 2082
2141 ret = btrfs_update_extent_ref(trans, root, lower->start, 2083 ret = btrfs_update_extent_ref(trans, root, lower->start,
2142 lower->start, c->start, 2084 lower->len, lower->start, c->start,
2143 root->root_key.objectid, 2085 root->root_key.objectid,
2144 trans->transid, level - 1); 2086 trans->transid, level - 1);
2145 BUG_ON(ret); 2087 BUG_ON(ret);
@@ -2221,7 +2163,7 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
2221 ret = insert_new_root(trans, root, path, level + 1); 2163 ret = insert_new_root(trans, root, path, level + 1);
2222 if (ret) 2164 if (ret)
2223 return ret; 2165 return ret;
2224 } else { 2166 } else if (!trans->transaction->delayed_refs.flushing) {
2225 ret = push_nodes_for_insert(trans, root, path, level); 2167 ret = push_nodes_for_insert(trans, root, path, level);
2226 c = path->nodes[level]; 2168 c = path->nodes[level];
2227 if (!ret && btrfs_header_nritems(c) < 2169 if (!ret && btrfs_header_nritems(c) <
@@ -2329,66 +2271,27 @@ noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2329 return ret; 2271 return ret;
2330} 2272}
2331 2273
2332/* 2274static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
2333 * push some data in the path leaf to the right, trying to free up at 2275 struct btrfs_root *root,
2334 * least data_size bytes. returns zero if the push worked, nonzero otherwise 2276 struct btrfs_path *path,
2335 * 2277 int data_size, int empty,
2336 * returns 1 if the push failed because the other node didn't have enough 2278 struct extent_buffer *right,
2337 * room, 0 if everything worked out and < 0 if there were major errors. 2279 int free_space, u32 left_nritems)
2338 */
2339static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2340 *root, struct btrfs_path *path, int data_size,
2341 int empty)
2342{ 2280{
2343 struct extent_buffer *left = path->nodes[0]; 2281 struct extent_buffer *left = path->nodes[0];
2344 struct extent_buffer *right; 2282 struct extent_buffer *upper = path->nodes[1];
2345 struct extent_buffer *upper;
2346 struct btrfs_disk_key disk_key; 2283 struct btrfs_disk_key disk_key;
2347 int slot; 2284 int slot;
2348 u32 i; 2285 u32 i;
2349 int free_space;
2350 int push_space = 0; 2286 int push_space = 0;
2351 int push_items = 0; 2287 int push_items = 0;
2352 struct btrfs_item *item; 2288 struct btrfs_item *item;
2353 u32 left_nritems;
2354 u32 nr; 2289 u32 nr;
2355 u32 right_nritems; 2290 u32 right_nritems;
2356 u32 data_end; 2291 u32 data_end;
2357 u32 this_item_size; 2292 u32 this_item_size;
2358 int ret; 2293 int ret;
2359 2294
2360 slot = path->slots[1];
2361 if (!path->nodes[1])
2362 return 1;
2363
2364 upper = path->nodes[1];
2365 if (slot >= btrfs_header_nritems(upper) - 1)
2366 return 1;
2367
2368 btrfs_assert_tree_locked(path->nodes[1]);
2369
2370 right = read_node_slot(root, upper, slot + 1);
2371 btrfs_tree_lock(right);
2372 btrfs_set_lock_blocking(right);
2373
2374 free_space = btrfs_leaf_free_space(root, right);
2375 if (free_space < data_size)
2376 goto out_unlock;
2377
2378 /* cow and double check */
2379 ret = btrfs_cow_block(trans, root, right, upper,
2380 slot + 1, &right, 0);
2381 if (ret)
2382 goto out_unlock;
2383
2384 free_space = btrfs_leaf_free_space(root, right);
2385 if (free_space < data_size)
2386 goto out_unlock;
2387
2388 left_nritems = btrfs_header_nritems(left);
2389 if (left_nritems == 0)
2390 goto out_unlock;
2391
2392 if (empty) 2295 if (empty)
2393 nr = 0; 2296 nr = 0;
2394 else 2297 else
@@ -2397,6 +2300,7 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2397 if (path->slots[0] >= left_nritems) 2300 if (path->slots[0] >= left_nritems)
2398 push_space += data_size; 2301 push_space += data_size;
2399 2302
2303 slot = path->slots[1];
2400 i = left_nritems - 1; 2304 i = left_nritems - 1;
2401 while (i >= nr) { 2305 while (i >= nr) {
2402 item = btrfs_item_nr(left, i); 2306 item = btrfs_item_nr(left, i);
@@ -2528,24 +2432,82 @@ out_unlock:
2528} 2432}
2529 2433
2530/* 2434/*
2435 * push some data in the path leaf to the right, trying to free up at
2436 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2437 *
2438 * returns 1 if the push failed because the other node didn't have enough
2439 * room, 0 if everything worked out and < 0 if there were major errors.
2440 */
2441static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2442 *root, struct btrfs_path *path, int data_size,
2443 int empty)
2444{
2445 struct extent_buffer *left = path->nodes[0];
2446 struct extent_buffer *right;
2447 struct extent_buffer *upper;
2448 int slot;
2449 int free_space;
2450 u32 left_nritems;
2451 int ret;
2452
2453 if (!path->nodes[1])
2454 return 1;
2455
2456 slot = path->slots[1];
2457 upper = path->nodes[1];
2458 if (slot >= btrfs_header_nritems(upper) - 1)
2459 return 1;
2460
2461 btrfs_assert_tree_locked(path->nodes[1]);
2462
2463 right = read_node_slot(root, upper, slot + 1);
2464 btrfs_tree_lock(right);
2465 btrfs_set_lock_blocking(right);
2466
2467 free_space = btrfs_leaf_free_space(root, right);
2468 if (free_space < data_size)
2469 goto out_unlock;
2470
2471 /* cow and double check */
2472 ret = btrfs_cow_block(trans, root, right, upper,
2473 slot + 1, &right);
2474 if (ret)
2475 goto out_unlock;
2476
2477 free_space = btrfs_leaf_free_space(root, right);
2478 if (free_space < data_size)
2479 goto out_unlock;
2480
2481 left_nritems = btrfs_header_nritems(left);
2482 if (left_nritems == 0)
2483 goto out_unlock;
2484
2485 return __push_leaf_right(trans, root, path, data_size, empty,
2486 right, free_space, left_nritems);
2487out_unlock:
2488 btrfs_tree_unlock(right);
2489 free_extent_buffer(right);
2490 return 1;
2491}
2492
2493/*
2531 * push some data in the path leaf to the left, trying to free up at 2494 * push some data in the path leaf to the left, trying to free up at
2532 * least data_size bytes. returns zero if the push worked, nonzero otherwise 2495 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2533 */ 2496 */
2534static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 2497static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
2535 *root, struct btrfs_path *path, int data_size, 2498 struct btrfs_root *root,
2536 int empty) 2499 struct btrfs_path *path, int data_size,
2500 int empty, struct extent_buffer *left,
2501 int free_space, int right_nritems)
2537{ 2502{
2538 struct btrfs_disk_key disk_key; 2503 struct btrfs_disk_key disk_key;
2539 struct extent_buffer *right = path->nodes[0]; 2504 struct extent_buffer *right = path->nodes[0];
2540 struct extent_buffer *left;
2541 int slot; 2505 int slot;
2542 int i; 2506 int i;
2543 int free_space;
2544 int push_space = 0; 2507 int push_space = 0;
2545 int push_items = 0; 2508 int push_items = 0;
2546 struct btrfs_item *item; 2509 struct btrfs_item *item;
2547 u32 old_left_nritems; 2510 u32 old_left_nritems;
2548 u32 right_nritems;
2549 u32 nr; 2511 u32 nr;
2550 int ret = 0; 2512 int ret = 0;
2551 int wret; 2513 int wret;
@@ -2553,41 +2515,6 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2553 u32 old_left_item_size; 2515 u32 old_left_item_size;
2554 2516
2555 slot = path->slots[1]; 2517 slot = path->slots[1];
2556 if (slot == 0)
2557 return 1;
2558 if (!path->nodes[1])
2559 return 1;
2560
2561 right_nritems = btrfs_header_nritems(right);
2562 if (right_nritems == 0)
2563 return 1;
2564
2565 btrfs_assert_tree_locked(path->nodes[1]);
2566
2567 left = read_node_slot(root, path->nodes[1], slot - 1);
2568 btrfs_tree_lock(left);
2569 btrfs_set_lock_blocking(left);
2570
2571 free_space = btrfs_leaf_free_space(root, left);
2572 if (free_space < data_size) {
2573 ret = 1;
2574 goto out;
2575 }
2576
2577 /* cow and double check */
2578 ret = btrfs_cow_block(trans, root, left,
2579 path->nodes[1], slot - 1, &left, 0);
2580 if (ret) {
2581 /* we hit -ENOSPC, but it isn't fatal here */
2582 ret = 1;
2583 goto out;
2584 }
2585
2586 free_space = btrfs_leaf_free_space(root, left);
2587 if (free_space < data_size) {
2588 ret = 1;
2589 goto out;
2590 }
2591 2518
2592 if (empty) 2519 if (empty)
2593 nr = right_nritems; 2520 nr = right_nritems;
@@ -2755,6 +2682,154 @@ out:
2755} 2682}
2756 2683
2757/* 2684/*
2685 * push some data in the path leaf to the left, trying to free up at
2686 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2687 */
2688static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2689 *root, struct btrfs_path *path, int data_size,
2690 int empty)
2691{
2692 struct extent_buffer *right = path->nodes[0];
2693 struct extent_buffer *left;
2694 int slot;
2695 int free_space;
2696 u32 right_nritems;
2697 int ret = 0;
2698
2699 slot = path->slots[1];
2700 if (slot == 0)
2701 return 1;
2702 if (!path->nodes[1])
2703 return 1;
2704
2705 right_nritems = btrfs_header_nritems(right);
2706 if (right_nritems == 0)
2707 return 1;
2708
2709 btrfs_assert_tree_locked(path->nodes[1]);
2710
2711 left = read_node_slot(root, path->nodes[1], slot - 1);
2712 btrfs_tree_lock(left);
2713 btrfs_set_lock_blocking(left);
2714
2715 free_space = btrfs_leaf_free_space(root, left);
2716 if (free_space < data_size) {
2717 ret = 1;
2718 goto out;
2719 }
2720
2721 /* cow and double check */
2722 ret = btrfs_cow_block(trans, root, left,
2723 path->nodes[1], slot - 1, &left);
2724 if (ret) {
2725 /* we hit -ENOSPC, but it isn't fatal here */
2726 ret = 1;
2727 goto out;
2728 }
2729
2730 free_space = btrfs_leaf_free_space(root, left);
2731 if (free_space < data_size) {
2732 ret = 1;
2733 goto out;
2734 }
2735
2736 return __push_leaf_left(trans, root, path, data_size,
2737 empty, left, free_space, right_nritems);
2738out:
2739 btrfs_tree_unlock(left);
2740 free_extent_buffer(left);
2741 return ret;
2742}
2743
2744/*
2745 * split the path's leaf in two, making sure there is at least data_size
2746 * available for the resulting leaf level of the path.
2747 *
2748 * returns 0 if all went well and < 0 on failure.
2749 */
2750static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2751 struct btrfs_root *root,
2752 struct btrfs_path *path,
2753 struct extent_buffer *l,
2754 struct extent_buffer *right,
2755 int slot, int mid, int nritems)
2756{
2757 int data_copy_size;
2758 int rt_data_off;
2759 int i;
2760 int ret = 0;
2761 int wret;
2762 struct btrfs_disk_key disk_key;
2763
2764 nritems = nritems - mid;
2765 btrfs_set_header_nritems(right, nritems);
2766 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2767
2768 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2769 btrfs_item_nr_offset(mid),
2770 nritems * sizeof(struct btrfs_item));
2771
2772 copy_extent_buffer(right, l,
2773 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2774 data_copy_size, btrfs_leaf_data(l) +
2775 leaf_data_end(root, l), data_copy_size);
2776
2777 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2778 btrfs_item_end_nr(l, mid);
2779
2780 for (i = 0; i < nritems; i++) {
2781 struct btrfs_item *item = btrfs_item_nr(right, i);
2782 u32 ioff;
2783
2784 if (!right->map_token) {
2785 map_extent_buffer(right, (unsigned long)item,
2786 sizeof(struct btrfs_item),
2787 &right->map_token, &right->kaddr,
2788 &right->map_start, &right->map_len,
2789 KM_USER1);
2790 }
2791
2792 ioff = btrfs_item_offset(right, item);
2793 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2794 }
2795
2796 if (right->map_token) {
2797 unmap_extent_buffer(right, right->map_token, KM_USER1);
2798 right->map_token = NULL;
2799 }
2800
2801 btrfs_set_header_nritems(l, mid);
2802 ret = 0;
2803 btrfs_item_key(right, &disk_key, 0);
2804 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2805 path->slots[1] + 1, 1);
2806 if (wret)
2807 ret = wret;
2808
2809 btrfs_mark_buffer_dirty(right);
2810 btrfs_mark_buffer_dirty(l);
2811 BUG_ON(path->slots[0] != slot);
2812
2813 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2814 BUG_ON(ret);
2815
2816 if (mid <= slot) {
2817 btrfs_tree_unlock(path->nodes[0]);
2818 free_extent_buffer(path->nodes[0]);
2819 path->nodes[0] = right;
2820 path->slots[0] -= mid;
2821 path->slots[1] += 1;
2822 } else {
2823 btrfs_tree_unlock(right);
2824 free_extent_buffer(right);
2825 }
2826
2827 BUG_ON(path->slots[0] < 0);
2828
2829 return ret;
2830}
2831
2832/*
2758 * split the path's leaf in two, making sure there is at least data_size 2833 * split the path's leaf in two, making sure there is at least data_size
2759 * available for the resulting leaf level of the path. 2834 * available for the resulting leaf level of the path.
2760 * 2835 *
@@ -2771,17 +2846,14 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
2771 int mid; 2846 int mid;
2772 int slot; 2847 int slot;
2773 struct extent_buffer *right; 2848 struct extent_buffer *right;
2774 int data_copy_size;
2775 int rt_data_off;
2776 int i;
2777 int ret = 0; 2849 int ret = 0;
2778 int wret; 2850 int wret;
2779 int double_split; 2851 int double_split;
2780 int num_doubles = 0; 2852 int num_doubles = 0;
2781 struct btrfs_disk_key disk_key;
2782 2853
2783 /* first try to make some room by pushing left and right */ 2854 /* first try to make some room by pushing left and right */
2784 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) { 2855 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY &&
2856 !trans->transaction->delayed_refs.flushing) {
2785 wret = push_leaf_right(trans, root, path, data_size, 0); 2857 wret = push_leaf_right(trans, root, path, data_size, 0);
2786 if (wret < 0) 2858 if (wret < 0)
2787 return wret; 2859 return wret;
@@ -2830,11 +2902,14 @@ again:
2830 write_extent_buffer(right, root->fs_info->chunk_tree_uuid, 2902 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2831 (unsigned long)btrfs_header_chunk_tree_uuid(right), 2903 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2832 BTRFS_UUID_SIZE); 2904 BTRFS_UUID_SIZE);
2905
2833 if (mid <= slot) { 2906 if (mid <= slot) {
2834 if (nritems == 1 || 2907 if (nritems == 1 ||
2835 leaf_space_used(l, mid, nritems - mid) + data_size > 2908 leaf_space_used(l, mid, nritems - mid) + data_size >
2836 BTRFS_LEAF_DATA_SIZE(root)) { 2909 BTRFS_LEAF_DATA_SIZE(root)) {
2837 if (slot >= nritems) { 2910 if (slot >= nritems) {
2911 struct btrfs_disk_key disk_key;
2912
2838 btrfs_cpu_key_to_disk(&disk_key, ins_key); 2913 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2839 btrfs_set_header_nritems(right, 0); 2914 btrfs_set_header_nritems(right, 0);
2840 wret = insert_ptr(trans, root, path, 2915 wret = insert_ptr(trans, root, path,
@@ -2862,6 +2937,8 @@ again:
2862 if (leaf_space_used(l, 0, mid) + data_size > 2937 if (leaf_space_used(l, 0, mid) + data_size >
2863 BTRFS_LEAF_DATA_SIZE(root)) { 2938 BTRFS_LEAF_DATA_SIZE(root)) {
2864 if (!extend && data_size && slot == 0) { 2939 if (!extend && data_size && slot == 0) {
2940 struct btrfs_disk_key disk_key;
2941
2865 btrfs_cpu_key_to_disk(&disk_key, ins_key); 2942 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2866 btrfs_set_header_nritems(right, 0); 2943 btrfs_set_header_nritems(right, 0);
2867 wret = insert_ptr(trans, root, path, 2944 wret = insert_ptr(trans, root, path,
@@ -2894,76 +2971,16 @@ again:
2894 } 2971 }
2895 } 2972 }
2896 } 2973 }
2897 nritems = nritems - mid;
2898 btrfs_set_header_nritems(right, nritems);
2899 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2900
2901 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2902 btrfs_item_nr_offset(mid),
2903 nritems * sizeof(struct btrfs_item));
2904
2905 copy_extent_buffer(right, l,
2906 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2907 data_copy_size, btrfs_leaf_data(l) +
2908 leaf_data_end(root, l), data_copy_size);
2909
2910 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2911 btrfs_item_end_nr(l, mid);
2912
2913 for (i = 0; i < nritems; i++) {
2914 struct btrfs_item *item = btrfs_item_nr(right, i);
2915 u32 ioff;
2916
2917 if (!right->map_token) {
2918 map_extent_buffer(right, (unsigned long)item,
2919 sizeof(struct btrfs_item),
2920 &right->map_token, &right->kaddr,
2921 &right->map_start, &right->map_len,
2922 KM_USER1);
2923 }
2924
2925 ioff = btrfs_item_offset(right, item);
2926 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2927 }
2928
2929 if (right->map_token) {
2930 unmap_extent_buffer(right, right->map_token, KM_USER1);
2931 right->map_token = NULL;
2932 }
2933
2934 btrfs_set_header_nritems(l, mid);
2935 ret = 0;
2936 btrfs_item_key(right, &disk_key, 0);
2937 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2938 path->slots[1] + 1, 1);
2939 if (wret)
2940 ret = wret;
2941
2942 btrfs_mark_buffer_dirty(right);
2943 btrfs_mark_buffer_dirty(l);
2944 BUG_ON(path->slots[0] != slot);
2945 2974
2946 ret = btrfs_update_ref(trans, root, l, right, 0, nritems); 2975 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2947 BUG_ON(ret); 2976 BUG_ON(ret);
2948 2977
2949 if (mid <= slot) {
2950 btrfs_tree_unlock(path->nodes[0]);
2951 free_extent_buffer(path->nodes[0]);
2952 path->nodes[0] = right;
2953 path->slots[0] -= mid;
2954 path->slots[1] += 1;
2955 } else {
2956 btrfs_tree_unlock(right);
2957 free_extent_buffer(right);
2958 }
2959
2960 BUG_ON(path->slots[0] < 0);
2961
2962 if (double_split) { 2978 if (double_split) {
2963 BUG_ON(num_doubles != 0); 2979 BUG_ON(num_doubles != 0);
2964 num_doubles++; 2980 num_doubles++;
2965 goto again; 2981 goto again;
2966 } 2982 }
2983
2967 return ret; 2984 return ret;
2968} 2985}
2969 2986
@@ -3021,26 +3038,27 @@ int btrfs_split_item(struct btrfs_trans_handle *trans,
3021 return -EAGAIN; 3038 return -EAGAIN;
3022 } 3039 }
3023 3040
3041 btrfs_set_path_blocking(path);
3024 ret = split_leaf(trans, root, &orig_key, path, 3042 ret = split_leaf(trans, root, &orig_key, path,
3025 sizeof(struct btrfs_item), 1); 3043 sizeof(struct btrfs_item), 1);
3026 path->keep_locks = 0; 3044 path->keep_locks = 0;
3027 BUG_ON(ret); 3045 BUG_ON(ret);
3028 3046
3047 btrfs_unlock_up_safe(path, 1);
3048 leaf = path->nodes[0];
3049 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3050
3051split:
3029 /* 3052 /*
3030 * make sure any changes to the path from split_leaf leave it 3053 * make sure any changes to the path from split_leaf leave it
3031 * in a blocking state 3054 * in a blocking state
3032 */ 3055 */
3033 btrfs_set_path_blocking(path); 3056 btrfs_set_path_blocking(path);
3034 3057
3035 leaf = path->nodes[0];
3036 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3037
3038split:
3039 item = btrfs_item_nr(leaf, path->slots[0]); 3058 item = btrfs_item_nr(leaf, path->slots[0]);
3040 orig_offset = btrfs_item_offset(leaf, item); 3059 orig_offset = btrfs_item_offset(leaf, item);
3041 item_size = btrfs_item_size(leaf, item); 3060 item_size = btrfs_item_size(leaf, item);
3042 3061
3043
3044 buf = kmalloc(item_size, GFP_NOFS); 3062 buf = kmalloc(item_size, GFP_NOFS);
3045 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 3063 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3046 path->slots[0]), item_size); 3064 path->slots[0]), item_size);
@@ -3445,39 +3463,27 @@ out:
3445} 3463}
3446 3464
3447/* 3465/*
3448 * Given a key and some data, insert items into the tree. 3466 * this is a helper for btrfs_insert_empty_items, the main goal here is
3449 * This does all the path init required, making room in the tree if needed. 3467 * to save stack depth by doing the bulk of the work in a function
3468 * that doesn't call btrfs_search_slot
3450 */ 3469 */
3451int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 3470static noinline_for_stack int
3452 struct btrfs_root *root, 3471setup_items_for_insert(struct btrfs_trans_handle *trans,
3453 struct btrfs_path *path, 3472 struct btrfs_root *root, struct btrfs_path *path,
3454 struct btrfs_key *cpu_key, u32 *data_size, 3473 struct btrfs_key *cpu_key, u32 *data_size,
3455 int nr) 3474 u32 total_data, u32 total_size, int nr)
3456{ 3475{
3457 struct extent_buffer *leaf;
3458 struct btrfs_item *item; 3476 struct btrfs_item *item;
3459 int ret = 0;
3460 int slot;
3461 int slot_orig;
3462 int i; 3477 int i;
3463 u32 nritems; 3478 u32 nritems;
3464 u32 total_size = 0;
3465 u32 total_data = 0;
3466 unsigned int data_end; 3479 unsigned int data_end;
3467 struct btrfs_disk_key disk_key; 3480 struct btrfs_disk_key disk_key;
3481 int ret;
3482 struct extent_buffer *leaf;
3483 int slot;
3468 3484
3469 for (i = 0; i < nr; i++)
3470 total_data += data_size[i];
3471
3472 total_size = total_data + (nr * sizeof(struct btrfs_item));
3473 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3474 if (ret == 0)
3475 return -EEXIST;
3476 if (ret < 0)
3477 goto out;
3478
3479 slot_orig = path->slots[0];
3480 leaf = path->nodes[0]; 3485 leaf = path->nodes[0];
3486 slot = path->slots[0];
3481 3487
3482 nritems = btrfs_header_nritems(leaf); 3488 nritems = btrfs_header_nritems(leaf);
3483 data_end = leaf_data_end(root, leaf); 3489 data_end = leaf_data_end(root, leaf);
@@ -3489,9 +3495,6 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3489 BUG(); 3495 BUG();
3490 } 3496 }
3491 3497
3492 slot = path->slots[0];
3493 BUG_ON(slot < 0);
3494
3495 if (slot != nritems) { 3498 if (slot != nritems) {
3496 unsigned int old_data = btrfs_item_end_nr(leaf, slot); 3499 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3497 3500
@@ -3547,21 +3550,60 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3547 data_end -= data_size[i]; 3550 data_end -= data_size[i];
3548 btrfs_set_item_size(leaf, item, data_size[i]); 3551 btrfs_set_item_size(leaf, item, data_size[i]);
3549 } 3552 }
3553
3550 btrfs_set_header_nritems(leaf, nritems + nr); 3554 btrfs_set_header_nritems(leaf, nritems + nr);
3551 btrfs_mark_buffer_dirty(leaf);
3552 3555
3553 ret = 0; 3556 ret = 0;
3554 if (slot == 0) { 3557 if (slot == 0) {
3558 struct btrfs_disk_key disk_key;
3555 btrfs_cpu_key_to_disk(&disk_key, cpu_key); 3559 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3556 ret = fixup_low_keys(trans, root, path, &disk_key, 1); 3560 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3557 } 3561 }
3562 btrfs_unlock_up_safe(path, 1);
3563 btrfs_mark_buffer_dirty(leaf);
3558 3564
3559 if (btrfs_leaf_free_space(root, leaf) < 0) { 3565 if (btrfs_leaf_free_space(root, leaf) < 0) {
3560 btrfs_print_leaf(root, leaf); 3566 btrfs_print_leaf(root, leaf);
3561 BUG(); 3567 BUG();
3562 } 3568 }
3569 return ret;
3570}
3571
3572/*
3573 * Given a key and some data, insert items into the tree.
3574 * This does all the path init required, making room in the tree if needed.
3575 */
3576int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3577 struct btrfs_root *root,
3578 struct btrfs_path *path,
3579 struct btrfs_key *cpu_key, u32 *data_size,
3580 int nr)
3581{
3582 struct extent_buffer *leaf;
3583 int ret = 0;
3584 int slot;
3585 int i;
3586 u32 total_size = 0;
3587 u32 total_data = 0;
3588
3589 for (i = 0; i < nr; i++)
3590 total_data += data_size[i];
3591
3592 total_size = total_data + (nr * sizeof(struct btrfs_item));
3593 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3594 if (ret == 0)
3595 return -EEXIST;
3596 if (ret < 0)
3597 goto out;
3598
3599 leaf = path->nodes[0];
3600 slot = path->slots[0];
3601 BUG_ON(slot < 0);
3602
3603 ret = setup_items_for_insert(trans, root, path, cpu_key, data_size,
3604 total_data, total_size, nr);
3605
3563out: 3606out:
3564 btrfs_unlock_up_safe(path, 1);
3565 return ret; 3607 return ret;
3566} 3608}
3567 3609
@@ -3749,7 +3791,8 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3749 } 3791 }
3750 3792
3751 /* delete the leaf if it is mostly empty */ 3793 /* delete the leaf if it is mostly empty */
3752 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) { 3794 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4 &&
3795 !trans->transaction->delayed_refs.flushing) {
3753 /* push_leaf_left fixes the path. 3796 /* push_leaf_left fixes the path.
3754 * make sure the path still points to our leaf 3797 * make sure the path still points to our leaf
3755 * for possible call to del_ptr below 3798 * for possible call to del_ptr below
@@ -3757,6 +3800,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3757 slot = path->slots[1]; 3800 slot = path->slots[1];
3758 extent_buffer_get(leaf); 3801 extent_buffer_get(leaf);
3759 3802
3803 btrfs_set_path_blocking(path);
3760 wret = push_leaf_left(trans, root, path, 1, 1); 3804 wret = push_leaf_left(trans, root, path, 1, 1);
3761 if (wret < 0 && wret != -ENOSPC) 3805 if (wret < 0 && wret != -ENOSPC)
3762 ret = wret; 3806 ret = wret;
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 7dd1b6d0bf32..9417713542a2 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -45,6 +45,13 @@ struct btrfs_ordered_sum;
45 45
46#define BTRFS_MAX_LEVEL 8 46#define BTRFS_MAX_LEVEL 8
47 47
48/*
49 * files bigger than this get some pre-flushing when they are added
50 * to the ordered operations list. That way we limit the total
51 * work done by the commit
52 */
53#define BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT (8 * 1024 * 1024)
54
48/* holds pointers to all of the tree roots */ 55/* holds pointers to all of the tree roots */
49#define BTRFS_ROOT_TREE_OBJECTID 1ULL 56#define BTRFS_ROOT_TREE_OBJECTID 1ULL
50 57
@@ -401,15 +408,16 @@ struct btrfs_path {
401 int locks[BTRFS_MAX_LEVEL]; 408 int locks[BTRFS_MAX_LEVEL];
402 int reada; 409 int reada;
403 /* keep some upper locks as we walk down */ 410 /* keep some upper locks as we walk down */
404 int keep_locks;
405 int skip_locking;
406 int lowest_level; 411 int lowest_level;
407 412
408 /* 413 /*
409 * set by btrfs_split_item, tells search_slot to keep all locks 414 * set by btrfs_split_item, tells search_slot to keep all locks
410 * and to force calls to keep space in the nodes 415 * and to force calls to keep space in the nodes
411 */ 416 */
412 int search_for_split; 417 unsigned int search_for_split:1;
418 unsigned int keep_locks:1;
419 unsigned int skip_locking:1;
420 unsigned int leave_spinning:1;
413}; 421};
414 422
415/* 423/*
@@ -688,15 +696,18 @@ struct btrfs_fs_info {
688 struct rb_root block_group_cache_tree; 696 struct rb_root block_group_cache_tree;
689 697
690 struct extent_io_tree pinned_extents; 698 struct extent_io_tree pinned_extents;
691 struct extent_io_tree pending_del;
692 struct extent_io_tree extent_ins;
693 699
694 /* logical->physical extent mapping */ 700 /* logical->physical extent mapping */
695 struct btrfs_mapping_tree mapping_tree; 701 struct btrfs_mapping_tree mapping_tree;
696 702
697 u64 generation; 703 u64 generation;
698 u64 last_trans_committed; 704 u64 last_trans_committed;
699 u64 last_trans_new_blockgroup; 705
706 /*
707 * this is updated to the current trans every time a full commit
708 * is required instead of the faster short fsync log commits
709 */
710 u64 last_trans_log_full_commit;
700 u64 open_ioctl_trans; 711 u64 open_ioctl_trans;
701 unsigned long mount_opt; 712 unsigned long mount_opt;
702 u64 max_extent; 713 u64 max_extent;
@@ -717,12 +728,21 @@ struct btrfs_fs_info {
717 struct mutex tree_log_mutex; 728 struct mutex tree_log_mutex;
718 struct mutex transaction_kthread_mutex; 729 struct mutex transaction_kthread_mutex;
719 struct mutex cleaner_mutex; 730 struct mutex cleaner_mutex;
720 struct mutex extent_ins_mutex;
721 struct mutex pinned_mutex; 731 struct mutex pinned_mutex;
722 struct mutex chunk_mutex; 732 struct mutex chunk_mutex;
723 struct mutex drop_mutex; 733 struct mutex drop_mutex;
724 struct mutex volume_mutex; 734 struct mutex volume_mutex;
725 struct mutex tree_reloc_mutex; 735 struct mutex tree_reloc_mutex;
736
737 /*
738 * this protects the ordered operations list only while we are
739 * processing all of the entries on it. This way we make
740 * sure the commit code doesn't find the list temporarily empty
741 * because another function happens to be doing non-waiting preflush
742 * before jumping into the main commit.
743 */
744 struct mutex ordered_operations_mutex;
745
726 struct list_head trans_list; 746 struct list_head trans_list;
727 struct list_head hashers; 747 struct list_head hashers;
728 struct list_head dead_roots; 748 struct list_head dead_roots;
@@ -737,10 +757,29 @@ struct btrfs_fs_info {
737 * ordered extents 757 * ordered extents
738 */ 758 */
739 spinlock_t ordered_extent_lock; 759 spinlock_t ordered_extent_lock;
760
761 /*
762 * all of the data=ordered extents pending writeback
763 * these can span multiple transactions and basically include
764 * every dirty data page that isn't from nodatacow
765 */
740 struct list_head ordered_extents; 766 struct list_head ordered_extents;
767
768 /*
769 * all of the inodes that have delalloc bytes. It is possible for
770 * this list to be empty even when there is still dirty data=ordered
771 * extents waiting to finish IO.
772 */
741 struct list_head delalloc_inodes; 773 struct list_head delalloc_inodes;
742 774
743 /* 775 /*
776 * special rename and truncate targets that must be on disk before
777 * we're allowed to commit. This is basically the ext3 style
778 * data=ordered list.
779 */
780 struct list_head ordered_operations;
781
782 /*
744 * there is a pool of worker threads for checksumming during writes 783 * there is a pool of worker threads for checksumming during writes
745 * and a pool for checksumming after reads. This is because readers 784 * and a pool for checksumming after reads. This is because readers
746 * can run with FS locks held, and the writers may be waiting for 785 * can run with FS locks held, and the writers may be waiting for
@@ -781,6 +820,11 @@ struct btrfs_fs_info {
781 atomic_t throttle_gen; 820 atomic_t throttle_gen;
782 821
783 u64 total_pinned; 822 u64 total_pinned;
823
824 /* protected by the delalloc lock, used to keep from writing
825 * metadata until there is a nice batch
826 */
827 u64 dirty_metadata_bytes;
784 struct list_head dirty_cowonly_roots; 828 struct list_head dirty_cowonly_roots;
785 829
786 struct btrfs_fs_devices *fs_devices; 830 struct btrfs_fs_devices *fs_devices;
@@ -1704,18 +1748,15 @@ static inline struct dentry *fdentry(struct file *file)
1704} 1748}
1705 1749
1706/* extent-tree.c */ 1750/* extent-tree.c */
1751int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1752 struct btrfs_root *root, unsigned long count);
1707int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len); 1753int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len);
1708int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root, u64 bytenr,
1710 u64 num_bytes, u32 *refs);
1711int btrfs_update_pinned_extents(struct btrfs_root *root, 1754int btrfs_update_pinned_extents(struct btrfs_root *root,
1712 u64 bytenr, u64 num, int pin); 1755 u64 bytenr, u64 num, int pin);
1713int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, 1756int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
1714 struct btrfs_root *root, struct extent_buffer *leaf); 1757 struct btrfs_root *root, struct extent_buffer *leaf);
1715int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, 1758int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1716 struct btrfs_root *root, u64 objectid, u64 bytenr); 1759 struct btrfs_root *root, u64 objectid, u64 bytenr);
1717int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1718 struct btrfs_root *root);
1719int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy); 1760int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy);
1720struct btrfs_block_group_cache *btrfs_lookup_block_group( 1761struct btrfs_block_group_cache *btrfs_lookup_block_group(
1721 struct btrfs_fs_info *info, 1762 struct btrfs_fs_info *info,
@@ -1777,7 +1818,7 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1777 u64 root_objectid, u64 ref_generation, 1818 u64 root_objectid, u64 ref_generation,
1778 u64 owner_objectid); 1819 u64 owner_objectid);
1779int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, 1820int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1780 struct btrfs_root *root, u64 bytenr, 1821 struct btrfs_root *root, u64 bytenr, u64 num_bytes,
1781 u64 orig_parent, u64 parent, 1822 u64 orig_parent, u64 parent,
1782 u64 root_objectid, u64 ref_generation, 1823 u64 root_objectid, u64 ref_generation,
1783 u64 owner_objectid); 1824 u64 owner_objectid);
@@ -1838,7 +1879,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
1838int btrfs_cow_block(struct btrfs_trans_handle *trans, 1879int btrfs_cow_block(struct btrfs_trans_handle *trans,
1839 struct btrfs_root *root, struct extent_buffer *buf, 1880 struct btrfs_root *root, struct extent_buffer *buf,
1840 struct extent_buffer *parent, int parent_slot, 1881 struct extent_buffer *parent, int parent_slot,
1841 struct extent_buffer **cow_ret, u64 prealloc_dest); 1882 struct extent_buffer **cow_ret);
1842int btrfs_copy_root(struct btrfs_trans_handle *trans, 1883int btrfs_copy_root(struct btrfs_trans_handle *trans,
1843 struct btrfs_root *root, 1884 struct btrfs_root *root,
1844 struct extent_buffer *buf, 1885 struct extent_buffer *buf,
diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
new file mode 100644
index 000000000000..cbf7dc8ae3ec
--- /dev/null
+++ b/fs/btrfs/delayed-ref.c
@@ -0,0 +1,669 @@
1/*
2 * Copyright (C) 2009 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 <linux/sort.h>
21#include <linux/ftrace.h>
22#include "ctree.h"
23#include "delayed-ref.h"
24#include "transaction.h"
25
26/*
27 * delayed back reference update tracking. For subvolume trees
28 * we queue up extent allocations and backref maintenance for
29 * delayed processing. This avoids deep call chains where we
30 * add extents in the middle of btrfs_search_slot, and it allows
31 * us to buffer up frequently modified backrefs in an rb tree instead
32 * of hammering updates on the extent allocation tree.
33 *
34 * Right now this code is only used for reference counted trees, but
35 * the long term goal is to get rid of the similar code for delayed
36 * extent tree modifications.
37 */
38
39/*
40 * entries in the rb tree are ordered by the byte number of the extent
41 * and by the byte number of the parent block.
42 */
43static int comp_entry(struct btrfs_delayed_ref_node *ref,
44 u64 bytenr, u64 parent)
45{
46 if (bytenr < ref->bytenr)
47 return -1;
48 if (bytenr > ref->bytenr)
49 return 1;
50 if (parent < ref->parent)
51 return -1;
52 if (parent > ref->parent)
53 return 1;
54 return 0;
55}
56
57/*
58 * insert a new ref into the rbtree. This returns any existing refs
59 * for the same (bytenr,parent) tuple, or NULL if the new node was properly
60 * inserted.
61 */
62static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
63 u64 bytenr, u64 parent,
64 struct rb_node *node)
65{
66 struct rb_node **p = &root->rb_node;
67 struct rb_node *parent_node = NULL;
68 struct btrfs_delayed_ref_node *entry;
69 int cmp;
70
71 while (*p) {
72 parent_node = *p;
73 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node,
74 rb_node);
75
76 cmp = comp_entry(entry, bytenr, parent);
77 if (cmp < 0)
78 p = &(*p)->rb_left;
79 else if (cmp > 0)
80 p = &(*p)->rb_right;
81 else
82 return entry;
83 }
84
85 entry = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
86 rb_link_node(node, parent_node, p);
87 rb_insert_color(node, root);
88 return NULL;
89}
90
91/*
92 * find an entry based on (bytenr,parent). This returns the delayed
93 * ref if it was able to find one, or NULL if nothing was in that spot
94 */
95static struct btrfs_delayed_ref_node *tree_search(struct rb_root *root,
96 u64 bytenr, u64 parent,
97 struct btrfs_delayed_ref_node **last)
98{
99 struct rb_node *n = root->rb_node;
100 struct btrfs_delayed_ref_node *entry;
101 int cmp;
102
103 while (n) {
104 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
105 WARN_ON(!entry->in_tree);
106 if (last)
107 *last = entry;
108
109 cmp = comp_entry(entry, bytenr, parent);
110 if (cmp < 0)
111 n = n->rb_left;
112 else if (cmp > 0)
113 n = n->rb_right;
114 else
115 return entry;
116 }
117 return NULL;
118}
119
120int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
121 struct btrfs_delayed_ref_head *head)
122{
123 struct btrfs_delayed_ref_root *delayed_refs;
124
125 delayed_refs = &trans->transaction->delayed_refs;
126 assert_spin_locked(&delayed_refs->lock);
127 if (mutex_trylock(&head->mutex))
128 return 0;
129
130 atomic_inc(&head->node.refs);
131 spin_unlock(&delayed_refs->lock);
132
133 mutex_lock(&head->mutex);
134 spin_lock(&delayed_refs->lock);
135 if (!head->node.in_tree) {
136 mutex_unlock(&head->mutex);
137 btrfs_put_delayed_ref(&head->node);
138 return -EAGAIN;
139 }
140 btrfs_put_delayed_ref(&head->node);
141 return 0;
142}
143
144int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
145 struct list_head *cluster, u64 start)
146{
147 int count = 0;
148 struct btrfs_delayed_ref_root *delayed_refs;
149 struct rb_node *node;
150 struct btrfs_delayed_ref_node *ref;
151 struct btrfs_delayed_ref_head *head;
152
153 delayed_refs = &trans->transaction->delayed_refs;
154 if (start == 0) {
155 node = rb_first(&delayed_refs->root);
156 } else {
157 ref = NULL;
158 tree_search(&delayed_refs->root, start, (u64)-1, &ref);
159 if (ref) {
160 struct btrfs_delayed_ref_node *tmp;
161
162 node = rb_prev(&ref->rb_node);
163 while (node) {
164 tmp = rb_entry(node,
165 struct btrfs_delayed_ref_node,
166 rb_node);
167 if (tmp->bytenr < start)
168 break;
169 ref = tmp;
170 node = rb_prev(&ref->rb_node);
171 }
172 node = &ref->rb_node;
173 } else
174 node = rb_first(&delayed_refs->root);
175 }
176again:
177 while (node && count < 32) {
178 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
179 if (btrfs_delayed_ref_is_head(ref)) {
180 head = btrfs_delayed_node_to_head(ref);
181 if (list_empty(&head->cluster)) {
182 list_add_tail(&head->cluster, cluster);
183 delayed_refs->run_delayed_start =
184 head->node.bytenr;
185 count++;
186
187 WARN_ON(delayed_refs->num_heads_ready == 0);
188 delayed_refs->num_heads_ready--;
189 } else if (count) {
190 /* the goal of the clustering is to find extents
191 * that are likely to end up in the same extent
192 * leaf on disk. So, we don't want them spread
193 * all over the tree. Stop now if we've hit
194 * a head that was already in use
195 */
196 break;
197 }
198 }
199 node = rb_next(node);
200 }
201 if (count) {
202 return 0;
203 } else if (start) {
204 /*
205 * we've gone to the end of the rbtree without finding any
206 * clusters. start from the beginning and try again
207 */
208 start = 0;
209 node = rb_first(&delayed_refs->root);
210 goto again;
211 }
212 return 1;
213}
214
215/*
216 * This checks to see if there are any delayed refs in the
217 * btree for a given bytenr. It returns one if it finds any
218 * and zero otherwise.
219 *
220 * If it only finds a head node, it returns 0.
221 *
222 * The idea is to use this when deciding if you can safely delete an
223 * extent from the extent allocation tree. There may be a pending
224 * ref in the rbtree that adds or removes references, so as long as this
225 * returns one you need to leave the BTRFS_EXTENT_ITEM in the extent
226 * allocation tree.
227 */
228int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr)
229{
230 struct btrfs_delayed_ref_node *ref;
231 struct btrfs_delayed_ref_root *delayed_refs;
232 struct rb_node *prev_node;
233 int ret = 0;
234
235 delayed_refs = &trans->transaction->delayed_refs;
236 spin_lock(&delayed_refs->lock);
237
238 ref = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
239 if (ref) {
240 prev_node = rb_prev(&ref->rb_node);
241 if (!prev_node)
242 goto out;
243 ref = rb_entry(prev_node, struct btrfs_delayed_ref_node,
244 rb_node);
245 if (ref->bytenr == bytenr)
246 ret = 1;
247 }
248out:
249 spin_unlock(&delayed_refs->lock);
250 return ret;
251}
252
253/*
254 * helper function to lookup reference count
255 *
256 * the head node for delayed ref is used to store the sum of all the
257 * reference count modifications queued up in the rbtree. This way you
258 * can check to see what the reference count would be if all of the
259 * delayed refs are processed.
260 */
261int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
262 struct btrfs_root *root, u64 bytenr,
263 u64 num_bytes, u32 *refs)
264{
265 struct btrfs_delayed_ref_node *ref;
266 struct btrfs_delayed_ref_head *head;
267 struct btrfs_delayed_ref_root *delayed_refs;
268 struct btrfs_path *path;
269 struct extent_buffer *leaf;
270 struct btrfs_extent_item *ei;
271 struct btrfs_key key;
272 u32 num_refs;
273 int ret;
274
275 path = btrfs_alloc_path();
276 if (!path)
277 return -ENOMEM;
278
279 key.objectid = bytenr;
280 key.type = BTRFS_EXTENT_ITEM_KEY;
281 key.offset = num_bytes;
282 delayed_refs = &trans->transaction->delayed_refs;
283again:
284 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
285 &key, path, 0, 0);
286 if (ret < 0)
287 goto out;
288
289 if (ret == 0) {
290 leaf = path->nodes[0];
291 ei = btrfs_item_ptr(leaf, path->slots[0],
292 struct btrfs_extent_item);
293 num_refs = btrfs_extent_refs(leaf, ei);
294 } else {
295 num_refs = 0;
296 ret = 0;
297 }
298
299 spin_lock(&delayed_refs->lock);
300 ref = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
301 if (ref) {
302 head = btrfs_delayed_node_to_head(ref);
303 if (mutex_trylock(&head->mutex)) {
304 num_refs += ref->ref_mod;
305 mutex_unlock(&head->mutex);
306 *refs = num_refs;
307 goto out;
308 }
309
310 atomic_inc(&ref->refs);
311 spin_unlock(&delayed_refs->lock);
312
313 btrfs_release_path(root->fs_info->extent_root, path);
314
315 mutex_lock(&head->mutex);
316 mutex_unlock(&head->mutex);
317 btrfs_put_delayed_ref(ref);
318 goto again;
319 } else {
320 *refs = num_refs;
321 }
322out:
323 spin_unlock(&delayed_refs->lock);
324 btrfs_free_path(path);
325 return ret;
326}
327
328/*
329 * helper function to update an extent delayed ref in the
330 * rbtree. existing and update must both have the same
331 * bytenr and parent
332 *
333 * This may free existing if the update cancels out whatever
334 * operation it was doing.
335 */
336static noinline void
337update_existing_ref(struct btrfs_trans_handle *trans,
338 struct btrfs_delayed_ref_root *delayed_refs,
339 struct btrfs_delayed_ref_node *existing,
340 struct btrfs_delayed_ref_node *update)
341{
342 struct btrfs_delayed_ref *existing_ref;
343 struct btrfs_delayed_ref *ref;
344
345 existing_ref = btrfs_delayed_node_to_ref(existing);
346 ref = btrfs_delayed_node_to_ref(update);
347
348 if (ref->pin)
349 existing_ref->pin = 1;
350
351 if (ref->action != existing_ref->action) {
352 /*
353 * this is effectively undoing either an add or a
354 * drop. We decrement the ref_mod, and if it goes
355 * down to zero we just delete the entry without
356 * every changing the extent allocation tree.
357 */
358 existing->ref_mod--;
359 if (existing->ref_mod == 0) {
360 rb_erase(&existing->rb_node,
361 &delayed_refs->root);
362 existing->in_tree = 0;
363 btrfs_put_delayed_ref(existing);
364 delayed_refs->num_entries--;
365 if (trans->delayed_ref_updates)
366 trans->delayed_ref_updates--;
367 }
368 } else {
369 if (existing_ref->action == BTRFS_ADD_DELAYED_REF) {
370 /* if we're adding refs, make sure all the
371 * details match up. The extent could
372 * have been totally freed and reallocated
373 * by a different owner before the delayed
374 * ref entries were removed.
375 */
376 existing_ref->owner_objectid = ref->owner_objectid;
377 existing_ref->generation = ref->generation;
378 existing_ref->root = ref->root;
379 existing->num_bytes = update->num_bytes;
380 }
381 /*
382 * the action on the existing ref matches
383 * the action on the ref we're trying to add.
384 * Bump the ref_mod by one so the backref that
385 * is eventually added/removed has the correct
386 * reference count
387 */
388 existing->ref_mod += update->ref_mod;
389 }
390}
391
392/*
393 * helper function to update the accounting in the head ref
394 * existing and update must have the same bytenr
395 */
396static noinline void
397update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
398 struct btrfs_delayed_ref_node *update)
399{
400 struct btrfs_delayed_ref_head *existing_ref;
401 struct btrfs_delayed_ref_head *ref;
402
403 existing_ref = btrfs_delayed_node_to_head(existing);
404 ref = btrfs_delayed_node_to_head(update);
405
406 if (ref->must_insert_reserved) {
407 /* if the extent was freed and then
408 * reallocated before the delayed ref
409 * entries were processed, we can end up
410 * with an existing head ref without
411 * the must_insert_reserved flag set.
412 * Set it again here
413 */
414 existing_ref->must_insert_reserved = ref->must_insert_reserved;
415
416 /*
417 * update the num_bytes so we make sure the accounting
418 * is done correctly
419 */
420 existing->num_bytes = update->num_bytes;
421
422 }
423
424 /*
425 * update the reference mod on the head to reflect this new operation
426 */
427 existing->ref_mod += update->ref_mod;
428}
429
430/*
431 * helper function to actually insert a delayed ref into the rbtree.
432 * this does all the dirty work in terms of maintaining the correct
433 * overall modification count in the head node and properly dealing
434 * with updating existing nodes as new modifications are queued.
435 */
436static noinline int __btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
437 struct btrfs_delayed_ref_node *ref,
438 u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
439 u64 ref_generation, u64 owner_objectid, int action,
440 int pin)
441{
442 struct btrfs_delayed_ref_node *existing;
443 struct btrfs_delayed_ref *full_ref;
444 struct btrfs_delayed_ref_head *head_ref = NULL;
445 struct btrfs_delayed_ref_root *delayed_refs;
446 int count_mod = 1;
447 int must_insert_reserved = 0;
448
449 /*
450 * the head node stores the sum of all the mods, so dropping a ref
451 * should drop the sum in the head node by one.
452 */
453 if (parent == (u64)-1) {
454 if (action == BTRFS_DROP_DELAYED_REF)
455 count_mod = -1;
456 else if (action == BTRFS_UPDATE_DELAYED_HEAD)
457 count_mod = 0;
458 }
459
460 /*
461 * BTRFS_ADD_DELAYED_EXTENT means that we need to update
462 * the reserved accounting when the extent is finally added, or
463 * if a later modification deletes the delayed ref without ever
464 * inserting the extent into the extent allocation tree.
465 * ref->must_insert_reserved is the flag used to record
466 * that accounting mods are required.
467 *
468 * Once we record must_insert_reserved, switch the action to
469 * BTRFS_ADD_DELAYED_REF because other special casing is not required.
470 */
471 if (action == BTRFS_ADD_DELAYED_EXTENT) {
472 must_insert_reserved = 1;
473 action = BTRFS_ADD_DELAYED_REF;
474 } else {
475 must_insert_reserved = 0;
476 }
477
478
479 delayed_refs = &trans->transaction->delayed_refs;
480
481 /* first set the basic ref node struct up */
482 atomic_set(&ref->refs, 1);
483 ref->bytenr = bytenr;
484 ref->parent = parent;
485 ref->ref_mod = count_mod;
486 ref->in_tree = 1;
487 ref->num_bytes = num_bytes;
488
489 if (btrfs_delayed_ref_is_head(ref)) {
490 head_ref = btrfs_delayed_node_to_head(ref);
491 head_ref->must_insert_reserved = must_insert_reserved;
492 INIT_LIST_HEAD(&head_ref->cluster);
493 mutex_init(&head_ref->mutex);
494 } else {
495 full_ref = btrfs_delayed_node_to_ref(ref);
496 full_ref->root = ref_root;
497 full_ref->generation = ref_generation;
498 full_ref->owner_objectid = owner_objectid;
499 full_ref->pin = pin;
500 full_ref->action = action;
501 }
502
503 existing = tree_insert(&delayed_refs->root, bytenr,
504 parent, &ref->rb_node);
505
506 if (existing) {
507 if (btrfs_delayed_ref_is_head(ref))
508 update_existing_head_ref(existing, ref);
509 else
510 update_existing_ref(trans, delayed_refs, existing, ref);
511
512 /*
513 * we've updated the existing ref, free the newly
514 * allocated ref
515 */
516 kfree(ref);
517 } else {
518 if (btrfs_delayed_ref_is_head(ref)) {
519 delayed_refs->num_heads++;
520 delayed_refs->num_heads_ready++;
521 }
522 delayed_refs->num_entries++;
523 trans->delayed_ref_updates++;
524 }
525 return 0;
526}
527
528/*
529 * add a delayed ref to the tree. This does all of the accounting required
530 * to make sure the delayed ref is eventually processed before this
531 * transaction commits.
532 */
533int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
534 u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
535 u64 ref_generation, u64 owner_objectid, int action,
536 int pin)
537{
538 struct btrfs_delayed_ref *ref;
539 struct btrfs_delayed_ref_head *head_ref;
540 struct btrfs_delayed_ref_root *delayed_refs;
541 int ret;
542
543 ref = kmalloc(sizeof(*ref), GFP_NOFS);
544 if (!ref)
545 return -ENOMEM;
546
547 /*
548 * the parent = 0 case comes from cases where we don't actually
549 * know the parent yet. It will get updated later via a add/drop
550 * pair.
551 */
552 if (parent == 0)
553 parent = bytenr;
554
555 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
556 if (!head_ref) {
557 kfree(ref);
558 return -ENOMEM;
559 }
560 delayed_refs = &trans->transaction->delayed_refs;
561 spin_lock(&delayed_refs->lock);
562
563 /*
564 * insert both the head node and the new ref without dropping
565 * the spin lock
566 */
567 ret = __btrfs_add_delayed_ref(trans, &head_ref->node, bytenr, num_bytes,
568 (u64)-1, 0, 0, 0, action, pin);
569 BUG_ON(ret);
570
571 ret = __btrfs_add_delayed_ref(trans, &ref->node, bytenr, num_bytes,
572 parent, ref_root, ref_generation,
573 owner_objectid, action, pin);
574 BUG_ON(ret);
575 spin_unlock(&delayed_refs->lock);
576 return 0;
577}
578
579/*
580 * this does a simple search for the head node for a given extent.
581 * It must be called with the delayed ref spinlock held, and it returns
582 * the head node if any where found, or NULL if not.
583 */
584struct btrfs_delayed_ref_head *
585btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
586{
587 struct btrfs_delayed_ref_node *ref;
588 struct btrfs_delayed_ref_root *delayed_refs;
589
590 delayed_refs = &trans->transaction->delayed_refs;
591 ref = tree_search(&delayed_refs->root, bytenr, (u64)-1, NULL);
592 if (ref)
593 return btrfs_delayed_node_to_head(ref);
594 return NULL;
595}
596
597/*
598 * add a delayed ref to the tree. This does all of the accounting required
599 * to make sure the delayed ref is eventually processed before this
600 * transaction commits.
601 *
602 * The main point of this call is to add and remove a backreference in a single
603 * shot, taking the lock only once, and only searching for the head node once.
604 *
605 * It is the same as doing a ref add and delete in two separate calls.
606 */
607int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
608 u64 bytenr, u64 num_bytes, u64 orig_parent,
609 u64 parent, u64 orig_ref_root, u64 ref_root,
610 u64 orig_ref_generation, u64 ref_generation,
611 u64 owner_objectid, int pin)
612{
613 struct btrfs_delayed_ref *ref;
614 struct btrfs_delayed_ref *old_ref;
615 struct btrfs_delayed_ref_head *head_ref;
616 struct btrfs_delayed_ref_root *delayed_refs;
617 int ret;
618
619 ref = kmalloc(sizeof(*ref), GFP_NOFS);
620 if (!ref)
621 return -ENOMEM;
622
623 old_ref = kmalloc(sizeof(*old_ref), GFP_NOFS);
624 if (!old_ref) {
625 kfree(ref);
626 return -ENOMEM;
627 }
628
629 /*
630 * the parent = 0 case comes from cases where we don't actually
631 * know the parent yet. It will get updated later via a add/drop
632 * pair.
633 */
634 if (parent == 0)
635 parent = bytenr;
636 if (orig_parent == 0)
637 orig_parent = bytenr;
638
639 head_ref = kmalloc(sizeof(*head_ref), GFP_NOFS);
640 if (!head_ref) {
641 kfree(ref);
642 kfree(old_ref);
643 return -ENOMEM;
644 }
645 delayed_refs = &trans->transaction->delayed_refs;
646 spin_lock(&delayed_refs->lock);
647
648 /*
649 * insert both the head node and the new ref without dropping
650 * the spin lock
651 */
652 ret = __btrfs_add_delayed_ref(trans, &head_ref->node, bytenr, num_bytes,
653 (u64)-1, 0, 0, 0,
654 BTRFS_UPDATE_DELAYED_HEAD, 0);
655 BUG_ON(ret);
656
657 ret = __btrfs_add_delayed_ref(trans, &ref->node, bytenr, num_bytes,
658 parent, ref_root, ref_generation,
659 owner_objectid, BTRFS_ADD_DELAYED_REF, 0);
660 BUG_ON(ret);
661
662 ret = __btrfs_add_delayed_ref(trans, &old_ref->node, bytenr, num_bytes,
663 orig_parent, orig_ref_root,
664 orig_ref_generation, owner_objectid,
665 BTRFS_DROP_DELAYED_REF, pin);
666 BUG_ON(ret);
667 spin_unlock(&delayed_refs->lock);
668 return 0;
669}
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
new file mode 100644
index 000000000000..3bec2ff0b15c
--- /dev/null
+++ b/fs/btrfs/delayed-ref.h
@@ -0,0 +1,193 @@
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#ifndef __DELAYED_REF__
19#define __DELAYED_REF__
20
21/* these are the possible values of struct btrfs_delayed_ref->action */
22#define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */
23#define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */
24#define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */
25#define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */
26
27struct btrfs_delayed_ref_node {
28 struct rb_node rb_node;
29
30 /* the starting bytenr of the extent */
31 u64 bytenr;
32
33 /* the parent our backref will point to */
34 u64 parent;
35
36 /* the size of the extent */
37 u64 num_bytes;
38
39 /* ref count on this data structure */
40 atomic_t refs;
41
42 /*
43 * how many refs is this entry adding or deleting. For
44 * head refs, this may be a negative number because it is keeping
45 * track of the total mods done to the reference count.
46 * For individual refs, this will always be a positive number
47 *
48 * It may be more than one, since it is possible for a single
49 * parent to have more than one ref on an extent
50 */
51 int ref_mod;
52
53 /* is this node still in the rbtree? */
54 unsigned int in_tree:1;
55};
56
57/*
58 * the head refs are used to hold a lock on a given extent, which allows us
59 * to make sure that only one process is running the delayed refs
60 * at a time for a single extent. They also store the sum of all the
61 * reference count modifications we've queued up.
62 */
63struct btrfs_delayed_ref_head {
64 struct btrfs_delayed_ref_node node;
65
66 /*
67 * the mutex is held while running the refs, and it is also
68 * held when checking the sum of reference modifications.
69 */
70 struct mutex mutex;
71
72 struct list_head cluster;
73
74 /*
75 * when a new extent is allocated, it is just reserved in memory
76 * The actual extent isn't inserted into the extent allocation tree
77 * until the delayed ref is processed. must_insert_reserved is
78 * used to flag a delayed ref so the accounting can be updated
79 * when a full insert is done.
80 *
81 * It is possible the extent will be freed before it is ever
82 * inserted into the extent allocation tree. In this case
83 * we need to update the in ram accounting to properly reflect
84 * the free has happened.
85 */
86 unsigned int must_insert_reserved:1;
87};
88
89struct btrfs_delayed_ref {
90 struct btrfs_delayed_ref_node node;
91
92 /* the root objectid our ref will point to */
93 u64 root;
94
95 /* the generation for the backref */
96 u64 generation;
97
98 /* owner_objectid of the backref */
99 u64 owner_objectid;
100
101 /* operation done by this entry in the rbtree */
102 u8 action;
103
104 /* if pin == 1, when the extent is freed it will be pinned until
105 * transaction commit
106 */
107 unsigned int pin:1;
108};
109
110struct btrfs_delayed_ref_root {
111 struct rb_root root;
112
113 /* this spin lock protects the rbtree and the entries inside */
114 spinlock_t lock;
115
116 /* how many delayed ref updates we've queued, used by the
117 * throttling code
118 */
119 unsigned long num_entries;
120
121 /* total number of head nodes in tree */
122 unsigned long num_heads;
123
124 /* total number of head nodes ready for processing */
125 unsigned long num_heads_ready;
126
127 /*
128 * set when the tree is flushing before a transaction commit,
129 * used by the throttling code to decide if new updates need
130 * to be run right away
131 */
132 int flushing;
133
134 u64 run_delayed_start;
135};
136
137static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
138{
139 WARN_ON(atomic_read(&ref->refs) == 0);
140 if (atomic_dec_and_test(&ref->refs)) {
141 WARN_ON(ref->in_tree);
142 kfree(ref);
143 }
144}
145
146int btrfs_add_delayed_ref(struct btrfs_trans_handle *trans,
147 u64 bytenr, u64 num_bytes, u64 parent, u64 ref_root,
148 u64 ref_generation, u64 owner_objectid, int action,
149 int pin);
150
151struct btrfs_delayed_ref_head *
152btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr);
153int btrfs_delayed_ref_pending(struct btrfs_trans_handle *trans, u64 bytenr);
154int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
155 struct btrfs_root *root, u64 bytenr,
156 u64 num_bytes, u32 *refs);
157int btrfs_update_delayed_ref(struct btrfs_trans_handle *trans,
158 u64 bytenr, u64 num_bytes, u64 orig_parent,
159 u64 parent, u64 orig_ref_root, u64 ref_root,
160 u64 orig_ref_generation, u64 ref_generation,
161 u64 owner_objectid, int pin);
162int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
163 struct btrfs_delayed_ref_head *head);
164int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
165 struct list_head *cluster, u64 search_start);
166/*
167 * a node might live in a head or a regular ref, this lets you
168 * test for the proper type to use.
169 */
170static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node)
171{
172 return node->parent == (u64)-1;
173}
174
175/*
176 * helper functions to cast a node into its container
177 */
178static inline struct btrfs_delayed_ref *
179btrfs_delayed_node_to_ref(struct btrfs_delayed_ref_node *node)
180{
181 WARN_ON(btrfs_delayed_ref_is_head(node));
182 return container_of(node, struct btrfs_delayed_ref, node);
183
184}
185
186static inline struct btrfs_delayed_ref_head *
187btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node)
188{
189 WARN_ON(!btrfs_delayed_ref_is_head(node));
190 return container_of(node, struct btrfs_delayed_ref_head, node);
191
192}
193#endif
diff --git a/fs/btrfs/dir-item.c b/fs/btrfs/dir-item.c
index 926a0b287a7d..1d70236ba00c 100644
--- a/fs/btrfs/dir-item.c
+++ b/fs/btrfs/dir-item.c
@@ -145,7 +145,10 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
145 key.objectid = dir; 145 key.objectid = dir;
146 btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY); 146 btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY);
147 key.offset = btrfs_name_hash(name, name_len); 147 key.offset = btrfs_name_hash(name, name_len);
148
148 path = btrfs_alloc_path(); 149 path = btrfs_alloc_path();
150 path->leave_spinning = 1;
151
149 data_size = sizeof(*dir_item) + name_len; 152 data_size = sizeof(*dir_item) + name_len;
150 dir_item = insert_with_overflow(trans, root, path, &key, data_size, 153 dir_item = insert_with_overflow(trans, root, path, &key, data_size,
151 name, name_len); 154 name, name_len);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 6ec80c0fc869..92d73929d381 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -668,14 +668,31 @@ static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
668static int btree_writepage(struct page *page, struct writeback_control *wbc) 668static int btree_writepage(struct page *page, struct writeback_control *wbc)
669{ 669{
670 struct extent_io_tree *tree; 670 struct extent_io_tree *tree;
671 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
672 struct extent_buffer *eb;
673 int was_dirty;
674
671 tree = &BTRFS_I(page->mapping->host)->io_tree; 675 tree = &BTRFS_I(page->mapping->host)->io_tree;
676 if (!(current->flags & PF_MEMALLOC)) {
677 return extent_write_full_page(tree, page,
678 btree_get_extent, wbc);
679 }
672 680
673 if (current->flags & PF_MEMALLOC) { 681 redirty_page_for_writepage(wbc, page);
674 redirty_page_for_writepage(wbc, page); 682 eb = btrfs_find_tree_block(root, page_offset(page),
675 unlock_page(page); 683 PAGE_CACHE_SIZE);
676 return 0; 684 WARN_ON(!eb);
685
686 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
687 if (!was_dirty) {
688 spin_lock(&root->fs_info->delalloc_lock);
689 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
690 spin_unlock(&root->fs_info->delalloc_lock);
677 } 691 }
678 return extent_write_full_page(tree, page, btree_get_extent, wbc); 692 free_extent_buffer(eb);
693
694 unlock_page(page);
695 return 0;
679} 696}
680 697
681static int btree_writepages(struct address_space *mapping, 698static int btree_writepages(struct address_space *mapping,
@@ -684,15 +701,15 @@ static int btree_writepages(struct address_space *mapping,
684 struct extent_io_tree *tree; 701 struct extent_io_tree *tree;
685 tree = &BTRFS_I(mapping->host)->io_tree; 702 tree = &BTRFS_I(mapping->host)->io_tree;
686 if (wbc->sync_mode == WB_SYNC_NONE) { 703 if (wbc->sync_mode == WB_SYNC_NONE) {
704 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
687 u64 num_dirty; 705 u64 num_dirty;
688 u64 start = 0;
689 unsigned long thresh = 32 * 1024 * 1024; 706 unsigned long thresh = 32 * 1024 * 1024;
690 707
691 if (wbc->for_kupdate) 708 if (wbc->for_kupdate)
692 return 0; 709 return 0;
693 710
694 num_dirty = count_range_bits(tree, &start, (u64)-1, 711 /* this is a bit racy, but that's ok */
695 thresh, EXTENT_DIRTY); 712 num_dirty = root->fs_info->dirty_metadata_bytes;
696 if (num_dirty < thresh) 713 if (num_dirty < thresh)
697 return 0; 714 return 0;
698 } 715 }
@@ -859,9 +876,17 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
859 root->fs_info->running_transaction->transid) { 876 root->fs_info->running_transaction->transid) {
860 btrfs_assert_tree_locked(buf); 877 btrfs_assert_tree_locked(buf);
861 878
862 /* ugh, clear_extent_buffer_dirty can be expensive */ 879 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
863 btrfs_set_lock_blocking(buf); 880 spin_lock(&root->fs_info->delalloc_lock);
881 if (root->fs_info->dirty_metadata_bytes >= buf->len)
882 root->fs_info->dirty_metadata_bytes -= buf->len;
883 else
884 WARN_ON(1);
885 spin_unlock(&root->fs_info->delalloc_lock);
886 }
864 887
888 /* ugh, clear_extent_buffer_dirty needs to lock the page */
889 btrfs_set_lock_blocking(buf);
865 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, 890 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
866 buf); 891 buf);
867 } 892 }
@@ -1471,12 +1496,6 @@ static int transaction_kthread(void *arg)
1471 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE); 1496 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472 mutex_lock(&root->fs_info->transaction_kthread_mutex); 1497 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1473 1498
1474 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1475 printk(KERN_INFO "btrfs: total reference cache "
1476 "size %llu\n",
1477 root->fs_info->total_ref_cache_size);
1478 }
1479
1480 mutex_lock(&root->fs_info->trans_mutex); 1499 mutex_lock(&root->fs_info->trans_mutex);
1481 cur = root->fs_info->running_transaction; 1500 cur = root->fs_info->running_transaction;
1482 if (!cur) { 1501 if (!cur) {
@@ -1493,6 +1512,7 @@ static int transaction_kthread(void *arg)
1493 mutex_unlock(&root->fs_info->trans_mutex); 1512 mutex_unlock(&root->fs_info->trans_mutex);
1494 trans = btrfs_start_transaction(root, 1); 1513 trans = btrfs_start_transaction(root, 1);
1495 ret = btrfs_commit_transaction(trans, root); 1514 ret = btrfs_commit_transaction(trans, root);
1515
1496sleep: 1516sleep:
1497 wake_up_process(root->fs_info->cleaner_kthread); 1517 wake_up_process(root->fs_info->cleaner_kthread);
1498 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1518 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
@@ -1552,6 +1572,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1552 INIT_LIST_HEAD(&fs_info->dead_roots); 1572 INIT_LIST_HEAD(&fs_info->dead_roots);
1553 INIT_LIST_HEAD(&fs_info->hashers); 1573 INIT_LIST_HEAD(&fs_info->hashers);
1554 INIT_LIST_HEAD(&fs_info->delalloc_inodes); 1574 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1575 INIT_LIST_HEAD(&fs_info->ordered_operations);
1555 spin_lock_init(&fs_info->delalloc_lock); 1576 spin_lock_init(&fs_info->delalloc_lock);
1556 spin_lock_init(&fs_info->new_trans_lock); 1577 spin_lock_init(&fs_info->new_trans_lock);
1557 spin_lock_init(&fs_info->ref_cache_lock); 1578 spin_lock_init(&fs_info->ref_cache_lock);
@@ -1611,10 +1632,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1611 1632
1612 extent_io_tree_init(&fs_info->pinned_extents, 1633 extent_io_tree_init(&fs_info->pinned_extents,
1613 fs_info->btree_inode->i_mapping, GFP_NOFS); 1634 fs_info->btree_inode->i_mapping, GFP_NOFS);
1614 extent_io_tree_init(&fs_info->pending_del,
1615 fs_info->btree_inode->i_mapping, GFP_NOFS);
1616 extent_io_tree_init(&fs_info->extent_ins,
1617 fs_info->btree_inode->i_mapping, GFP_NOFS);
1618 fs_info->do_barriers = 1; 1635 fs_info->do_barriers = 1;
1619 1636
1620 INIT_LIST_HEAD(&fs_info->dead_reloc_roots); 1637 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
@@ -1627,9 +1644,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
1627 insert_inode_hash(fs_info->btree_inode); 1644 insert_inode_hash(fs_info->btree_inode);
1628 1645
1629 mutex_init(&fs_info->trans_mutex); 1646 mutex_init(&fs_info->trans_mutex);
1647 mutex_init(&fs_info->ordered_operations_mutex);
1630 mutex_init(&fs_info->tree_log_mutex); 1648 mutex_init(&fs_info->tree_log_mutex);
1631 mutex_init(&fs_info->drop_mutex); 1649 mutex_init(&fs_info->drop_mutex);
1632 mutex_init(&fs_info->extent_ins_mutex);
1633 mutex_init(&fs_info->pinned_mutex); 1650 mutex_init(&fs_info->pinned_mutex);
1634 mutex_init(&fs_info->chunk_mutex); 1651 mutex_init(&fs_info->chunk_mutex);
1635 mutex_init(&fs_info->transaction_kthread_mutex); 1652 mutex_init(&fs_info->transaction_kthread_mutex);
@@ -2358,8 +2375,7 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2358 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 2375 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2359 u64 transid = btrfs_header_generation(buf); 2376 u64 transid = btrfs_header_generation(buf);
2360 struct inode *btree_inode = root->fs_info->btree_inode; 2377 struct inode *btree_inode = root->fs_info->btree_inode;
2361 2378 int was_dirty;
2362 btrfs_set_lock_blocking(buf);
2363 2379
2364 btrfs_assert_tree_locked(buf); 2380 btrfs_assert_tree_locked(buf);
2365 if (transid != root->fs_info->generation) { 2381 if (transid != root->fs_info->generation) {
@@ -2370,7 +2386,13 @@ void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2370 (unsigned long long)root->fs_info->generation); 2386 (unsigned long long)root->fs_info->generation);
2371 WARN_ON(1); 2387 WARN_ON(1);
2372 } 2388 }
2373 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf); 2389 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2390 buf);
2391 if (!was_dirty) {
2392 spin_lock(&root->fs_info->delalloc_lock);
2393 root->fs_info->dirty_metadata_bytes += buf->len;
2394 spin_unlock(&root->fs_info->delalloc_lock);
2395 }
2374} 2396}
2375 2397
2376void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr) 2398void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
@@ -2410,6 +2432,7 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2410int btree_lock_page_hook(struct page *page) 2432int btree_lock_page_hook(struct page *page)
2411{ 2433{
2412 struct inode *inode = page->mapping->host; 2434 struct inode *inode = page->mapping->host;
2435 struct btrfs_root *root = BTRFS_I(inode)->root;
2413 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 2436 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2414 struct extent_buffer *eb; 2437 struct extent_buffer *eb;
2415 unsigned long len; 2438 unsigned long len;
@@ -2425,6 +2448,16 @@ int btree_lock_page_hook(struct page *page)
2425 2448
2426 btrfs_tree_lock(eb); 2449 btrfs_tree_lock(eb);
2427 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); 2450 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2451
2452 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2453 spin_lock(&root->fs_info->delalloc_lock);
2454 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2455 root->fs_info->dirty_metadata_bytes -= eb->len;
2456 else
2457 WARN_ON(1);
2458 spin_unlock(&root->fs_info->delalloc_lock);
2459 }
2460
2428 btrfs_tree_unlock(eb); 2461 btrfs_tree_unlock(eb);
2429 free_extent_buffer(eb); 2462 free_extent_buffer(eb);
2430out: 2463out:
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 95029db227be..c958ecbc1916 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -72,6 +72,7 @@ int btrfs_insert_dev_radix(struct btrfs_root *root,
72void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr); 72void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr);
73int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root); 73int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root);
74void btrfs_mark_buffer_dirty(struct extent_buffer *buf); 74void btrfs_mark_buffer_dirty(struct extent_buffer *buf);
75void btrfs_mark_buffer_dirty_nonblocking(struct extent_buffer *buf);
75int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid); 76int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid);
76int btrfs_set_buffer_uptodate(struct extent_buffer *buf); 77int btrfs_set_buffer_uptodate(struct extent_buffer *buf);
77int wait_on_tree_block_writeback(struct btrfs_root *root, 78int wait_on_tree_block_writeback(struct btrfs_root *root,
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index fefe83ad2059..f5e7cae63d80 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -49,17 +49,23 @@ struct pending_extent_op {
49 int del; 49 int del;
50}; 50};
51 51
52static int finish_current_insert(struct btrfs_trans_handle *trans, 52static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *extent_root, int all); 53 struct btrfs_root *root, u64 parent,
54static int del_pending_extents(struct btrfs_trans_handle *trans, 54 u64 root_objectid, u64 ref_generation,
55 struct btrfs_root *extent_root, int all); 55 u64 owner, struct btrfs_key *ins,
56static int pin_down_bytes(struct btrfs_trans_handle *trans, 56 int ref_mod);
57 struct btrfs_root *root, 57static int update_reserved_extents(struct btrfs_root *root,
58 u64 bytenr, u64 num_bytes, int is_data); 58 u64 bytenr, u64 num, int reserve);
59static int update_block_group(struct btrfs_trans_handle *trans, 59static int update_block_group(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root, 60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int alloc, 61 u64 bytenr, u64 num_bytes, int alloc,
62 int mark_free); 62 int mark_free);
63static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 u64 bytenr, u64 num_bytes, u64 parent,
66 u64 root_objectid, u64 ref_generation,
67 u64 owner_objectid, int pin,
68 int ref_to_drop);
63 69
64static int do_chunk_alloc(struct btrfs_trans_handle *trans, 70static int do_chunk_alloc(struct btrfs_trans_handle *trans,
65 struct btrfs_root *extent_root, u64 alloc_bytes, 71 struct btrfs_root *extent_root, u64 alloc_bytes,
@@ -554,262 +560,13 @@ out:
554 return ret; 560 return ret;
555} 561}
556 562
557/*
558 * updates all the backrefs that are pending on update_list for the
559 * extent_root
560 */
561static noinline int update_backrefs(struct btrfs_trans_handle *trans,
562 struct btrfs_root *extent_root,
563 struct btrfs_path *path,
564 struct list_head *update_list)
565{
566 struct btrfs_key key;
567 struct btrfs_extent_ref *ref;
568 struct btrfs_fs_info *info = extent_root->fs_info;
569 struct pending_extent_op *op;
570 struct extent_buffer *leaf;
571 int ret = 0;
572 struct list_head *cur = update_list->next;
573 u64 ref_objectid;
574 u64 ref_root = extent_root->root_key.objectid;
575
576 op = list_entry(cur, struct pending_extent_op, list);
577
578search:
579 key.objectid = op->bytenr;
580 key.type = BTRFS_EXTENT_REF_KEY;
581 key.offset = op->orig_parent;
582
583 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
584 BUG_ON(ret);
585
586 leaf = path->nodes[0];
587
588loop:
589 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
590
591 ref_objectid = btrfs_ref_objectid(leaf, ref);
592
593 if (btrfs_ref_root(leaf, ref) != ref_root ||
594 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
595 (ref_objectid != op->level &&
596 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
597 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
598 "root %llu, owner %u\n",
599 (unsigned long long)op->bytenr,
600 (unsigned long long)op->orig_parent,
601 (unsigned long long)ref_root, op->level);
602 btrfs_print_leaf(extent_root, leaf);
603 BUG();
604 }
605
606 key.objectid = op->bytenr;
607 key.offset = op->parent;
608 key.type = BTRFS_EXTENT_REF_KEY;
609 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
610 BUG_ON(ret);
611 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
612 btrfs_set_ref_generation(leaf, ref, op->generation);
613
614 cur = cur->next;
615
616 list_del_init(&op->list);
617 unlock_extent(&info->extent_ins, op->bytenr,
618 op->bytenr + op->num_bytes - 1, GFP_NOFS);
619 kfree(op);
620
621 if (cur == update_list) {
622 btrfs_mark_buffer_dirty(path->nodes[0]);
623 btrfs_release_path(extent_root, path);
624 goto out;
625 }
626
627 op = list_entry(cur, struct pending_extent_op, list);
628
629 path->slots[0]++;
630 while (path->slots[0] < btrfs_header_nritems(leaf)) {
631 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
632 if (key.objectid == op->bytenr &&
633 key.type == BTRFS_EXTENT_REF_KEY)
634 goto loop;
635 path->slots[0]++;
636 }
637
638 btrfs_mark_buffer_dirty(path->nodes[0]);
639 btrfs_release_path(extent_root, path);
640 goto search;
641
642out:
643 return 0;
644}
645
646static noinline int insert_extents(struct btrfs_trans_handle *trans,
647 struct btrfs_root *extent_root,
648 struct btrfs_path *path,
649 struct list_head *insert_list, int nr)
650{
651 struct btrfs_key *keys;
652 u32 *data_size;
653 struct pending_extent_op *op;
654 struct extent_buffer *leaf;
655 struct list_head *cur = insert_list->next;
656 struct btrfs_fs_info *info = extent_root->fs_info;
657 u64 ref_root = extent_root->root_key.objectid;
658 int i = 0, last = 0, ret;
659 int total = nr * 2;
660
661 if (!nr)
662 return 0;
663
664 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
665 if (!keys)
666 return -ENOMEM;
667
668 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
669 if (!data_size) {
670 kfree(keys);
671 return -ENOMEM;
672 }
673
674 list_for_each_entry(op, insert_list, list) {
675 keys[i].objectid = op->bytenr;
676 keys[i].offset = op->num_bytes;
677 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
678 data_size[i] = sizeof(struct btrfs_extent_item);
679 i++;
680
681 keys[i].objectid = op->bytenr;
682 keys[i].offset = op->parent;
683 keys[i].type = BTRFS_EXTENT_REF_KEY;
684 data_size[i] = sizeof(struct btrfs_extent_ref);
685 i++;
686 }
687
688 op = list_entry(cur, struct pending_extent_op, list);
689 i = 0;
690 while (i < total) {
691 int c;
692 ret = btrfs_insert_some_items(trans, extent_root, path,
693 keys+i, data_size+i, total-i);
694 BUG_ON(ret < 0);
695
696 if (last && ret > 1)
697 BUG();
698
699 leaf = path->nodes[0];
700 for (c = 0; c < ret; c++) {
701 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
702
703 /*
704 * if the first item we inserted was a backref, then
705 * the EXTENT_ITEM will be the odd c's, else it will
706 * be the even c's
707 */
708 if ((ref_first && (c % 2)) ||
709 (!ref_first && !(c % 2))) {
710 struct btrfs_extent_item *itm;
711
712 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
713 struct btrfs_extent_item);
714 btrfs_set_extent_refs(path->nodes[0], itm, 1);
715 op->del++;
716 } else {
717 struct btrfs_extent_ref *ref;
718
719 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
720 struct btrfs_extent_ref);
721 btrfs_set_ref_root(leaf, ref, ref_root);
722 btrfs_set_ref_generation(leaf, ref,
723 op->generation);
724 btrfs_set_ref_objectid(leaf, ref, op->level);
725 btrfs_set_ref_num_refs(leaf, ref, 1);
726 op->del++;
727 }
728
729 /*
730 * using del to see when its ok to free up the
731 * pending_extent_op. In the case where we insert the
732 * last item on the list in order to help do batching
733 * we need to not free the extent op until we actually
734 * insert the extent_item
735 */
736 if (op->del == 2) {
737 unlock_extent(&info->extent_ins, op->bytenr,
738 op->bytenr + op->num_bytes - 1,
739 GFP_NOFS);
740 cur = cur->next;
741 list_del_init(&op->list);
742 kfree(op);
743 if (cur != insert_list)
744 op = list_entry(cur,
745 struct pending_extent_op,
746 list);
747 }
748 }
749 btrfs_mark_buffer_dirty(leaf);
750 btrfs_release_path(extent_root, path);
751
752 /*
753 * Ok backref's and items usually go right next to eachother,
754 * but if we could only insert 1 item that means that we
755 * inserted on the end of a leaf, and we have no idea what may
756 * be on the next leaf so we just play it safe. In order to
757 * try and help this case we insert the last thing on our
758 * insert list so hopefully it will end up being the last
759 * thing on the leaf and everything else will be before it,
760 * which will let us insert a whole bunch of items at the same
761 * time.
762 */
763 if (ret == 1 && !last && (i + ret < total)) {
764 /*
765 * last: where we will pick up the next time around
766 * i: our current key to insert, will be total - 1
767 * cur: the current op we are screwing with
768 * op: duh
769 */
770 last = i + ret;
771 i = total - 1;
772 cur = insert_list->prev;
773 op = list_entry(cur, struct pending_extent_op, list);
774 } else if (last) {
775 /*
776 * ok we successfully inserted the last item on the
777 * list, lets reset everything
778 *
779 * i: our current key to insert, so where we left off
780 * last time
781 * last: done with this
782 * cur: the op we are messing with
783 * op: duh
784 * total: since we inserted the last key, we need to
785 * decrement total so we dont overflow
786 */
787 i = last;
788 last = 0;
789 total--;
790 if (i < total) {
791 cur = insert_list->next;
792 op = list_entry(cur, struct pending_extent_op,
793 list);
794 }
795 } else {
796 i += ret;
797 }
798
799 cond_resched();
800 }
801 ret = 0;
802 kfree(keys);
803 kfree(data_size);
804 return ret;
805}
806
807static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, 563static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
808 struct btrfs_root *root, 564 struct btrfs_root *root,
809 struct btrfs_path *path, 565 struct btrfs_path *path,
810 u64 bytenr, u64 parent, 566 u64 bytenr, u64 parent,
811 u64 ref_root, u64 ref_generation, 567 u64 ref_root, u64 ref_generation,
812 u64 owner_objectid) 568 u64 owner_objectid,
569 int refs_to_add)
813{ 570{
814 struct btrfs_key key; 571 struct btrfs_key key;
815 struct extent_buffer *leaf; 572 struct extent_buffer *leaf;
@@ -829,9 +586,10 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
829 btrfs_set_ref_root(leaf, ref, ref_root); 586 btrfs_set_ref_root(leaf, ref, ref_root);
830 btrfs_set_ref_generation(leaf, ref, ref_generation); 587 btrfs_set_ref_generation(leaf, ref, ref_generation);
831 btrfs_set_ref_objectid(leaf, ref, owner_objectid); 588 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
832 btrfs_set_ref_num_refs(leaf, ref, 1); 589 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
833 } else if (ret == -EEXIST) { 590 } else if (ret == -EEXIST) {
834 u64 existing_owner; 591 u64 existing_owner;
592
835 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); 593 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
836 leaf = path->nodes[0]; 594 leaf = path->nodes[0];
837 ref = btrfs_item_ptr(leaf, path->slots[0], 595 ref = btrfs_item_ptr(leaf, path->slots[0],
@@ -845,7 +603,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
845 603
846 num_refs = btrfs_ref_num_refs(leaf, ref); 604 num_refs = btrfs_ref_num_refs(leaf, ref);
847 BUG_ON(num_refs == 0); 605 BUG_ON(num_refs == 0);
848 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); 606 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
849 607
850 existing_owner = btrfs_ref_objectid(leaf, ref); 608 existing_owner = btrfs_ref_objectid(leaf, ref);
851 if (existing_owner != owner_objectid && 609 if (existing_owner != owner_objectid &&
@@ -857,6 +615,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
857 } else { 615 } else {
858 goto out; 616 goto out;
859 } 617 }
618 btrfs_unlock_up_safe(path, 1);
860 btrfs_mark_buffer_dirty(path->nodes[0]); 619 btrfs_mark_buffer_dirty(path->nodes[0]);
861out: 620out:
862 btrfs_release_path(root, path); 621 btrfs_release_path(root, path);
@@ -865,7 +624,8 @@ out:
865 624
866static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, 625static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
867 struct btrfs_root *root, 626 struct btrfs_root *root,
868 struct btrfs_path *path) 627 struct btrfs_path *path,
628 int refs_to_drop)
869{ 629{
870 struct extent_buffer *leaf; 630 struct extent_buffer *leaf;
871 struct btrfs_extent_ref *ref; 631 struct btrfs_extent_ref *ref;
@@ -875,8 +635,8 @@ static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
875 leaf = path->nodes[0]; 635 leaf = path->nodes[0];
876 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); 636 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
877 num_refs = btrfs_ref_num_refs(leaf, ref); 637 num_refs = btrfs_ref_num_refs(leaf, ref);
878 BUG_ON(num_refs == 0); 638 BUG_ON(num_refs < refs_to_drop);
879 num_refs -= 1; 639 num_refs -= refs_to_drop;
880 if (num_refs == 0) { 640 if (num_refs == 0) {
881 ret = btrfs_del_item(trans, root, path); 641 ret = btrfs_del_item(trans, root, path);
882 } else { 642 } else {
@@ -927,332 +687,28 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
927#endif 687#endif
928} 688}
929 689
930static noinline int free_extents(struct btrfs_trans_handle *trans,
931 struct btrfs_root *extent_root,
932 struct list_head *del_list)
933{
934 struct btrfs_fs_info *info = extent_root->fs_info;
935 struct btrfs_path *path;
936 struct btrfs_key key, found_key;
937 struct extent_buffer *leaf;
938 struct list_head *cur;
939 struct pending_extent_op *op;
940 struct btrfs_extent_item *ei;
941 int ret, num_to_del, extent_slot = 0, found_extent = 0;
942 u32 refs;
943 u64 bytes_freed = 0;
944
945 path = btrfs_alloc_path();
946 if (!path)
947 return -ENOMEM;
948 path->reada = 1;
949
950search:
951 /* search for the backref for the current ref we want to delete */
952 cur = del_list->next;
953 op = list_entry(cur, struct pending_extent_op, list);
954 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
955 op->orig_parent,
956 extent_root->root_key.objectid,
957 op->orig_generation, op->level, 1);
958 if (ret) {
959 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
960 "root %llu gen %llu owner %u\n",
961 (unsigned long long)op->bytenr,
962 (unsigned long long)extent_root->root_key.objectid,
963 (unsigned long long)op->orig_generation, op->level);
964 btrfs_print_leaf(extent_root, path->nodes[0]);
965 WARN_ON(1);
966 goto out;
967 }
968
969 extent_slot = path->slots[0];
970 num_to_del = 1;
971 found_extent = 0;
972
973 /*
974 * if we aren't the first item on the leaf we can move back one and see
975 * if our ref is right next to our extent item
976 */
977 if (likely(extent_slot)) {
978 extent_slot--;
979 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
980 extent_slot);
981 if (found_key.objectid == op->bytenr &&
982 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
983 found_key.offset == op->num_bytes) {
984 num_to_del++;
985 found_extent = 1;
986 }
987 }
988
989 /*
990 * if we didn't find the extent we need to delete the backref and then
991 * search for the extent item key so we can update its ref count
992 */
993 if (!found_extent) {
994 key.objectid = op->bytenr;
995 key.type = BTRFS_EXTENT_ITEM_KEY;
996 key.offset = op->num_bytes;
997
998 ret = remove_extent_backref(trans, extent_root, path);
999 BUG_ON(ret);
1000 btrfs_release_path(extent_root, path);
1001 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1002 BUG_ON(ret);
1003 extent_slot = path->slots[0];
1004 }
1005
1006 /* this is where we update the ref count for the extent */
1007 leaf = path->nodes[0];
1008 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
1009 refs = btrfs_extent_refs(leaf, ei);
1010 BUG_ON(refs == 0);
1011 refs--;
1012 btrfs_set_extent_refs(leaf, ei, refs);
1013
1014 btrfs_mark_buffer_dirty(leaf);
1015
1016 /*
1017 * This extent needs deleting. The reason cur_slot is extent_slot +
1018 * num_to_del is because extent_slot points to the slot where the extent
1019 * is, and if the backref was not right next to the extent we will be
1020 * deleting at least 1 item, and will want to start searching at the
1021 * slot directly next to extent_slot. However if we did find the
1022 * backref next to the extent item them we will be deleting at least 2
1023 * items and will want to start searching directly after the ref slot
1024 */
1025 if (!refs) {
1026 struct list_head *pos, *n, *end;
1027 int cur_slot = extent_slot+num_to_del;
1028 u64 super_used;
1029 u64 root_used;
1030
1031 path->slots[0] = extent_slot;
1032 bytes_freed = op->num_bytes;
1033
1034 mutex_lock(&info->pinned_mutex);
1035 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1036 op->num_bytes, op->level >=
1037 BTRFS_FIRST_FREE_OBJECTID);
1038 mutex_unlock(&info->pinned_mutex);
1039 BUG_ON(ret < 0);
1040 op->del = ret;
1041
1042 /*
1043 * we need to see if we can delete multiple things at once, so
1044 * start looping through the list of extents we are wanting to
1045 * delete and see if their extent/backref's are right next to
1046 * eachother and the extents only have 1 ref
1047 */
1048 for (pos = cur->next; pos != del_list; pos = pos->next) {
1049 struct pending_extent_op *tmp;
1050
1051 tmp = list_entry(pos, struct pending_extent_op, list);
1052
1053 /* we only want to delete extent+ref at this stage */
1054 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1055 break;
1056
1057 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1058 if (found_key.objectid != tmp->bytenr ||
1059 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1060 found_key.offset != tmp->num_bytes)
1061 break;
1062
1063 /* check to make sure this extent only has one ref */
1064 ei = btrfs_item_ptr(leaf, cur_slot,
1065 struct btrfs_extent_item);
1066 if (btrfs_extent_refs(leaf, ei) != 1)
1067 break;
1068
1069 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1070 if (found_key.objectid != tmp->bytenr ||
1071 found_key.type != BTRFS_EXTENT_REF_KEY ||
1072 found_key.offset != tmp->orig_parent)
1073 break;
1074
1075 /*
1076 * the ref is right next to the extent, we can set the
1077 * ref count to 0 since we will delete them both now
1078 */
1079 btrfs_set_extent_refs(leaf, ei, 0);
1080
1081 /* pin down the bytes for this extent */
1082 mutex_lock(&info->pinned_mutex);
1083 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1084 tmp->num_bytes, tmp->level >=
1085 BTRFS_FIRST_FREE_OBJECTID);
1086 mutex_unlock(&info->pinned_mutex);
1087 BUG_ON(ret < 0);
1088
1089 /*
1090 * use the del field to tell if we need to go ahead and
1091 * free up the extent when we delete the item or not.
1092 */
1093 tmp->del = ret;
1094 bytes_freed += tmp->num_bytes;
1095
1096 num_to_del += 2;
1097 cur_slot += 2;
1098 }
1099 end = pos;
1100
1101 /* update the free space counters */
1102 spin_lock(&info->delalloc_lock);
1103 super_used = btrfs_super_bytes_used(&info->super_copy);
1104 btrfs_set_super_bytes_used(&info->super_copy,
1105 super_used - bytes_freed);
1106
1107 root_used = btrfs_root_used(&extent_root->root_item);
1108 btrfs_set_root_used(&extent_root->root_item,
1109 root_used - bytes_freed);
1110 spin_unlock(&info->delalloc_lock);
1111
1112 /* delete the items */
1113 ret = btrfs_del_items(trans, extent_root, path,
1114 path->slots[0], num_to_del);
1115 BUG_ON(ret);
1116
1117 /*
1118 * loop through the extents we deleted and do the cleanup work
1119 * on them
1120 */
1121 for (pos = cur, n = pos->next; pos != end;
1122 pos = n, n = pos->next) {
1123 struct pending_extent_op *tmp;
1124 tmp = list_entry(pos, struct pending_extent_op, list);
1125
1126 /*
1127 * remember tmp->del tells us wether or not we pinned
1128 * down the extent
1129 */
1130 ret = update_block_group(trans, extent_root,
1131 tmp->bytenr, tmp->num_bytes, 0,
1132 tmp->del);
1133 BUG_ON(ret);
1134
1135 list_del_init(&tmp->list);
1136 unlock_extent(&info->extent_ins, tmp->bytenr,
1137 tmp->bytenr + tmp->num_bytes - 1,
1138 GFP_NOFS);
1139 kfree(tmp);
1140 }
1141 } else if (refs && found_extent) {
1142 /*
1143 * the ref and extent were right next to eachother, but the
1144 * extent still has a ref, so just free the backref and keep
1145 * going
1146 */
1147 ret = remove_extent_backref(trans, extent_root, path);
1148 BUG_ON(ret);
1149
1150 list_del_init(&op->list);
1151 unlock_extent(&info->extent_ins, op->bytenr,
1152 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1153 kfree(op);
1154 } else {
1155 /*
1156 * the extent has multiple refs and the backref we were looking
1157 * for was not right next to it, so just unlock and go next,
1158 * we're good to go
1159 */
1160 list_del_init(&op->list);
1161 unlock_extent(&info->extent_ins, op->bytenr,
1162 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1163 kfree(op);
1164 }
1165
1166 btrfs_release_path(extent_root, path);
1167 if (!list_empty(del_list))
1168 goto search;
1169
1170out:
1171 btrfs_free_path(path);
1172 return ret;
1173}
1174
1175static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, 690static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1176 struct btrfs_root *root, u64 bytenr, 691 struct btrfs_root *root, u64 bytenr,
692 u64 num_bytes,
1177 u64 orig_parent, u64 parent, 693 u64 orig_parent, u64 parent,
1178 u64 orig_root, u64 ref_root, 694 u64 orig_root, u64 ref_root,
1179 u64 orig_generation, u64 ref_generation, 695 u64 orig_generation, u64 ref_generation,
1180 u64 owner_objectid) 696 u64 owner_objectid)
1181{ 697{
1182 int ret; 698 int ret;
1183 struct btrfs_root *extent_root = root->fs_info->extent_root; 699 int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
1184 struct btrfs_path *path;
1185
1186 if (root == root->fs_info->extent_root) {
1187 struct pending_extent_op *extent_op;
1188 u64 num_bytes;
1189
1190 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1191 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1192 mutex_lock(&root->fs_info->extent_ins_mutex);
1193 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1194 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1195 u64 priv;
1196 ret = get_state_private(&root->fs_info->extent_ins,
1197 bytenr, &priv);
1198 BUG_ON(ret);
1199 extent_op = (struct pending_extent_op *)
1200 (unsigned long)priv;
1201 BUG_ON(extent_op->parent != orig_parent);
1202 BUG_ON(extent_op->generation != orig_generation);
1203 700
1204 extent_op->parent = parent; 701 ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
1205 extent_op->generation = ref_generation; 702 orig_parent, parent, orig_root,
1206 } else { 703 ref_root, orig_generation,
1207 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 704 ref_generation, owner_objectid, pin);
1208 BUG_ON(!extent_op);
1209
1210 extent_op->type = PENDING_BACKREF_UPDATE;
1211 extent_op->bytenr = bytenr;
1212 extent_op->num_bytes = num_bytes;
1213 extent_op->parent = parent;
1214 extent_op->orig_parent = orig_parent;
1215 extent_op->generation = ref_generation;
1216 extent_op->orig_generation = orig_generation;
1217 extent_op->level = (int)owner_objectid;
1218 INIT_LIST_HEAD(&extent_op->list);
1219 extent_op->del = 0;
1220
1221 set_extent_bits(&root->fs_info->extent_ins,
1222 bytenr, bytenr + num_bytes - 1,
1223 EXTENT_WRITEBACK, GFP_NOFS);
1224 set_state_private(&root->fs_info->extent_ins,
1225 bytenr, (unsigned long)extent_op);
1226 }
1227 mutex_unlock(&root->fs_info->extent_ins_mutex);
1228 return 0;
1229 }
1230
1231 path = btrfs_alloc_path();
1232 if (!path)
1233 return -ENOMEM;
1234 ret = lookup_extent_backref(trans, extent_root, path,
1235 bytenr, orig_parent, orig_root,
1236 orig_generation, owner_objectid, 1);
1237 if (ret)
1238 goto out;
1239 ret = remove_extent_backref(trans, extent_root, path);
1240 if (ret)
1241 goto out;
1242 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1243 parent, ref_root, ref_generation,
1244 owner_objectid);
1245 BUG_ON(ret); 705 BUG_ON(ret);
1246 finish_current_insert(trans, extent_root, 0);
1247 del_pending_extents(trans, extent_root, 0);
1248out:
1249 btrfs_free_path(path);
1250 return ret; 706 return ret;
1251} 707}
1252 708
1253int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, 709int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1254 struct btrfs_root *root, u64 bytenr, 710 struct btrfs_root *root, u64 bytenr,
1255 u64 orig_parent, u64 parent, 711 u64 num_bytes, u64 orig_parent, u64 parent,
1256 u64 ref_root, u64 ref_generation, 712 u64 ref_root, u64 ref_generation,
1257 u64 owner_objectid) 713 u64 owner_objectid)
1258{ 714{
@@ -1260,20 +716,36 @@ int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1260 if (ref_root == BTRFS_TREE_LOG_OBJECTID && 716 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1261 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) 717 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1262 return 0; 718 return 0;
1263 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, 719
1264 parent, ref_root, ref_root, 720 ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
1265 ref_generation, ref_generation, 721 orig_parent, parent, ref_root,
1266 owner_objectid); 722 ref_root, ref_generation,
723 ref_generation, owner_objectid);
1267 return ret; 724 return ret;
1268} 725}
1269
1270static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, 726static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root, u64 bytenr, 727 struct btrfs_root *root, u64 bytenr,
728 u64 num_bytes,
1272 u64 orig_parent, u64 parent, 729 u64 orig_parent, u64 parent,
1273 u64 orig_root, u64 ref_root, 730 u64 orig_root, u64 ref_root,
1274 u64 orig_generation, u64 ref_generation, 731 u64 orig_generation, u64 ref_generation,
1275 u64 owner_objectid) 732 u64 owner_objectid)
1276{ 733{
734 int ret;
735
736 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
737 ref_generation, owner_objectid,
738 BTRFS_ADD_DELAYED_REF, 0);
739 BUG_ON(ret);
740 return ret;
741}
742
743static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 num_bytes, u64 parent, u64 ref_root,
746 u64 ref_generation, u64 owner_objectid,
747 int refs_to_add)
748{
1277 struct btrfs_path *path; 749 struct btrfs_path *path;
1278 int ret; 750 int ret;
1279 struct btrfs_key key; 751 struct btrfs_key key;
@@ -1286,17 +758,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1286 return -ENOMEM; 758 return -ENOMEM;
1287 759
1288 path->reada = 1; 760 path->reada = 1;
761 path->leave_spinning = 1;
1289 key.objectid = bytenr; 762 key.objectid = bytenr;
1290 key.type = BTRFS_EXTENT_ITEM_KEY; 763 key.type = BTRFS_EXTENT_ITEM_KEY;
1291 key.offset = (u64)-1; 764 key.offset = num_bytes;
1292 765
1293 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, 766 /* first find the extent item and update its reference count */
1294 0, 1); 767 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1295 if (ret < 0) 768 path, 0, 1);
769 if (ret < 0) {
770 btrfs_set_path_blocking(path);
1296 return ret; 771 return ret;
1297 BUG_ON(ret == 0 || path->slots[0] == 0); 772 }
1298 773
1299 path->slots[0]--; 774 if (ret > 0) {
775 WARN_ON(1);
776 btrfs_free_path(path);
777 return -EIO;
778 }
1300 l = path->nodes[0]; 779 l = path->nodes[0];
1301 780
1302 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 781 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
@@ -1310,21 +789,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1310 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); 789 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1311 790
1312 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); 791 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
792
1313 refs = btrfs_extent_refs(l, item); 793 refs = btrfs_extent_refs(l, item);
1314 btrfs_set_extent_refs(l, item, refs + 1); 794 btrfs_set_extent_refs(l, item, refs + refs_to_add);
795 btrfs_unlock_up_safe(path, 1);
796
1315 btrfs_mark_buffer_dirty(path->nodes[0]); 797 btrfs_mark_buffer_dirty(path->nodes[0]);
1316 798
1317 btrfs_release_path(root->fs_info->extent_root, path); 799 btrfs_release_path(root->fs_info->extent_root, path);
1318 800
1319 path->reada = 1; 801 path->reada = 1;
802 path->leave_spinning = 1;
803
804 /* now insert the actual backref */
1320 ret = insert_extent_backref(trans, root->fs_info->extent_root, 805 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1321 path, bytenr, parent, 806 path, bytenr, parent,
1322 ref_root, ref_generation, 807 ref_root, ref_generation,
1323 owner_objectid); 808 owner_objectid, refs_to_add);
1324 BUG_ON(ret); 809 BUG_ON(ret);
1325 finish_current_insert(trans, root->fs_info->extent_root, 0);
1326 del_pending_extents(trans, root->fs_info->extent_root, 0);
1327
1328 btrfs_free_path(path); 810 btrfs_free_path(path);
1329 return 0; 811 return 0;
1330} 812}
@@ -1339,68 +821,278 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1339 if (ref_root == BTRFS_TREE_LOG_OBJECTID && 821 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1340 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) 822 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1341 return 0; 823 return 0;
1342 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, 824
825 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
1343 0, ref_root, 0, ref_generation, 826 0, ref_root, 0, ref_generation,
1344 owner_objectid); 827 owner_objectid);
1345 return ret; 828 return ret;
1346} 829}
1347 830
1348int btrfs_extent_post_op(struct btrfs_trans_handle *trans, 831static int drop_delayed_ref(struct btrfs_trans_handle *trans,
1349 struct btrfs_root *root) 832 struct btrfs_root *root,
833 struct btrfs_delayed_ref_node *node)
834{
835 int ret = 0;
836 struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
837
838 BUG_ON(node->ref_mod == 0);
839 ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
840 node->parent, ref->root, ref->generation,
841 ref->owner_objectid, ref->pin, node->ref_mod);
842
843 return ret;
844}
845
846/* helper function to actually process a single delayed ref entry */
847static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
848 struct btrfs_root *root,
849 struct btrfs_delayed_ref_node *node,
850 int insert_reserved)
1350{ 851{
1351 u64 start;
1352 u64 end;
1353 int ret; 852 int ret;
853 struct btrfs_delayed_ref *ref;
854
855 if (node->parent == (u64)-1) {
856 struct btrfs_delayed_ref_head *head;
857 /*
858 * we've hit the end of the chain and we were supposed
859 * to insert this extent into the tree. But, it got
860 * deleted before we ever needed to insert it, so all
861 * we have to do is clean up the accounting
862 */
863 if (insert_reserved) {
864 update_reserved_extents(root, node->bytenr,
865 node->num_bytes, 0);
866 }
867 head = btrfs_delayed_node_to_head(node);
868 mutex_unlock(&head->mutex);
869 return 0;
870 }
1354 871
1355 while(1) { 872 ref = btrfs_delayed_node_to_ref(node);
1356 finish_current_insert(trans, root->fs_info->extent_root, 1); 873 if (ref->action == BTRFS_ADD_DELAYED_REF) {
1357 del_pending_extents(trans, root->fs_info->extent_root, 1); 874 if (insert_reserved) {
875 struct btrfs_key ins;
1358 876
1359 /* is there more work to do? */ 877 ins.objectid = node->bytenr;
1360 ret = find_first_extent_bit(&root->fs_info->pending_del, 878 ins.offset = node->num_bytes;
1361 0, &start, &end, EXTENT_WRITEBACK); 879 ins.type = BTRFS_EXTENT_ITEM_KEY;
1362 if (!ret) 880
1363 continue; 881 /* record the full extent allocation */
1364 ret = find_first_extent_bit(&root->fs_info->extent_ins, 882 ret = __btrfs_alloc_reserved_extent(trans, root,
1365 0, &start, &end, EXTENT_WRITEBACK); 883 node->parent, ref->root,
1366 if (!ret) 884 ref->generation, ref->owner_objectid,
1367 continue; 885 &ins, node->ref_mod);
1368 break; 886 update_reserved_extents(root, node->bytenr,
887 node->num_bytes, 0);
888 } else {
889 /* just add one backref */
890 ret = add_extent_ref(trans, root, node->bytenr,
891 node->num_bytes,
892 node->parent, ref->root, ref->generation,
893 ref->owner_objectid, node->ref_mod);
894 }
895 BUG_ON(ret);
896 } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
897 WARN_ON(insert_reserved);
898 ret = drop_delayed_ref(trans, root, node);
1369 } 899 }
1370 return 0; 900 return 0;
1371} 901}
1372 902
1373int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, 903static noinline struct btrfs_delayed_ref_node *
1374 struct btrfs_root *root, u64 bytenr, 904select_delayed_ref(struct btrfs_delayed_ref_head *head)
1375 u64 num_bytes, u32 *refs)
1376{ 905{
1377 struct btrfs_path *path; 906 struct rb_node *node;
907 struct btrfs_delayed_ref_node *ref;
908 int action = BTRFS_ADD_DELAYED_REF;
909again:
910 /*
911 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
912 * this prevents ref count from going down to zero when
913 * there still are pending delayed ref.
914 */
915 node = rb_prev(&head->node.rb_node);
916 while (1) {
917 if (!node)
918 break;
919 ref = rb_entry(node, struct btrfs_delayed_ref_node,
920 rb_node);
921 if (ref->bytenr != head->node.bytenr)
922 break;
923 if (btrfs_delayed_node_to_ref(ref)->action == action)
924 return ref;
925 node = rb_prev(node);
926 }
927 if (action == BTRFS_ADD_DELAYED_REF) {
928 action = BTRFS_DROP_DELAYED_REF;
929 goto again;
930 }
931 return NULL;
932}
933
934static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
935 struct btrfs_root *root,
936 struct list_head *cluster)
937{
938 struct btrfs_delayed_ref_root *delayed_refs;
939 struct btrfs_delayed_ref_node *ref;
940 struct btrfs_delayed_ref_head *locked_ref = NULL;
1378 int ret; 941 int ret;
1379 struct btrfs_key key; 942 int count = 0;
1380 struct extent_buffer *l; 943 int must_insert_reserved = 0;
1381 struct btrfs_extent_item *item;
1382 944
1383 WARN_ON(num_bytes < root->sectorsize); 945 delayed_refs = &trans->transaction->delayed_refs;
1384 path = btrfs_alloc_path(); 946 while (1) {
1385 path->reada = 1; 947 if (!locked_ref) {
1386 key.objectid = bytenr; 948 /* pick a new head ref from the cluster list */
1387 key.offset = num_bytes; 949 if (list_empty(cluster))
1388 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); 950 break;
1389 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, 951
1390 0, 0); 952 locked_ref = list_entry(cluster->next,
1391 if (ret < 0) 953 struct btrfs_delayed_ref_head, cluster);
1392 goto out; 954
1393 if (ret != 0) { 955 /* grab the lock that says we are going to process
1394 btrfs_print_leaf(root, path->nodes[0]); 956 * all the refs for this head */
1395 printk(KERN_INFO "btrfs failed to find block number %llu\n", 957 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1396 (unsigned long long)bytenr); 958
1397 BUG(); 959 /*
960 * we may have dropped the spin lock to get the head
961 * mutex lock, and that might have given someone else
962 * time to free the head. If that's true, it has been
963 * removed from our list and we can move on.
964 */
965 if (ret == -EAGAIN) {
966 locked_ref = NULL;
967 count++;
968 continue;
969 }
970 }
971
972 /*
973 * record the must insert reserved flag before we
974 * drop the spin lock.
975 */
976 must_insert_reserved = locked_ref->must_insert_reserved;
977 locked_ref->must_insert_reserved = 0;
978
979 /*
980 * locked_ref is the head node, so we have to go one
981 * node back for any delayed ref updates
982 */
983 ref = select_delayed_ref(locked_ref);
984 if (!ref) {
985 /* All delayed refs have been processed, Go ahead
986 * and send the head node to run_one_delayed_ref,
987 * so that any accounting fixes can happen
988 */
989 ref = &locked_ref->node;
990 list_del_init(&locked_ref->cluster);
991 locked_ref = NULL;
992 }
993
994 ref->in_tree = 0;
995 rb_erase(&ref->rb_node, &delayed_refs->root);
996 delayed_refs->num_entries--;
997 spin_unlock(&delayed_refs->lock);
998
999 ret = run_one_delayed_ref(trans, root, ref,
1000 must_insert_reserved);
1001 BUG_ON(ret);
1002 btrfs_put_delayed_ref(ref);
1003
1004 count++;
1005 cond_resched();
1006 spin_lock(&delayed_refs->lock);
1007 }
1008 return count;
1009}
1010
1011/*
1012 * this starts processing the delayed reference count updates and
1013 * extent insertions we have queued up so far. count can be
1014 * 0, which means to process everything in the tree at the start
1015 * of the run (but not newly added entries), or it can be some target
1016 * number you'd like to process.
1017 */
1018int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root, unsigned long count)
1020{
1021 struct rb_node *node;
1022 struct btrfs_delayed_ref_root *delayed_refs;
1023 struct btrfs_delayed_ref_node *ref;
1024 struct list_head cluster;
1025 int ret;
1026 int run_all = count == (unsigned long)-1;
1027 int run_most = 0;
1028
1029 if (root == root->fs_info->extent_root)
1030 root = root->fs_info->tree_root;
1031
1032 delayed_refs = &trans->transaction->delayed_refs;
1033 INIT_LIST_HEAD(&cluster);
1034again:
1035 spin_lock(&delayed_refs->lock);
1036 if (count == 0) {
1037 count = delayed_refs->num_entries * 2;
1038 run_most = 1;
1039 }
1040 while (1) {
1041 if (!(run_all || run_most) &&
1042 delayed_refs->num_heads_ready < 64)
1043 break;
1044
1045 /*
1046 * go find something we can process in the rbtree. We start at
1047 * the beginning of the tree, and then build a cluster
1048 * of refs to process starting at the first one we are able to
1049 * lock
1050 */
1051 ret = btrfs_find_ref_cluster(trans, &cluster,
1052 delayed_refs->run_delayed_start);
1053 if (ret)
1054 break;
1055
1056 ret = run_clustered_refs(trans, root, &cluster);
1057 BUG_ON(ret < 0);
1058
1059 count -= min_t(unsigned long, ret, count);
1060
1061 if (count == 0)
1062 break;
1063 }
1064
1065 if (run_all) {
1066 node = rb_first(&delayed_refs->root);
1067 if (!node)
1068 goto out;
1069 count = (unsigned long)-1;
1070
1071 while (node) {
1072 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1073 rb_node);
1074 if (btrfs_delayed_ref_is_head(ref)) {
1075 struct btrfs_delayed_ref_head *head;
1076
1077 head = btrfs_delayed_node_to_head(ref);
1078 atomic_inc(&ref->refs);
1079
1080 spin_unlock(&delayed_refs->lock);
1081 mutex_lock(&head->mutex);
1082 mutex_unlock(&head->mutex);
1083
1084 btrfs_put_delayed_ref(ref);
1085 cond_resched();
1086 goto again;
1087 }
1088 node = rb_next(node);
1089 }
1090 spin_unlock(&delayed_refs->lock);
1091 schedule_timeout(1);
1092 goto again;
1398 } 1093 }
1399 l = path->nodes[0];
1400 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1401 *refs = btrfs_extent_refs(l, item);
1402out: 1094out:
1403 btrfs_free_path(path); 1095 spin_unlock(&delayed_refs->lock);
1404 return 0; 1096 return 0;
1405} 1097}
1406 1098
@@ -1624,7 +1316,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1624 int refi = 0; 1316 int refi = 0;
1625 int slot; 1317 int slot;
1626 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, 1318 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1627 u64, u64, u64, u64, u64, u64, u64, u64); 1319 u64, u64, u64, u64, u64, u64, u64, u64, u64);
1628 1320
1629 ref_root = btrfs_header_owner(buf); 1321 ref_root = btrfs_header_owner(buf);
1630 ref_generation = btrfs_header_generation(buf); 1322 ref_generation = btrfs_header_generation(buf);
@@ -1696,12 +1388,19 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1696 1388
1697 if (level == 0) { 1389 if (level == 0) {
1698 btrfs_item_key_to_cpu(buf, &key, slot); 1390 btrfs_item_key_to_cpu(buf, &key, slot);
1391 fi = btrfs_item_ptr(buf, slot,
1392 struct btrfs_file_extent_item);
1393
1394 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1395 if (bytenr == 0)
1396 continue;
1699 1397
1700 ret = process_func(trans, root, bytenr, 1398 ret = process_func(trans, root, bytenr,
1701 orig_buf->start, buf->start, 1399 btrfs_file_extent_disk_num_bytes(buf, fi),
1702 orig_root, ref_root, 1400 orig_buf->start, buf->start,
1703 orig_generation, ref_generation, 1401 orig_root, ref_root,
1704 key.objectid); 1402 orig_generation, ref_generation,
1403 key.objectid);
1705 1404
1706 if (ret) { 1405 if (ret) {
1707 faili = slot; 1406 faili = slot;
@@ -1709,7 +1408,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1709 goto fail; 1408 goto fail;
1710 } 1409 }
1711 } else { 1410 } else {
1712 ret = process_func(trans, root, bytenr, 1411 ret = process_func(trans, root, bytenr, buf->len,
1713 orig_buf->start, buf->start, 1412 orig_buf->start, buf->start,
1714 orig_root, ref_root, 1413 orig_root, ref_root,
1715 orig_generation, ref_generation, 1414 orig_generation, ref_generation,
@@ -1786,17 +1485,17 @@ int btrfs_update_ref(struct btrfs_trans_handle *trans,
1786 if (bytenr == 0) 1485 if (bytenr == 0)
1787 continue; 1486 continue;
1788 ret = __btrfs_update_extent_ref(trans, root, bytenr, 1487 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1789 orig_buf->start, buf->start, 1488 btrfs_file_extent_disk_num_bytes(buf, fi),
1790 orig_root, ref_root, 1489 orig_buf->start, buf->start,
1791 orig_generation, ref_generation, 1490 orig_root, ref_root, orig_generation,
1792 key.objectid); 1491 ref_generation, key.objectid);
1793 if (ret) 1492 if (ret)
1794 goto fail; 1493 goto fail;
1795 } else { 1494 } else {
1796 bytenr = btrfs_node_blockptr(buf, slot); 1495 bytenr = btrfs_node_blockptr(buf, slot);
1797 ret = __btrfs_update_extent_ref(trans, root, bytenr, 1496 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1798 orig_buf->start, buf->start, 1497 buf->len, orig_buf->start,
1799 orig_root, ref_root, 1498 buf->start, orig_root, ref_root,
1800 orig_generation, ref_generation, 1499 orig_generation, ref_generation,
1801 level - 1); 1500 level - 1);
1802 if (ret) 1501 if (ret)
@@ -1815,7 +1514,6 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
1815 struct btrfs_block_group_cache *cache) 1514 struct btrfs_block_group_cache *cache)
1816{ 1515{
1817 int ret; 1516 int ret;
1818 int pending_ret;
1819 struct btrfs_root *extent_root = root->fs_info->extent_root; 1517 struct btrfs_root *extent_root = root->fs_info->extent_root;
1820 unsigned long bi; 1518 unsigned long bi;
1821 struct extent_buffer *leaf; 1519 struct extent_buffer *leaf;
@@ -1831,12 +1529,8 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
1831 btrfs_mark_buffer_dirty(leaf); 1529 btrfs_mark_buffer_dirty(leaf);
1832 btrfs_release_path(extent_root, path); 1530 btrfs_release_path(extent_root, path);
1833fail: 1531fail:
1834 finish_current_insert(trans, extent_root, 0);
1835 pending_ret = del_pending_extents(trans, extent_root, 0);
1836 if (ret) 1532 if (ret)
1837 return ret; 1533 return ret;
1838 if (pending_ret)
1839 return pending_ret;
1840 return 0; 1534 return 0;
1841 1535
1842} 1536}
@@ -2361,6 +2055,8 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
2361 clear_extent_dirty(&fs_info->pinned_extents, 2055 clear_extent_dirty(&fs_info->pinned_extents,
2362 bytenr, bytenr + num - 1, GFP_NOFS); 2056 bytenr, bytenr + num - 1, GFP_NOFS);
2363 } 2057 }
2058 mutex_unlock(&root->fs_info->pinned_mutex);
2059
2364 while (num > 0) { 2060 while (num > 0) {
2365 cache = btrfs_lookup_block_group(fs_info, bytenr); 2061 cache = btrfs_lookup_block_group(fs_info, bytenr);
2366 BUG_ON(!cache); 2062 BUG_ON(!cache);
@@ -2452,8 +2148,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2452 u64 end; 2148 u64 end;
2453 int ret; 2149 int ret;
2454 2150
2455 mutex_lock(&root->fs_info->pinned_mutex);
2456 while (1) { 2151 while (1) {
2152 mutex_lock(&root->fs_info->pinned_mutex);
2457 ret = find_first_extent_bit(unpin, 0, &start, &end, 2153 ret = find_first_extent_bit(unpin, 0, &start, &end,
2458 EXTENT_DIRTY); 2154 EXTENT_DIRTY);
2459 if (ret) 2155 if (ret)
@@ -2461,209 +2157,21 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2461 2157
2462 ret = btrfs_discard_extent(root, start, end + 1 - start); 2158 ret = btrfs_discard_extent(root, start, end + 1 - start);
2463 2159
2160 /* unlocks the pinned mutex */
2464 btrfs_update_pinned_extents(root, start, end + 1 - start, 0); 2161 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2465 clear_extent_dirty(unpin, start, end, GFP_NOFS); 2162 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2466 2163
2467 if (need_resched()) { 2164 cond_resched();
2468 mutex_unlock(&root->fs_info->pinned_mutex);
2469 cond_resched();
2470 mutex_lock(&root->fs_info->pinned_mutex);
2471 }
2472 } 2165 }
2473 mutex_unlock(&root->fs_info->pinned_mutex); 2166 mutex_unlock(&root->fs_info->pinned_mutex);
2474 return ret; 2167 return ret;
2475} 2168}
2476 2169
2477static int finish_current_insert(struct btrfs_trans_handle *trans,
2478 struct btrfs_root *extent_root, int all)
2479{
2480 u64 start;
2481 u64 end;
2482 u64 priv;
2483 u64 search = 0;
2484 struct btrfs_fs_info *info = extent_root->fs_info;
2485 struct btrfs_path *path;
2486 struct pending_extent_op *extent_op, *tmp;
2487 struct list_head insert_list, update_list;
2488 int ret;
2489 int num_inserts = 0, max_inserts, restart = 0;
2490
2491 path = btrfs_alloc_path();
2492 INIT_LIST_HEAD(&insert_list);
2493 INIT_LIST_HEAD(&update_list);
2494
2495 max_inserts = extent_root->leafsize /
2496 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2497 sizeof(struct btrfs_extent_ref) +
2498 sizeof(struct btrfs_extent_item));
2499again:
2500 mutex_lock(&info->extent_ins_mutex);
2501 while (1) {
2502 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2503 &end, EXTENT_WRITEBACK);
2504 if (ret) {
2505 if (restart && !num_inserts &&
2506 list_empty(&update_list)) {
2507 restart = 0;
2508 search = 0;
2509 continue;
2510 }
2511 break;
2512 }
2513
2514 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2515 if (!ret) {
2516 if (all)
2517 restart = 1;
2518 search = end + 1;
2519 if (need_resched()) {
2520 mutex_unlock(&info->extent_ins_mutex);
2521 cond_resched();
2522 mutex_lock(&info->extent_ins_mutex);
2523 }
2524 continue;
2525 }
2526
2527 ret = get_state_private(&info->extent_ins, start, &priv);
2528 BUG_ON(ret);
2529 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2530
2531 if (extent_op->type == PENDING_EXTENT_INSERT) {
2532 num_inserts++;
2533 list_add_tail(&extent_op->list, &insert_list);
2534 search = end + 1;
2535 if (num_inserts == max_inserts) {
2536 restart = 1;
2537 break;
2538 }
2539 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2540 list_add_tail(&extent_op->list, &update_list);
2541 search = end + 1;
2542 } else {
2543 BUG();
2544 }
2545 }
2546
2547 /*
2548 * process the update list, clear the writeback bit for it, and if
2549 * somebody marked this thing for deletion then just unlock it and be
2550 * done, the free_extents will handle it
2551 */
2552 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2553 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2554 extent_op->bytenr + extent_op->num_bytes - 1,
2555 EXTENT_WRITEBACK, GFP_NOFS);
2556 if (extent_op->del) {
2557 list_del_init(&extent_op->list);
2558 unlock_extent(&info->extent_ins, extent_op->bytenr,
2559 extent_op->bytenr + extent_op->num_bytes
2560 - 1, GFP_NOFS);
2561 kfree(extent_op);
2562 }
2563 }
2564 mutex_unlock(&info->extent_ins_mutex);
2565
2566 /*
2567 * still have things left on the update list, go ahead an update
2568 * everything
2569 */
2570 if (!list_empty(&update_list)) {
2571 ret = update_backrefs(trans, extent_root, path, &update_list);
2572 BUG_ON(ret);
2573
2574 /* we may have COW'ed new blocks, so lets start over */
2575 if (all)
2576 restart = 1;
2577 }
2578
2579 /*
2580 * if no inserts need to be done, but we skipped some extents and we
2581 * need to make sure everything is cleaned then reset everything and
2582 * go back to the beginning
2583 */
2584 if (!num_inserts && restart) {
2585 search = 0;
2586 restart = 0;
2587 INIT_LIST_HEAD(&update_list);
2588 INIT_LIST_HEAD(&insert_list);
2589 goto again;
2590 } else if (!num_inserts) {
2591 goto out;
2592 }
2593
2594 /*
2595 * process the insert extents list. Again if we are deleting this
2596 * extent, then just unlock it, pin down the bytes if need be, and be
2597 * done with it. Saves us from having to actually insert the extent
2598 * into the tree and then subsequently come along and delete it
2599 */
2600 mutex_lock(&info->extent_ins_mutex);
2601 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2602 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2603 extent_op->bytenr + extent_op->num_bytes - 1,
2604 EXTENT_WRITEBACK, GFP_NOFS);
2605 if (extent_op->del) {
2606 u64 used;
2607 list_del_init(&extent_op->list);
2608 unlock_extent(&info->extent_ins, extent_op->bytenr,
2609 extent_op->bytenr + extent_op->num_bytes
2610 - 1, GFP_NOFS);
2611
2612 mutex_lock(&extent_root->fs_info->pinned_mutex);
2613 ret = pin_down_bytes(trans, extent_root,
2614 extent_op->bytenr,
2615 extent_op->num_bytes, 0);
2616 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2617
2618 spin_lock(&info->delalloc_lock);
2619 used = btrfs_super_bytes_used(&info->super_copy);
2620 btrfs_set_super_bytes_used(&info->super_copy,
2621 used - extent_op->num_bytes);
2622 used = btrfs_root_used(&extent_root->root_item);
2623 btrfs_set_root_used(&extent_root->root_item,
2624 used - extent_op->num_bytes);
2625 spin_unlock(&info->delalloc_lock);
2626
2627 ret = update_block_group(trans, extent_root,
2628 extent_op->bytenr,
2629 extent_op->num_bytes,
2630 0, ret > 0);
2631 BUG_ON(ret);
2632 kfree(extent_op);
2633 num_inserts--;
2634 }
2635 }
2636 mutex_unlock(&info->extent_ins_mutex);
2637
2638 ret = insert_extents(trans, extent_root, path, &insert_list,
2639 num_inserts);
2640 BUG_ON(ret);
2641
2642 /*
2643 * if restart is set for whatever reason we need to go back and start
2644 * searching through the pending list again.
2645 *
2646 * We just inserted some extents, which could have resulted in new
2647 * blocks being allocated, which would result in new blocks needing
2648 * updates, so if all is set we _must_ restart to get the updated
2649 * blocks.
2650 */
2651 if (restart || all) {
2652 INIT_LIST_HEAD(&insert_list);
2653 INIT_LIST_HEAD(&update_list);
2654 search = 0;
2655 restart = 0;
2656 num_inserts = 0;
2657 goto again;
2658 }
2659out:
2660 btrfs_free_path(path);
2661 return 0;
2662}
2663
2664static int pin_down_bytes(struct btrfs_trans_handle *trans, 2170static int pin_down_bytes(struct btrfs_trans_handle *trans,
2665 struct btrfs_root *root, 2171 struct btrfs_root *root,
2666 u64 bytenr, u64 num_bytes, int is_data) 2172 struct btrfs_path *path,
2173 u64 bytenr, u64 num_bytes, int is_data,
2174 struct extent_buffer **must_clean)
2667{ 2175{
2668 int err = 0; 2176 int err = 0;
2669 struct extent_buffer *buf; 2177 struct extent_buffer *buf;
@@ -2686,17 +2194,19 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
2686 u64 header_transid = btrfs_header_generation(buf); 2194 u64 header_transid = btrfs_header_generation(buf);
2687 if (header_owner != BTRFS_TREE_LOG_OBJECTID && 2195 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2688 header_owner != BTRFS_TREE_RELOC_OBJECTID && 2196 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2197 header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2689 header_transid == trans->transid && 2198 header_transid == trans->transid &&
2690 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { 2199 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2691 clean_tree_block(NULL, root, buf); 2200 *must_clean = buf;
2692 btrfs_tree_unlock(buf);
2693 free_extent_buffer(buf);
2694 return 1; 2201 return 1;
2695 } 2202 }
2696 btrfs_tree_unlock(buf); 2203 btrfs_tree_unlock(buf);
2697 } 2204 }
2698 free_extent_buffer(buf); 2205 free_extent_buffer(buf);
2699pinit: 2206pinit:
2207 btrfs_set_path_blocking(path);
2208 mutex_lock(&root->fs_info->pinned_mutex);
2209 /* unlocks the pinned mutex */
2700 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 2210 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2701 2211
2702 BUG_ON(err < 0); 2212 BUG_ON(err < 0);
@@ -2710,7 +2220,8 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *root, 2220 struct btrfs_root *root,
2711 u64 bytenr, u64 num_bytes, u64 parent, 2221 u64 bytenr, u64 num_bytes, u64 parent,
2712 u64 root_objectid, u64 ref_generation, 2222 u64 root_objectid, u64 ref_generation,
2713 u64 owner_objectid, int pin, int mark_free) 2223 u64 owner_objectid, int pin, int mark_free,
2224 int refs_to_drop)
2714{ 2225{
2715 struct btrfs_path *path; 2226 struct btrfs_path *path;
2716 struct btrfs_key key; 2227 struct btrfs_key key;
@@ -2732,6 +2243,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2732 return -ENOMEM; 2243 return -ENOMEM;
2733 2244
2734 path->reada = 1; 2245 path->reada = 1;
2246 path->leave_spinning = 1;
2735 ret = lookup_extent_backref(trans, extent_root, path, 2247 ret = lookup_extent_backref(trans, extent_root, path,
2736 bytenr, parent, root_objectid, 2248 bytenr, parent, root_objectid,
2737 ref_generation, owner_objectid, 1); 2249 ref_generation, owner_objectid, 1);
@@ -2753,9 +2265,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2753 break; 2265 break;
2754 } 2266 }
2755 if (!found_extent) { 2267 if (!found_extent) {
2756 ret = remove_extent_backref(trans, extent_root, path); 2268 ret = remove_extent_backref(trans, extent_root, path,
2269 refs_to_drop);
2757 BUG_ON(ret); 2270 BUG_ON(ret);
2758 btrfs_release_path(extent_root, path); 2271 btrfs_release_path(extent_root, path);
2272 path->leave_spinning = 1;
2759 ret = btrfs_search_slot(trans, extent_root, 2273 ret = btrfs_search_slot(trans, extent_root,
2760 &key, path, -1, 1); 2274 &key, path, -1, 1);
2761 if (ret) { 2275 if (ret) {
@@ -2771,8 +2285,9 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2771 btrfs_print_leaf(extent_root, path->nodes[0]); 2285 btrfs_print_leaf(extent_root, path->nodes[0]);
2772 WARN_ON(1); 2286 WARN_ON(1);
2773 printk(KERN_ERR "btrfs unable to find ref byte nr %llu " 2287 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2774 "root %llu gen %llu owner %llu\n", 2288 "parent %llu root %llu gen %llu owner %llu\n",
2775 (unsigned long long)bytenr, 2289 (unsigned long long)bytenr,
2290 (unsigned long long)parent,
2776 (unsigned long long)root_objectid, 2291 (unsigned long long)root_objectid,
2777 (unsigned long long)ref_generation, 2292 (unsigned long long)ref_generation,
2778 (unsigned long long)owner_objectid); 2293 (unsigned long long)owner_objectid);
@@ -2782,17 +2297,23 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2782 ei = btrfs_item_ptr(leaf, extent_slot, 2297 ei = btrfs_item_ptr(leaf, extent_slot,
2783 struct btrfs_extent_item); 2298 struct btrfs_extent_item);
2784 refs = btrfs_extent_refs(leaf, ei); 2299 refs = btrfs_extent_refs(leaf, ei);
2785 BUG_ON(refs == 0);
2786 refs -= 1;
2787 btrfs_set_extent_refs(leaf, ei, refs);
2788 2300
2301 /*
2302 * we're not allowed to delete the extent item if there
2303 * are other delayed ref updates pending
2304 */
2305
2306 BUG_ON(refs < refs_to_drop);
2307 refs -= refs_to_drop;
2308 btrfs_set_extent_refs(leaf, ei, refs);
2789 btrfs_mark_buffer_dirty(leaf); 2309 btrfs_mark_buffer_dirty(leaf);
2790 2310
2791 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { 2311 if (refs == 0 && found_extent &&
2312 path->slots[0] == extent_slot + 1) {
2792 struct btrfs_extent_ref *ref; 2313 struct btrfs_extent_ref *ref;
2793 ref = btrfs_item_ptr(leaf, path->slots[0], 2314 ref = btrfs_item_ptr(leaf, path->slots[0],
2794 struct btrfs_extent_ref); 2315 struct btrfs_extent_ref);
2795 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); 2316 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2796 /* if the back ref and the extent are next to each other 2317 /* if the back ref and the extent are next to each other
2797 * they get deleted below in one shot 2318 * they get deleted below in one shot
2798 */ 2319 */
@@ -2800,11 +2321,13 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2800 num_to_del = 2; 2321 num_to_del = 2;
2801 } else if (found_extent) { 2322 } else if (found_extent) {
2802 /* otherwise delete the extent back ref */ 2323 /* otherwise delete the extent back ref */
2803 ret = remove_extent_backref(trans, extent_root, path); 2324 ret = remove_extent_backref(trans, extent_root, path,
2325 refs_to_drop);
2804 BUG_ON(ret); 2326 BUG_ON(ret);
2805 /* if refs are 0, we need to setup the path for deletion */ 2327 /* if refs are 0, we need to setup the path for deletion */
2806 if (refs == 0) { 2328 if (refs == 0) {
2807 btrfs_release_path(extent_root, path); 2329 btrfs_release_path(extent_root, path);
2330 path->leave_spinning = 1;
2808 ret = btrfs_search_slot(trans, extent_root, &key, path, 2331 ret = btrfs_search_slot(trans, extent_root, &key, path,
2809 -1, 1); 2332 -1, 1);
2810 BUG_ON(ret); 2333 BUG_ON(ret);
@@ -2814,16 +2337,18 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2814 if (refs == 0) { 2337 if (refs == 0) {
2815 u64 super_used; 2338 u64 super_used;
2816 u64 root_used; 2339 u64 root_used;
2340 struct extent_buffer *must_clean = NULL;
2817 2341
2818 if (pin) { 2342 if (pin) {
2819 mutex_lock(&root->fs_info->pinned_mutex); 2343 ret = pin_down_bytes(trans, root, path,
2820 ret = pin_down_bytes(trans, root, bytenr, num_bytes, 2344 bytenr, num_bytes,
2821 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); 2345 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID,
2822 mutex_unlock(&root->fs_info->pinned_mutex); 2346 &must_clean);
2823 if (ret > 0) 2347 if (ret > 0)
2824 mark_free = 1; 2348 mark_free = 1;
2825 BUG_ON(ret < 0); 2349 BUG_ON(ret < 0);
2826 } 2350 }
2351
2827 /* block accounting for super block */ 2352 /* block accounting for super block */
2828 spin_lock(&info->delalloc_lock); 2353 spin_lock(&info->delalloc_lock);
2829 super_used = btrfs_super_bytes_used(&info->super_copy); 2354 super_used = btrfs_super_bytes_used(&info->super_copy);
@@ -2835,14 +2360,34 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2835 btrfs_set_root_used(&root->root_item, 2360 btrfs_set_root_used(&root->root_item,
2836 root_used - num_bytes); 2361 root_used - num_bytes);
2837 spin_unlock(&info->delalloc_lock); 2362 spin_unlock(&info->delalloc_lock);
2363
2364 /*
2365 * it is going to be very rare for someone to be waiting
2366 * on the block we're freeing. del_items might need to
2367 * schedule, so rather than get fancy, just force it
2368 * to blocking here
2369 */
2370 if (must_clean)
2371 btrfs_set_lock_blocking(must_clean);
2372
2838 ret = btrfs_del_items(trans, extent_root, path, path->slots[0], 2373 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2839 num_to_del); 2374 num_to_del);
2840 BUG_ON(ret); 2375 BUG_ON(ret);
2841 btrfs_release_path(extent_root, path); 2376 btrfs_release_path(extent_root, path);
2842 2377
2378 if (must_clean) {
2379 clean_tree_block(NULL, root, must_clean);
2380 btrfs_tree_unlock(must_clean);
2381 free_extent_buffer(must_clean);
2382 }
2383
2843 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { 2384 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2844 ret = btrfs_del_csums(trans, root, bytenr, num_bytes); 2385 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2845 BUG_ON(ret); 2386 BUG_ON(ret);
2387 } else {
2388 invalidate_mapping_pages(info->btree_inode->i_mapping,
2389 bytenr >> PAGE_CACHE_SHIFT,
2390 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
2846 } 2391 }
2847 2392
2848 ret = update_block_group(trans, root, bytenr, num_bytes, 0, 2393 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
@@ -2850,218 +2395,103 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2850 BUG_ON(ret); 2395 BUG_ON(ret);
2851 } 2396 }
2852 btrfs_free_path(path); 2397 btrfs_free_path(path);
2853 finish_current_insert(trans, extent_root, 0);
2854 return ret; 2398 return ret;
2855} 2399}
2856 2400
2857/* 2401/*
2858 * find all the blocks marked as pending in the radix tree and remove 2402 * remove an extent from the root, returns 0 on success
2859 * them from the extent map
2860 */ 2403 */
2861static int del_pending_extents(struct btrfs_trans_handle *trans, 2404static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2862 struct btrfs_root *extent_root, int all) 2405 struct btrfs_root *root,
2406 u64 bytenr, u64 num_bytes, u64 parent,
2407 u64 root_objectid, u64 ref_generation,
2408 u64 owner_objectid, int pin,
2409 int refs_to_drop)
2863{ 2410{
2864 int ret; 2411 WARN_ON(num_bytes < root->sectorsize);
2865 int err = 0;
2866 u64 start;
2867 u64 end;
2868 u64 priv;
2869 u64 search = 0;
2870 int nr = 0, skipped = 0;
2871 struct extent_io_tree *pending_del;
2872 struct extent_io_tree *extent_ins;
2873 struct pending_extent_op *extent_op;
2874 struct btrfs_fs_info *info = extent_root->fs_info;
2875 struct list_head delete_list;
2876
2877 INIT_LIST_HEAD(&delete_list);
2878 extent_ins = &extent_root->fs_info->extent_ins;
2879 pending_del = &extent_root->fs_info->pending_del;
2880
2881again:
2882 mutex_lock(&info->extent_ins_mutex);
2883 while (1) {
2884 ret = find_first_extent_bit(pending_del, search, &start, &end,
2885 EXTENT_WRITEBACK);
2886 if (ret) {
2887 if (all && skipped && !nr) {
2888 search = 0;
2889 skipped = 0;
2890 continue;
2891 }
2892 mutex_unlock(&info->extent_ins_mutex);
2893 break;
2894 }
2895
2896 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2897 if (!ret) {
2898 search = end+1;
2899 skipped = 1;
2900
2901 if (need_resched()) {
2902 mutex_unlock(&info->extent_ins_mutex);
2903 cond_resched();
2904 mutex_lock(&info->extent_ins_mutex);
2905 }
2906
2907 continue;
2908 }
2909 BUG_ON(ret < 0);
2910
2911 ret = get_state_private(pending_del, start, &priv);
2912 BUG_ON(ret);
2913 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2914
2915 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2916 GFP_NOFS);
2917 if (!test_range_bit(extent_ins, start, end,
2918 EXTENT_WRITEBACK, 0)) {
2919 list_add_tail(&extent_op->list, &delete_list);
2920 nr++;
2921 } else {
2922 kfree(extent_op);
2923
2924 ret = get_state_private(&info->extent_ins, start,
2925 &priv);
2926 BUG_ON(ret);
2927 extent_op = (struct pending_extent_op *)
2928 (unsigned long)priv;
2929
2930 clear_extent_bits(&info->extent_ins, start, end,
2931 EXTENT_WRITEBACK, GFP_NOFS);
2932
2933 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2934 list_add_tail(&extent_op->list, &delete_list);
2935 search = end + 1;
2936 nr++;
2937 continue;
2938 }
2939
2940 mutex_lock(&extent_root->fs_info->pinned_mutex);
2941 ret = pin_down_bytes(trans, extent_root, start,
2942 end + 1 - start, 0);
2943 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2944
2945 ret = update_block_group(trans, extent_root, start,
2946 end + 1 - start, 0, ret > 0);
2947
2948 unlock_extent(extent_ins, start, end, GFP_NOFS);
2949 BUG_ON(ret);
2950 kfree(extent_op);
2951 }
2952 if (ret)
2953 err = ret;
2954
2955 search = end + 1;
2956
2957 if (need_resched()) {
2958 mutex_unlock(&info->extent_ins_mutex);
2959 cond_resched();
2960 mutex_lock(&info->extent_ins_mutex);
2961 }
2962 }
2963 2412
2964 if (nr) { 2413 /*
2965 ret = free_extents(trans, extent_root, &delete_list); 2414 * if metadata always pin
2966 BUG_ON(ret); 2415 * if data pin when any transaction has committed this
2967 } 2416 */
2417 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2418 ref_generation != trans->transid)
2419 pin = 1;
2968 2420
2969 if (all && skipped) { 2421 if (ref_generation != trans->transid)
2970 INIT_LIST_HEAD(&delete_list); 2422 pin = 1;
2971 search = 0;
2972 nr = 0;
2973 goto again;
2974 }
2975 2423
2976 if (!err) 2424 return __free_extent(trans, root, bytenr, num_bytes, parent,
2977 finish_current_insert(trans, extent_root, 0); 2425 root_objectid, ref_generation,
2978 return err; 2426 owner_objectid, pin, pin == 0, refs_to_drop);
2979} 2427}
2980 2428
2981/* 2429/*
2982 * remove an extent from the root, returns 0 on success 2430 * when we free an extent, it is possible (and likely) that we free the last
2431 * delayed ref for that extent as well. This searches the delayed ref tree for
2432 * a given extent, and if there are no other delayed refs to be processed, it
2433 * removes it from the tree.
2983 */ 2434 */
2984static int __btrfs_free_extent(struct btrfs_trans_handle *trans, 2435static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2985 struct btrfs_root *root, 2436 struct btrfs_root *root, u64 bytenr)
2986 u64 bytenr, u64 num_bytes, u64 parent,
2987 u64 root_objectid, u64 ref_generation,
2988 u64 owner_objectid, int pin)
2989{ 2437{
2990 struct btrfs_root *extent_root = root->fs_info->extent_root; 2438 struct btrfs_delayed_ref_head *head;
2991 int pending_ret; 2439 struct btrfs_delayed_ref_root *delayed_refs;
2440 struct btrfs_delayed_ref_node *ref;
2441 struct rb_node *node;
2992 int ret; 2442 int ret;
2993 2443
2994 WARN_ON(num_bytes < root->sectorsize); 2444 delayed_refs = &trans->transaction->delayed_refs;
2995 if (root == extent_root) { 2445 spin_lock(&delayed_refs->lock);
2996 struct pending_extent_op *extent_op = NULL; 2446 head = btrfs_find_delayed_ref_head(trans, bytenr);
2997 2447 if (!head)
2998 mutex_lock(&root->fs_info->extent_ins_mutex); 2448 goto out;
2999 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
3000 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
3001 u64 priv;
3002 ret = get_state_private(&root->fs_info->extent_ins,
3003 bytenr, &priv);
3004 BUG_ON(ret);
3005 extent_op = (struct pending_extent_op *)
3006 (unsigned long)priv;
3007 2449
3008 extent_op->del = 1; 2450 node = rb_prev(&head->node.rb_node);
3009 if (extent_op->type == PENDING_EXTENT_INSERT) { 2451 if (!node)
3010 mutex_unlock(&root->fs_info->extent_ins_mutex); 2452 goto out;
3011 return 0;
3012 }
3013 }
3014 2453
3015 if (extent_op) { 2454 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3016 ref_generation = extent_op->orig_generation;
3017 parent = extent_op->orig_parent;
3018 }
3019 2455
3020 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 2456 /* there are still entries for this ref, we can't drop it */
3021 BUG_ON(!extent_op); 2457 if (ref->bytenr == bytenr)
3022 2458 goto out;
3023 extent_op->type = PENDING_EXTENT_DELETE;
3024 extent_op->bytenr = bytenr;
3025 extent_op->num_bytes = num_bytes;
3026 extent_op->parent = parent;
3027 extent_op->orig_parent = parent;
3028 extent_op->generation = ref_generation;
3029 extent_op->orig_generation = ref_generation;
3030 extent_op->level = (int)owner_objectid;
3031 INIT_LIST_HEAD(&extent_op->list);
3032 extent_op->del = 0;
3033
3034 set_extent_bits(&root->fs_info->pending_del,
3035 bytenr, bytenr + num_bytes - 1,
3036 EXTENT_WRITEBACK, GFP_NOFS);
3037 set_state_private(&root->fs_info->pending_del,
3038 bytenr, (unsigned long)extent_op);
3039 mutex_unlock(&root->fs_info->extent_ins_mutex);
3040 return 0;
3041 }
3042 /* if metadata always pin */
3043 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3044 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3045 mutex_lock(&root->fs_info->pinned_mutex);
3046 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3047 mutex_unlock(&root->fs_info->pinned_mutex);
3048 update_reserved_extents(root, bytenr, num_bytes, 0);
3049 return 0;
3050 }
3051 pin = 1;
3052 }
3053 2459
3054 /* if data pin when any transaction has committed this */ 2460 /*
3055 if (ref_generation != trans->transid) 2461 * waiting for the lock here would deadlock. If someone else has it
3056 pin = 1; 2462 * locked they are already in the process of dropping it anyway
2463 */
2464 if (!mutex_trylock(&head->mutex))
2465 goto out;
3057 2466
3058 ret = __free_extent(trans, root, bytenr, num_bytes, parent, 2467 /*
3059 root_objectid, ref_generation, 2468 * at this point we have a head with no other entries. Go
3060 owner_objectid, pin, pin == 0); 2469 * ahead and process it.
2470 */
2471 head->node.in_tree = 0;
2472 rb_erase(&head->node.rb_node, &delayed_refs->root);
3061 2473
3062 finish_current_insert(trans, root->fs_info->extent_root, 0); 2474 delayed_refs->num_entries--;
3063 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); 2475
3064 return ret ? ret : pending_ret; 2476 /*
2477 * we don't take a ref on the node because we're removing it from the
2478 * tree, so we just steal the ref the tree was holding.
2479 */
2480 delayed_refs->num_heads--;
2481 if (list_empty(&head->cluster))
2482 delayed_refs->num_heads_ready--;
2483
2484 list_del_init(&head->cluster);
2485 spin_unlock(&delayed_refs->lock);
2486
2487 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2488 &head->node, head->must_insert_reserved);
2489 BUG_ON(ret);
2490 btrfs_put_delayed_ref(&head->node);
2491 return 0;
2492out:
2493 spin_unlock(&delayed_refs->lock);
2494 return 0;
3065} 2495}
3066 2496
3067int btrfs_free_extent(struct btrfs_trans_handle *trans, 2497int btrfs_free_extent(struct btrfs_trans_handle *trans,
@@ -3072,9 +2502,30 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
3072{ 2502{
3073 int ret; 2503 int ret;
3074 2504
3075 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, 2505 /*
3076 root_objectid, ref_generation, 2506 * tree log blocks never actually go into the extent allocation
3077 owner_objectid, pin); 2507 * tree, just update pinning info and exit early.
2508 *
2509 * data extents referenced by the tree log do need to have
2510 * their reference counts bumped.
2511 */
2512 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2513 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2514 mutex_lock(&root->fs_info->pinned_mutex);
2515
2516 /* unlocks the pinned mutex */
2517 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2518 update_reserved_extents(root, bytenr, num_bytes, 0);
2519 ret = 0;
2520 } else {
2521 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2522 root_objectid, ref_generation,
2523 owner_objectid,
2524 BTRFS_DROP_DELAYED_REF, 1);
2525 BUG_ON(ret);
2526 ret = check_ref_cleanup(trans, root, bytenr);
2527 BUG_ON(ret);
2528 }
3078 return ret; 2529 return ret;
3079} 2530}
3080 2531
@@ -3475,10 +2926,10 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3475static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, 2926static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3476 struct btrfs_root *root, u64 parent, 2927 struct btrfs_root *root, u64 parent,
3477 u64 root_objectid, u64 ref_generation, 2928 u64 root_objectid, u64 ref_generation,
3478 u64 owner, struct btrfs_key *ins) 2929 u64 owner, struct btrfs_key *ins,
2930 int ref_mod)
3479{ 2931{
3480 int ret; 2932 int ret;
3481 int pending_ret;
3482 u64 super_used; 2933 u64 super_used;
3483 u64 root_used; 2934 u64 root_used;
3484 u64 num_bytes = ins->offset; 2935 u64 num_bytes = ins->offset;
@@ -3503,33 +2954,6 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3503 btrfs_set_root_used(&root->root_item, root_used + num_bytes); 2954 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3504 spin_unlock(&info->delalloc_lock); 2955 spin_unlock(&info->delalloc_lock);
3505 2956
3506 if (root == extent_root) {
3507 struct pending_extent_op *extent_op;
3508
3509 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3510 BUG_ON(!extent_op);
3511
3512 extent_op->type = PENDING_EXTENT_INSERT;
3513 extent_op->bytenr = ins->objectid;
3514 extent_op->num_bytes = ins->offset;
3515 extent_op->parent = parent;
3516 extent_op->orig_parent = 0;
3517 extent_op->generation = ref_generation;
3518 extent_op->orig_generation = 0;
3519 extent_op->level = (int)owner;
3520 INIT_LIST_HEAD(&extent_op->list);
3521 extent_op->del = 0;
3522
3523 mutex_lock(&root->fs_info->extent_ins_mutex);
3524 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3525 ins->objectid + ins->offset - 1,
3526 EXTENT_WRITEBACK, GFP_NOFS);
3527 set_state_private(&root->fs_info->extent_ins,
3528 ins->objectid, (unsigned long)extent_op);
3529 mutex_unlock(&root->fs_info->extent_ins_mutex);
3530 goto update_block;
3531 }
3532
3533 memcpy(&keys[0], ins, sizeof(*ins)); 2957 memcpy(&keys[0], ins, sizeof(*ins));
3534 keys[1].objectid = ins->objectid; 2958 keys[1].objectid = ins->objectid;
3535 keys[1].type = BTRFS_EXTENT_REF_KEY; 2959 keys[1].type = BTRFS_EXTENT_REF_KEY;
@@ -3540,37 +2964,31 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3540 path = btrfs_alloc_path(); 2964 path = btrfs_alloc_path();
3541 BUG_ON(!path); 2965 BUG_ON(!path);
3542 2966
2967 path->leave_spinning = 1;
3543 ret = btrfs_insert_empty_items(trans, extent_root, path, keys, 2968 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3544 sizes, 2); 2969 sizes, 2);
3545 BUG_ON(ret); 2970 BUG_ON(ret);
3546 2971
3547 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], 2972 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3548 struct btrfs_extent_item); 2973 struct btrfs_extent_item);
3549 btrfs_set_extent_refs(path->nodes[0], extent_item, 1); 2974 btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
3550 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, 2975 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3551 struct btrfs_extent_ref); 2976 struct btrfs_extent_ref);
3552 2977
3553 btrfs_set_ref_root(path->nodes[0], ref, root_objectid); 2978 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3554 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); 2979 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3555 btrfs_set_ref_objectid(path->nodes[0], ref, owner); 2980 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3556 btrfs_set_ref_num_refs(path->nodes[0], ref, 1); 2981 btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
3557 2982
3558 btrfs_mark_buffer_dirty(path->nodes[0]); 2983 btrfs_mark_buffer_dirty(path->nodes[0]);
3559 2984
3560 trans->alloc_exclude_start = 0; 2985 trans->alloc_exclude_start = 0;
3561 trans->alloc_exclude_nr = 0; 2986 trans->alloc_exclude_nr = 0;
3562 btrfs_free_path(path); 2987 btrfs_free_path(path);
3563 finish_current_insert(trans, extent_root, 0);
3564 pending_ret = del_pending_extents(trans, extent_root, 0);
3565 2988
3566 if (ret) 2989 if (ret)
3567 goto out; 2990 goto out;
3568 if (pending_ret) {
3569 ret = pending_ret;
3570 goto out;
3571 }
3572 2991
3573update_block:
3574 ret = update_block_group(trans, root, ins->objectid, 2992 ret = update_block_group(trans, root, ins->objectid,
3575 ins->offset, 1, 0); 2993 ins->offset, 1, 0);
3576 if (ret) { 2994 if (ret) {
@@ -3592,9 +3010,12 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3592 3010
3593 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) 3011 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3594 return 0; 3012 return 0;
3595 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, 3013
3596 ref_generation, owner, ins); 3014 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3597 update_reserved_extents(root, ins->objectid, ins->offset, 0); 3015 ins->offset, parent, root_objectid,
3016 ref_generation, owner,
3017 BTRFS_ADD_DELAYED_EXTENT, 0);
3018 BUG_ON(ret);
3598 return ret; 3019 return ret;
3599} 3020}
3600 3021
@@ -3621,7 +3042,7 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3621 BUG_ON(ret); 3042 BUG_ON(ret);
3622 put_block_group(block_group); 3043 put_block_group(block_group);
3623 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, 3044 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3624 ref_generation, owner, ins); 3045 ref_generation, owner, ins, 1);
3625 return ret; 3046 return ret;
3626} 3047}
3627 3048
@@ -3640,20 +3061,18 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3640 u64 search_end, struct btrfs_key *ins, u64 data) 3061 u64 search_end, struct btrfs_key *ins, u64 data)
3641{ 3062{
3642 int ret; 3063 int ret;
3643
3644 ret = __btrfs_reserve_extent(trans, root, num_bytes, 3064 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3645 min_alloc_size, empty_size, hint_byte, 3065 min_alloc_size, empty_size, hint_byte,
3646 search_end, ins, data); 3066 search_end, ins, data);
3647 BUG_ON(ret); 3067 BUG_ON(ret);
3648 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { 3068 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3649 ret = __btrfs_alloc_reserved_extent(trans, root, parent, 3069 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3650 root_objectid, ref_generation, 3070 ins->offset, parent, root_objectid,
3651 owner_objectid, ins); 3071 ref_generation, owner_objectid,
3072 BTRFS_ADD_DELAYED_EXTENT, 0);
3652 BUG_ON(ret); 3073 BUG_ON(ret);
3653
3654 } else {
3655 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3656 } 3074 }
3075 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3657 return ret; 3076 return ret;
3658} 3077}
3659 3078
@@ -3789,7 +3208,7 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3789 3208
3790 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 3209 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3791 3210
3792 ret = __btrfs_free_extent(trans, root, disk_bytenr, 3211 ret = btrfs_free_extent(trans, root, disk_bytenr,
3793 btrfs_file_extent_disk_num_bytes(leaf, fi), 3212 btrfs_file_extent_disk_num_bytes(leaf, fi),
3794 leaf->start, leaf_owner, leaf_generation, 3213 leaf->start, leaf_owner, leaf_generation,
3795 key.objectid, 0); 3214 key.objectid, 0);
@@ -3829,7 +3248,7 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3829 */ 3248 */
3830 for (i = 0; i < ref->nritems; i++) { 3249 for (i = 0; i < ref->nritems; i++) {
3831 info = ref->extents + sorted[i].slot; 3250 info = ref->extents + sorted[i].slot;
3832 ret = __btrfs_free_extent(trans, root, info->bytenr, 3251 ret = btrfs_free_extent(trans, root, info->bytenr,
3833 info->num_bytes, ref->bytenr, 3252 info->num_bytes, ref->bytenr,
3834 ref->owner, ref->generation, 3253 ref->owner, ref->generation,
3835 info->objectid, 0); 3254 info->objectid, 0);
@@ -3846,12 +3265,13 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3846 return 0; 3265 return 0;
3847} 3266}
3848 3267
3849static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, 3268static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3269 struct btrfs_root *root, u64 start,
3850 u64 len, u32 *refs) 3270 u64 len, u32 *refs)
3851{ 3271{
3852 int ret; 3272 int ret;
3853 3273
3854 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); 3274 ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3855 BUG_ON(ret); 3275 BUG_ON(ret);
3856 3276
3857#if 0 /* some debugging code in case we see problems here */ 3277#if 0 /* some debugging code in case we see problems here */
@@ -3959,7 +3379,8 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3959 * we just decrement it below and don't update any 3379 * we just decrement it below and don't update any
3960 * of the refs the leaf points to. 3380 * of the refs the leaf points to.
3961 */ 3381 */
3962 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); 3382 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3383 blocksize, &refs);
3963 BUG_ON(ret); 3384 BUG_ON(ret);
3964 if (refs != 1) 3385 if (refs != 1)
3965 continue; 3386 continue;
@@ -4010,7 +3431,7 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4010 */ 3431 */
4011 for (i = 0; i < refi; i++) { 3432 for (i = 0; i < refi; i++) {
4012 bytenr = sorted[i].bytenr; 3433 bytenr = sorted[i].bytenr;
4013 ret = __btrfs_free_extent(trans, root, bytenr, 3434 ret = btrfs_free_extent(trans, root, bytenr,
4014 blocksize, eb->start, 3435 blocksize, eb->start,
4015 root_owner, root_gen, 0, 1); 3436 root_owner, root_gen, 0, 1);
4016 BUG_ON(ret); 3437 BUG_ON(ret);
@@ -4053,7 +3474,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4053 3474
4054 WARN_ON(*level < 0); 3475 WARN_ON(*level < 0);
4055 WARN_ON(*level >= BTRFS_MAX_LEVEL); 3476 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4056 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, 3477 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4057 path->nodes[*level]->len, &refs); 3478 path->nodes[*level]->len, &refs);
4058 BUG_ON(ret); 3479 BUG_ON(ret);
4059 if (refs > 1) 3480 if (refs > 1)
@@ -4104,7 +3525,8 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4104 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); 3525 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4105 blocksize = btrfs_level_size(root, *level - 1); 3526 blocksize = btrfs_level_size(root, *level - 1);
4106 3527
4107 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); 3528 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3529 blocksize, &refs);
4108 BUG_ON(ret); 3530 BUG_ON(ret);
4109 3531
4110 /* 3532 /*
@@ -4119,7 +3541,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4119 root_gen = btrfs_header_generation(parent); 3541 root_gen = btrfs_header_generation(parent);
4120 path->slots[*level]++; 3542 path->slots[*level]++;
4121 3543
4122 ret = __btrfs_free_extent(trans, root, bytenr, 3544 ret = btrfs_free_extent(trans, root, bytenr,
4123 blocksize, parent->start, 3545 blocksize, parent->start,
4124 root_owner, root_gen, 3546 root_owner, root_gen,
4125 *level - 1, 1); 3547 *level - 1, 1);
@@ -4165,7 +3587,7 @@ out:
4165 * cleanup and free the reference on the last node 3587 * cleanup and free the reference on the last node
4166 * we processed 3588 * we processed
4167 */ 3589 */
4168 ret = __btrfs_free_extent(trans, root, bytenr, blocksize, 3590 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4169 parent->start, root_owner, root_gen, 3591 parent->start, root_owner, root_gen,
4170 *level, 1); 3592 *level, 1);
4171 free_extent_buffer(path->nodes[*level]); 3593 free_extent_buffer(path->nodes[*level]);
@@ -4354,6 +3776,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4354 struct btrfs_path *path; 3776 struct btrfs_path *path;
4355 int i; 3777 int i;
4356 int orig_level; 3778 int orig_level;
3779 int update_count;
4357 struct btrfs_root_item *root_item = &root->root_item; 3780 struct btrfs_root_item *root_item = &root->root_item;
4358 3781
4359 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); 3782 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
@@ -4395,6 +3818,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4395 } 3818 }
4396 } 3819 }
4397 while (1) { 3820 while (1) {
3821 unsigned long update;
4398 wret = walk_down_tree(trans, root, path, &level); 3822 wret = walk_down_tree(trans, root, path, &level);
4399 if (wret > 0) 3823 if (wret > 0)
4400 break; 3824 break;
@@ -4407,12 +3831,21 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4407 break; 3831 break;
4408 if (wret < 0) 3832 if (wret < 0)
4409 ret = wret; 3833 ret = wret;
4410 if (trans->transaction->in_commit) { 3834 if (trans->transaction->in_commit ||
3835 trans->transaction->delayed_refs.flushing) {
4411 ret = -EAGAIN; 3836 ret = -EAGAIN;
4412 break; 3837 break;
4413 } 3838 }
4414 atomic_inc(&root->fs_info->throttle_gen); 3839 atomic_inc(&root->fs_info->throttle_gen);
4415 wake_up(&root->fs_info->transaction_throttle); 3840 wake_up(&root->fs_info->transaction_throttle);
3841 for (update_count = 0; update_count < 16; update_count++) {
3842 update = trans->delayed_ref_updates;
3843 trans->delayed_ref_updates = 0;
3844 if (update)
3845 btrfs_run_delayed_refs(trans, root, update);
3846 else
3847 break;
3848 }
4416 } 3849 }
4417 for (i = 0; i <= orig_level; i++) { 3850 for (i = 0; i <= orig_level; i++) {
4418 if (path->nodes[i]) { 3851 if (path->nodes[i]) {
@@ -5457,6 +4890,7 @@ static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5457 root->root_key.objectid, 4890 root->root_key.objectid,
5458 trans->transid, key.objectid); 4891 trans->transid, key.objectid);
5459 BUG_ON(ret); 4892 BUG_ON(ret);
4893
5460 ret = btrfs_free_extent(trans, root, 4894 ret = btrfs_free_extent(trans, root,
5461 bytenr, num_bytes, leaf->start, 4895 bytenr, num_bytes, leaf->start,
5462 btrfs_header_owner(leaf), 4896 btrfs_header_owner(leaf),
@@ -5768,9 +5202,6 @@ static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5768 ref_path, NULL, NULL); 5202 ref_path, NULL, NULL);
5769 BUG_ON(ret); 5203 BUG_ON(ret);
5770 5204
5771 if (root == root->fs_info->extent_root)
5772 btrfs_extent_post_op(trans, root);
5773
5774 return 0; 5205 return 0;
5775} 5206}
5776 5207
@@ -6038,6 +5469,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6038 if (!path) 5469 if (!path)
6039 return -ENOMEM; 5470 return -ENOMEM;
6040 5471
5472 path->leave_spinning = 1;
6041 ret = btrfs_insert_empty_inode(trans, root, path, objectid); 5473 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6042 if (ret) 5474 if (ret)
6043 goto out; 5475 goto out;
@@ -6208,6 +5640,9 @@ again:
6208 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); 5640 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6209 mutex_unlock(&root->fs_info->cleaner_mutex); 5641 mutex_unlock(&root->fs_info->cleaner_mutex);
6210 5642
5643 trans = btrfs_start_transaction(info->tree_root, 1);
5644 btrfs_commit_transaction(trans, info->tree_root);
5645
6211 while (1) { 5646 while (1) {
6212 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5647 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6213 if (ret < 0) 5648 if (ret < 0)
@@ -6466,7 +5901,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6466 5901
6467 extent_root = root->fs_info->extent_root; 5902 extent_root = root->fs_info->extent_root;
6468 5903
6469 root->fs_info->last_trans_new_blockgroup = trans->transid; 5904 root->fs_info->last_trans_log_full_commit = trans->transid;
6470 5905
6471 cache = kzalloc(sizeof(*cache), GFP_NOFS); 5906 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6472 if (!cache) 5907 if (!cache)
@@ -6500,9 +5935,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6500 sizeof(cache->item)); 5935 sizeof(cache->item));
6501 BUG_ON(ret); 5936 BUG_ON(ret);
6502 5937
6503 finish_current_insert(trans, extent_root, 0);
6504 ret = del_pending_extents(trans, extent_root, 0);
6505 BUG_ON(ret);
6506 set_avail_alloc_bits(extent_root->fs_info, type); 5938 set_avail_alloc_bits(extent_root->fs_info, type);
6507 5939
6508 return 0; 5940 return 0;
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index ebe6b29e6069..08085af089e2 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3124,20 +3124,15 @@ void free_extent_buffer(struct extent_buffer *eb)
3124int clear_extent_buffer_dirty(struct extent_io_tree *tree, 3124int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3125 struct extent_buffer *eb) 3125 struct extent_buffer *eb)
3126{ 3126{
3127 int set;
3128 unsigned long i; 3127 unsigned long i;
3129 unsigned long num_pages; 3128 unsigned long num_pages;
3130 struct page *page; 3129 struct page *page;
3131 3130
3132 u64 start = eb->start;
3133 u64 end = start + eb->len - 1;
3134
3135 set = clear_extent_dirty(tree, start, end, GFP_NOFS);
3136 num_pages = num_extent_pages(eb->start, eb->len); 3131 num_pages = num_extent_pages(eb->start, eb->len);
3137 3132
3138 for (i = 0; i < num_pages; i++) { 3133 for (i = 0; i < num_pages; i++) {
3139 page = extent_buffer_page(eb, i); 3134 page = extent_buffer_page(eb, i);
3140 if (!set && !PageDirty(page)) 3135 if (!PageDirty(page))
3141 continue; 3136 continue;
3142 3137
3143 lock_page(page); 3138 lock_page(page);
@@ -3146,22 +3141,6 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree,
3146 else 3141 else
3147 set_page_private(page, EXTENT_PAGE_PRIVATE); 3142 set_page_private(page, EXTENT_PAGE_PRIVATE);
3148 3143
3149 /*
3150 * if we're on the last page or the first page and the
3151 * block isn't aligned on a page boundary, do extra checks
3152 * to make sure we don't clean page that is partially dirty
3153 */
3154 if ((i == 0 && (eb->start & (PAGE_CACHE_SIZE - 1))) ||
3155 ((i == num_pages - 1) &&
3156 ((eb->start + eb->len) & (PAGE_CACHE_SIZE - 1)))) {
3157 start = (u64)page->index << PAGE_CACHE_SHIFT;
3158 end = start + PAGE_CACHE_SIZE - 1;
3159 if (test_range_bit(tree, start, end,
3160 EXTENT_DIRTY, 0)) {
3161 unlock_page(page);
3162 continue;
3163 }
3164 }
3165 clear_page_dirty_for_io(page); 3144 clear_page_dirty_for_io(page);
3166 spin_lock_irq(&page->mapping->tree_lock); 3145 spin_lock_irq(&page->mapping->tree_lock);
3167 if (!PageDirty(page)) { 3146 if (!PageDirty(page)) {
@@ -3187,29 +3166,13 @@ int set_extent_buffer_dirty(struct extent_io_tree *tree,
3187{ 3166{
3188 unsigned long i; 3167 unsigned long i;
3189 unsigned long num_pages; 3168 unsigned long num_pages;
3169 int was_dirty = 0;
3190 3170
3171 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
3191 num_pages = num_extent_pages(eb->start, eb->len); 3172 num_pages = num_extent_pages(eb->start, eb->len);
3192 for (i = 0; i < num_pages; i++) { 3173 for (i = 0; i < num_pages; i++)
3193 struct page *page = extent_buffer_page(eb, i);
3194 /* writepage may need to do something special for the
3195 * first page, we have to make sure page->private is
3196 * properly set. releasepage may drop page->private
3197 * on us if the page isn't already dirty.
3198 */
3199 lock_page(page);
3200 if (i == 0) {
3201 set_page_extent_head(page, eb->len);
3202 } else if (PagePrivate(page) &&
3203 page->private != EXTENT_PAGE_PRIVATE) {
3204 set_page_extent_mapped(page);
3205 }
3206 __set_page_dirty_nobuffers(extent_buffer_page(eb, i)); 3174 __set_page_dirty_nobuffers(extent_buffer_page(eb, i));
3207 set_extent_dirty(tree, page_offset(page), 3175 return was_dirty;
3208 page_offset(page) + PAGE_CACHE_SIZE - 1,
3209 GFP_NOFS);
3210 unlock_page(page);
3211 }
3212 return 0;
3213} 3176}
3214 3177
3215int clear_extent_buffer_uptodate(struct extent_io_tree *tree, 3178int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
@@ -3789,6 +3752,10 @@ int try_release_extent_buffer(struct extent_io_tree *tree, struct page *page)
3789 ret = 0; 3752 ret = 0;
3790 goto out; 3753 goto out;
3791 } 3754 }
3755 if (test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3756 ret = 0;
3757 goto out;
3758 }
3792 /* at this point we can safely release the extent buffer */ 3759 /* at this point we can safely release the extent buffer */
3793 num_pages = num_extent_pages(eb->start, eb->len); 3760 num_pages = num_extent_pages(eb->start, eb->len);
3794 for (i = 0; i < num_pages; i++) 3761 for (i = 0; i < num_pages; i++)
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 1f9df88afbf6..5bc20abf3f3d 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -25,6 +25,7 @@
25/* these are bit numbers for test/set bit */ 25/* these are bit numbers for test/set bit */
26#define EXTENT_BUFFER_UPTODATE 0 26#define EXTENT_BUFFER_UPTODATE 0
27#define EXTENT_BUFFER_BLOCKING 1 27#define EXTENT_BUFFER_BLOCKING 1
28#define EXTENT_BUFFER_DIRTY 2
28 29
29/* 30/*
30 * page->private values. Every page that is controlled by the extent 31 * page->private values. Every page that is controlled by the extent
@@ -254,6 +255,8 @@ int clear_extent_buffer_dirty(struct extent_io_tree *tree,
254 struct extent_buffer *eb); 255 struct extent_buffer *eb);
255int set_extent_buffer_dirty(struct extent_io_tree *tree, 256int set_extent_buffer_dirty(struct extent_io_tree *tree,
256 struct extent_buffer *eb); 257 struct extent_buffer *eb);
258int test_extent_buffer_dirty(struct extent_io_tree *tree,
259 struct extent_buffer *eb);
257int set_extent_buffer_uptodate(struct extent_io_tree *tree, 260int set_extent_buffer_uptodate(struct extent_io_tree *tree,
258 struct extent_buffer *eb); 261 struct extent_buffer *eb);
259int clear_extent_buffer_uptodate(struct extent_io_tree *tree, 262int clear_extent_buffer_uptodate(struct extent_io_tree *tree,
diff --git a/fs/btrfs/file-item.c b/fs/btrfs/file-item.c
index 964652435fd1..9b99886562d0 100644
--- a/fs/btrfs/file-item.c
+++ b/fs/btrfs/file-item.c
@@ -52,6 +52,7 @@ int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
52 file_key.offset = pos; 52 file_key.offset = pos;
53 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY); 53 btrfs_set_key_type(&file_key, BTRFS_EXTENT_DATA_KEY);
54 54
55 path->leave_spinning = 1;
55 ret = btrfs_insert_empty_item(trans, root, path, &file_key, 56 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
56 sizeof(*item)); 57 sizeof(*item));
57 if (ret < 0) 58 if (ret < 0)
@@ -523,6 +524,7 @@ int btrfs_del_csums(struct btrfs_trans_handle *trans,
523 key.offset = end_byte - 1; 524 key.offset = end_byte - 1;
524 key.type = BTRFS_EXTENT_CSUM_KEY; 525 key.type = BTRFS_EXTENT_CSUM_KEY;
525 526
527 path->leave_spinning = 1;
526 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 528 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
527 if (ret > 0) { 529 if (ret > 0) {
528 if (path->slots[0] == 0) 530 if (path->slots[0] == 0)
@@ -757,8 +759,10 @@ insert:
757 } else { 759 } else {
758 ins_size = csum_size; 760 ins_size = csum_size;
759 } 761 }
762 path->leave_spinning = 1;
760 ret = btrfs_insert_empty_item(trans, root, path, &file_key, 763 ret = btrfs_insert_empty_item(trans, root, path, &file_key,
761 ins_size); 764 ins_size);
765 path->leave_spinning = 0;
762 if (ret < 0) 766 if (ret < 0)
763 goto fail_unlock; 767 goto fail_unlock;
764 if (ret != 0) { 768 if (ret != 0) {
@@ -776,7 +780,6 @@ found:
776 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end + 780 item_end = (struct btrfs_csum_item *)((unsigned char *)item_end +
777 btrfs_item_size_nr(leaf, path->slots[0])); 781 btrfs_item_size_nr(leaf, path->slots[0]));
778 eb_token = NULL; 782 eb_token = NULL;
779 cond_resched();
780next_sector: 783next_sector:
781 784
782 if (!eb_token || 785 if (!eb_token ||
@@ -817,9 +820,9 @@ next_sector:
817 eb_token = NULL; 820 eb_token = NULL;
818 } 821 }
819 btrfs_mark_buffer_dirty(path->nodes[0]); 822 btrfs_mark_buffer_dirty(path->nodes[0]);
820 cond_resched();
821 if (total_bytes < sums->len) { 823 if (total_bytes < sums->len) {
822 btrfs_release_path(root, path); 824 btrfs_release_path(root, path);
825 cond_resched();
823 goto again; 826 goto again;
824 } 827 }
825out: 828out:
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index dc78954861b3..9c9fb46ccd08 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -606,6 +606,7 @@ next_slot:
606 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY); 606 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
607 607
608 btrfs_release_path(root, path); 608 btrfs_release_path(root, path);
609 path->leave_spinning = 1;
609 ret = btrfs_insert_empty_item(trans, root, path, &ins, 610 ret = btrfs_insert_empty_item(trans, root, path, &ins,
610 sizeof(*extent)); 611 sizeof(*extent));
611 BUG_ON(ret); 612 BUG_ON(ret);
@@ -639,17 +640,22 @@ next_slot:
639 ram_bytes); 640 ram_bytes);
640 btrfs_set_file_extent_type(leaf, extent, found_type); 641 btrfs_set_file_extent_type(leaf, extent, found_type);
641 642
643 btrfs_unlock_up_safe(path, 1);
642 btrfs_mark_buffer_dirty(path->nodes[0]); 644 btrfs_mark_buffer_dirty(path->nodes[0]);
645 btrfs_set_lock_blocking(path->nodes[0]);
643 646
644 if (disk_bytenr != 0) { 647 if (disk_bytenr != 0) {
645 ret = btrfs_update_extent_ref(trans, root, 648 ret = btrfs_update_extent_ref(trans, root,
646 disk_bytenr, orig_parent, 649 disk_bytenr,
650 le64_to_cpu(old.disk_num_bytes),
651 orig_parent,
647 leaf->start, 652 leaf->start,
648 root->root_key.objectid, 653 root->root_key.objectid,
649 trans->transid, ins.objectid); 654 trans->transid, ins.objectid);
650 655
651 BUG_ON(ret); 656 BUG_ON(ret);
652 } 657 }
658 path->leave_spinning = 0;
653 btrfs_release_path(root, path); 659 btrfs_release_path(root, path);
654 if (disk_bytenr != 0) 660 if (disk_bytenr != 0)
655 inode_add_bytes(inode, extent_end - end); 661 inode_add_bytes(inode, extent_end - end);
@@ -912,7 +918,7 @@ again:
912 btrfs_set_file_extent_other_encoding(leaf, fi, 0); 918 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
913 919
914 if (orig_parent != leaf->start) { 920 if (orig_parent != leaf->start) {
915 ret = btrfs_update_extent_ref(trans, root, bytenr, 921 ret = btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
916 orig_parent, leaf->start, 922 orig_parent, leaf->start,
917 root->root_key.objectid, 923 root->root_key.objectid,
918 trans->transid, inode->i_ino); 924 trans->transid, inode->i_ino);
@@ -1155,6 +1161,20 @@ out_nolock:
1155 page_cache_release(pinned[1]); 1161 page_cache_release(pinned[1]);
1156 *ppos = pos; 1162 *ppos = pos;
1157 1163
1164 /*
1165 * we want to make sure fsync finds this change
1166 * but we haven't joined a transaction running right now.
1167 *
1168 * Later on, someone is sure to update the inode and get the
1169 * real transid recorded.
1170 *
1171 * We set last_trans now to the fs_info generation + 1,
1172 * this will either be one more than the running transaction
1173 * or the generation used for the next transaction if there isn't
1174 * one running right now.
1175 */
1176 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1177
1158 if (num_written > 0 && will_write) { 1178 if (num_written > 0 && will_write) {
1159 struct btrfs_trans_handle *trans; 1179 struct btrfs_trans_handle *trans;
1160 1180
@@ -1167,8 +1187,11 @@ out_nolock:
1167 ret = btrfs_log_dentry_safe(trans, root, 1187 ret = btrfs_log_dentry_safe(trans, root,
1168 file->f_dentry); 1188 file->f_dentry);
1169 if (ret == 0) { 1189 if (ret == 0) {
1170 btrfs_sync_log(trans, root); 1190 ret = btrfs_sync_log(trans, root);
1171 btrfs_end_transaction(trans, root); 1191 if (ret == 0)
1192 btrfs_end_transaction(trans, root);
1193 else
1194 btrfs_commit_transaction(trans, root);
1172 } else { 1195 } else {
1173 btrfs_commit_transaction(trans, root); 1196 btrfs_commit_transaction(trans, root);
1174 } 1197 }
@@ -1185,6 +1208,18 @@ out_nolock:
1185 1208
1186int btrfs_release_file(struct inode *inode, struct file *filp) 1209int btrfs_release_file(struct inode *inode, struct file *filp)
1187{ 1210{
1211 /*
1212 * ordered_data_close is set by settattr when we are about to truncate
1213 * a file from a non-zero size to a zero size. This tries to
1214 * flush down new bytes that may have been written if the
1215 * application were using truncate to replace a file in place.
1216 */
1217 if (BTRFS_I(inode)->ordered_data_close) {
1218 BTRFS_I(inode)->ordered_data_close = 0;
1219 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1220 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1221 filemap_flush(inode->i_mapping);
1222 }
1188 if (filp->private_data) 1223 if (filp->private_data)
1189 btrfs_ioctl_trans_end(filp); 1224 btrfs_ioctl_trans_end(filp);
1190 return 0; 1225 return 0;
@@ -1260,8 +1295,11 @@ int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1260 if (ret > 0) { 1295 if (ret > 0) {
1261 ret = btrfs_commit_transaction(trans, root); 1296 ret = btrfs_commit_transaction(trans, root);
1262 } else { 1297 } else {
1263 btrfs_sync_log(trans, root); 1298 ret = btrfs_sync_log(trans, root);
1264 ret = btrfs_end_transaction(trans, root); 1299 if (ret == 0)
1300 ret = btrfs_end_transaction(trans, root);
1301 else
1302 ret = btrfs_commit_transaction(trans, root);
1265 } 1303 }
1266 mutex_lock(&dentry->d_inode->i_mutex); 1304 mutex_lock(&dentry->d_inode->i_mutex);
1267out: 1305out:
diff --git a/fs/btrfs/inode-item.c b/fs/btrfs/inode-item.c
index 3d46fa1f29a4..6b627c611808 100644
--- a/fs/btrfs/inode-item.c
+++ b/fs/btrfs/inode-item.c
@@ -73,6 +73,8 @@ int btrfs_del_inode_ref(struct btrfs_trans_handle *trans,
73 if (!path) 73 if (!path)
74 return -ENOMEM; 74 return -ENOMEM;
75 75
76 path->leave_spinning = 1;
77
76 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 78 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
77 if (ret > 0) { 79 if (ret > 0) {
78 ret = -ENOENT; 80 ret = -ENOENT;
@@ -127,6 +129,7 @@ int btrfs_insert_inode_ref(struct btrfs_trans_handle *trans,
127 if (!path) 129 if (!path)
128 return -ENOMEM; 130 return -ENOMEM;
129 131
132 path->leave_spinning = 1;
130 ret = btrfs_insert_empty_item(trans, root, path, &key, 133 ret = btrfs_insert_empty_item(trans, root, path, &key,
131 ins_len); 134 ins_len);
132 if (ret == -EEXIST) { 135 if (ret == -EEXIST) {
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 17e608c4dc70..06d8db5afb08 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -134,6 +134,7 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
134 if (!path) 134 if (!path)
135 return -ENOMEM; 135 return -ENOMEM;
136 136
137 path->leave_spinning = 1;
137 btrfs_set_trans_block_group(trans, inode); 138 btrfs_set_trans_block_group(trans, inode);
138 139
139 key.objectid = inode->i_ino; 140 key.objectid = inode->i_ino;
@@ -167,9 +168,9 @@ static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
167 cur_size = min_t(unsigned long, compressed_size, 168 cur_size = min_t(unsigned long, compressed_size,
168 PAGE_CACHE_SIZE); 169 PAGE_CACHE_SIZE);
169 170
170 kaddr = kmap(cpage); 171 kaddr = kmap_atomic(cpage, KM_USER0);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size); 172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap(cpage); 173 kunmap_atomic(kaddr, KM_USER0);
173 174
174 i++; 175 i++;
175 ptr += cur_size; 176 ptr += cur_size;
@@ -204,7 +205,7 @@ fail:
204 * does the checks required to make sure the data is small enough 205 * does the checks required to make sure the data is small enough
205 * to fit as an inline extent. 206 * to fit as an inline extent.
206 */ 207 */
207static int cow_file_range_inline(struct btrfs_trans_handle *trans, 208static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
208 struct btrfs_root *root, 209 struct btrfs_root *root,
209 struct inode *inode, u64 start, u64 end, 210 struct inode *inode, u64 start, u64 end,
210 size_t compressed_size, 211 size_t compressed_size,
@@ -854,11 +855,6 @@ static int cow_file_range_async(struct inode *inode, struct page *locked_page,
854 u64 cur_end; 855 u64 cur_end;
855 int limit = 10 * 1024 * 1042; 856 int limit = 10 * 1024 * 1042;
856 857
857 if (!btrfs_test_opt(root, COMPRESS)) {
858 return cow_file_range(inode, locked_page, start, end,
859 page_started, nr_written, 1);
860 }
861
862 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED | 858 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
863 EXTENT_DELALLOC, 1, 0, GFP_NOFS); 859 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
864 while (start < end) { 860 while (start < end) {
@@ -935,7 +931,8 @@ static noinline int csum_exist_in_range(struct btrfs_root *root,
935 * If no cow copies or snapshots exist, we write directly to the existing 931 * If no cow copies or snapshots exist, we write directly to the existing
936 * blocks on disk 932 * blocks on disk
937 */ 933 */
938static int run_delalloc_nocow(struct inode *inode, struct page *locked_page, 934static noinline int run_delalloc_nocow(struct inode *inode,
935 struct page *locked_page,
939 u64 start, u64 end, int *page_started, int force, 936 u64 start, u64 end, int *page_started, int force,
940 unsigned long *nr_written) 937 unsigned long *nr_written)
941{ 938{
@@ -1133,6 +1130,7 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1133 unsigned long *nr_written) 1130 unsigned long *nr_written)
1134{ 1131{
1135 int ret; 1132 int ret;
1133 struct btrfs_root *root = BTRFS_I(inode)->root;
1136 1134
1137 if (btrfs_test_flag(inode, NODATACOW)) 1135 if (btrfs_test_flag(inode, NODATACOW))
1138 ret = run_delalloc_nocow(inode, locked_page, start, end, 1136 ret = run_delalloc_nocow(inode, locked_page, start, end,
@@ -1140,10 +1138,12 @@ static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1140 else if (btrfs_test_flag(inode, PREALLOC)) 1138 else if (btrfs_test_flag(inode, PREALLOC))
1141 ret = run_delalloc_nocow(inode, locked_page, start, end, 1139 ret = run_delalloc_nocow(inode, locked_page, start, end,
1142 page_started, 0, nr_written); 1140 page_started, 0, nr_written);
1141 else if (!btrfs_test_opt(root, COMPRESS))
1142 ret = cow_file_range(inode, locked_page, start, end,
1143 page_started, nr_written, 1);
1143 else 1144 else
1144 ret = cow_file_range_async(inode, locked_page, start, end, 1145 ret = cow_file_range_async(inode, locked_page, start, end,
1145 page_started, nr_written); 1146 page_started, nr_written);
1146
1147 return ret; 1147 return ret;
1148} 1148}
1149 1149
@@ -1453,6 +1453,7 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1453 path = btrfs_alloc_path(); 1453 path = btrfs_alloc_path();
1454 BUG_ON(!path); 1454 BUG_ON(!path);
1455 1455
1456 path->leave_spinning = 1;
1456 ret = btrfs_drop_extents(trans, root, inode, file_pos, 1457 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1457 file_pos + num_bytes, file_pos, &hint); 1458 file_pos + num_bytes, file_pos, &hint);
1458 BUG_ON(ret); 1459 BUG_ON(ret);
@@ -1475,6 +1476,10 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1475 btrfs_set_file_extent_compression(leaf, fi, compression); 1476 btrfs_set_file_extent_compression(leaf, fi, compression);
1476 btrfs_set_file_extent_encryption(leaf, fi, encryption); 1477 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1477 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding); 1478 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1479
1480 btrfs_unlock_up_safe(path, 1);
1481 btrfs_set_lock_blocking(leaf);
1482
1478 btrfs_mark_buffer_dirty(leaf); 1483 btrfs_mark_buffer_dirty(leaf);
1479 1484
1480 inode_add_bytes(inode, num_bytes); 1485 inode_add_bytes(inode, num_bytes);
@@ -1487,11 +1492,35 @@ static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1487 root->root_key.objectid, 1492 root->root_key.objectid,
1488 trans->transid, inode->i_ino, &ins); 1493 trans->transid, inode->i_ino, &ins);
1489 BUG_ON(ret); 1494 BUG_ON(ret);
1490
1491 btrfs_free_path(path); 1495 btrfs_free_path(path);
1496
1492 return 0; 1497 return 0;
1493} 1498}
1494 1499
1500/*
1501 * helper function for btrfs_finish_ordered_io, this
1502 * just reads in some of the csum leaves to prime them into ram
1503 * before we start the transaction. It limits the amount of btree
1504 * reads required while inside the transaction.
1505 */
1506static noinline void reada_csum(struct btrfs_root *root,
1507 struct btrfs_path *path,
1508 struct btrfs_ordered_extent *ordered_extent)
1509{
1510 struct btrfs_ordered_sum *sum;
1511 u64 bytenr;
1512
1513 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1514 list);
1515 bytenr = sum->sums[0].bytenr;
1516
1517 /*
1518 * we don't care about the results, the point of this search is
1519 * just to get the btree leaves into ram
1520 */
1521 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1522}
1523
1495/* as ordered data IO finishes, this gets called so we can finish 1524/* as ordered data IO finishes, this gets called so we can finish
1496 * an ordered extent if the range of bytes in the file it covers are 1525 * an ordered extent if the range of bytes in the file it covers are
1497 * fully written. 1526 * fully written.
@@ -1500,8 +1529,9 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1500{ 1529{
1501 struct btrfs_root *root = BTRFS_I(inode)->root; 1530 struct btrfs_root *root = BTRFS_I(inode)->root;
1502 struct btrfs_trans_handle *trans; 1531 struct btrfs_trans_handle *trans;
1503 struct btrfs_ordered_extent *ordered_extent; 1532 struct btrfs_ordered_extent *ordered_extent = NULL;
1504 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 1533 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1534 struct btrfs_path *path;
1505 int compressed = 0; 1535 int compressed = 0;
1506 int ret; 1536 int ret;
1507 1537
@@ -1509,9 +1539,33 @@ static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1509 if (!ret) 1539 if (!ret)
1510 return 0; 1540 return 0;
1511 1541
1542 /*
1543 * before we join the transaction, try to do some of our IO.
1544 * This will limit the amount of IO that we have to do with
1545 * the transaction running. We're unlikely to need to do any
1546 * IO if the file extents are new, the disk_i_size checks
1547 * covers the most common case.
1548 */
1549 if (start < BTRFS_I(inode)->disk_i_size) {
1550 path = btrfs_alloc_path();
1551 if (path) {
1552 ret = btrfs_lookup_file_extent(NULL, root, path,
1553 inode->i_ino,
1554 start, 0);
1555 ordered_extent = btrfs_lookup_ordered_extent(inode,
1556 start);
1557 if (!list_empty(&ordered_extent->list)) {
1558 btrfs_release_path(root, path);
1559 reada_csum(root, path, ordered_extent);
1560 }
1561 btrfs_free_path(path);
1562 }
1563 }
1564
1512 trans = btrfs_join_transaction(root, 1); 1565 trans = btrfs_join_transaction(root, 1);
1513 1566
1514 ordered_extent = btrfs_lookup_ordered_extent(inode, start); 1567 if (!ordered_extent)
1568 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1515 BUG_ON(!ordered_extent); 1569 BUG_ON(!ordered_extent);
1516 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) 1570 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1517 goto nocow; 1571 goto nocow;
@@ -2101,6 +2155,7 @@ noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2101 2155
2102 path = btrfs_alloc_path(); 2156 path = btrfs_alloc_path();
2103 BUG_ON(!path); 2157 BUG_ON(!path);
2158 path->leave_spinning = 1;
2104 ret = btrfs_lookup_inode(trans, root, path, 2159 ret = btrfs_lookup_inode(trans, root, path,
2105 &BTRFS_I(inode)->location, 1); 2160 &BTRFS_I(inode)->location, 1);
2106 if (ret) { 2161 if (ret) {
@@ -2147,6 +2202,7 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2147 goto err; 2202 goto err;
2148 } 2203 }
2149 2204
2205 path->leave_spinning = 1;
2150 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino, 2206 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2151 name, name_len, -1); 2207 name, name_len, -1);
2152 if (IS_ERR(di)) { 2208 if (IS_ERR(di)) {
@@ -2190,8 +2246,6 @@ int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2190 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, 2246 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2191 inode, dir->i_ino); 2247 inode, dir->i_ino);
2192 BUG_ON(ret != 0 && ret != -ENOENT); 2248 BUG_ON(ret != 0 && ret != -ENOENT);
2193 if (ret != -ENOENT)
2194 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2195 2249
2196 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, 2250 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2197 dir, index); 2251 dir, index);
@@ -2224,6 +2278,9 @@ static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2224 trans = btrfs_start_transaction(root, 1); 2278 trans = btrfs_start_transaction(root, 1);
2225 2279
2226 btrfs_set_trans_block_group(trans, dir); 2280 btrfs_set_trans_block_group(trans, dir);
2281
2282 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2283
2227 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode, 2284 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2228 dentry->d_name.name, dentry->d_name.len); 2285 dentry->d_name.name, dentry->d_name.len);
2229 2286
@@ -2498,6 +2555,7 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2498 key.type = (u8)-1; 2555 key.type = (u8)-1;
2499 2556
2500search_again: 2557search_again:
2558 path->leave_spinning = 1;
2501 ret = btrfs_search_slot(trans, root, &key, path, -1, 1); 2559 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2502 if (ret < 0) 2560 if (ret < 0)
2503 goto error; 2561 goto error;
@@ -2644,6 +2702,7 @@ delete:
2644 break; 2702 break;
2645 } 2703 }
2646 if (found_extent) { 2704 if (found_extent) {
2705 btrfs_set_path_blocking(path);
2647 ret = btrfs_free_extent(trans, root, extent_start, 2706 ret = btrfs_free_extent(trans, root, extent_start,
2648 extent_num_bytes, 2707 extent_num_bytes,
2649 leaf->start, root_owner, 2708 leaf->start, root_owner,
@@ -2848,11 +2907,21 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2848 if (err) 2907 if (err)
2849 return err; 2908 return err;
2850 2909
2851 if (S_ISREG(inode->i_mode) && 2910 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
2852 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) { 2911 if (attr->ia_size > inode->i_size) {
2853 err = btrfs_cont_expand(inode, attr->ia_size); 2912 err = btrfs_cont_expand(inode, attr->ia_size);
2854 if (err) 2913 if (err)
2855 return err; 2914 return err;
2915 } else if (inode->i_size > 0 &&
2916 attr->ia_size == 0) {
2917
2918 /* we're truncating a file that used to have good
2919 * data down to zero. Make sure it gets into
2920 * the ordered flush list so that any new writes
2921 * get down to disk quickly.
2922 */
2923 BTRFS_I(inode)->ordered_data_close = 1;
2924 }
2856 } 2925 }
2857 2926
2858 err = inode_setattr(inode, attr); 2927 err = inode_setattr(inode, attr);
@@ -2984,13 +3053,14 @@ static noinline void init_btrfs_i(struct inode *inode)
2984 bi->disk_i_size = 0; 3053 bi->disk_i_size = 0;
2985 bi->flags = 0; 3054 bi->flags = 0;
2986 bi->index_cnt = (u64)-1; 3055 bi->index_cnt = (u64)-1;
2987 bi->log_dirty_trans = 0; 3056 bi->last_unlink_trans = 0;
2988 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS); 3057 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2989 extent_io_tree_init(&BTRFS_I(inode)->io_tree, 3058 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2990 inode->i_mapping, GFP_NOFS); 3059 inode->i_mapping, GFP_NOFS);
2991 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree, 3060 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2992 inode->i_mapping, GFP_NOFS); 3061 inode->i_mapping, GFP_NOFS);
2993 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes); 3062 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3063 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
2994 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree); 3064 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2995 mutex_init(&BTRFS_I(inode)->extent_mutex); 3065 mutex_init(&BTRFS_I(inode)->extent_mutex);
2996 mutex_init(&BTRFS_I(inode)->log_mutex); 3066 mutex_init(&BTRFS_I(inode)->log_mutex);
@@ -3449,6 +3519,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3449 sizes[0] = sizeof(struct btrfs_inode_item); 3519 sizes[0] = sizeof(struct btrfs_inode_item);
3450 sizes[1] = name_len + sizeof(*ref); 3520 sizes[1] = name_len + sizeof(*ref);
3451 3521
3522 path->leave_spinning = 1;
3452 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2); 3523 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3453 if (ret != 0) 3524 if (ret != 0)
3454 goto fail; 3525 goto fail;
@@ -3727,6 +3798,8 @@ static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3727 drop_inode = 1; 3798 drop_inode = 1;
3728 3799
3729 nr = trans->blocks_used; 3800 nr = trans->blocks_used;
3801
3802 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
3730 btrfs_end_transaction_throttle(trans, root); 3803 btrfs_end_transaction_throttle(trans, root);
3731fail: 3804fail:
3732 if (drop_inode) { 3805 if (drop_inode) {
@@ -4363,6 +4436,8 @@ again:
4363 } 4436 }
4364 ClearPageChecked(page); 4437 ClearPageChecked(page);
4365 set_page_dirty(page); 4438 set_page_dirty(page);
4439
4440 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
4366 unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 4441 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4367 4442
4368out_unlock: 4443out_unlock:
@@ -4388,6 +4463,27 @@ static void btrfs_truncate(struct inode *inode)
4388 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1); 4463 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4389 4464
4390 trans = btrfs_start_transaction(root, 1); 4465 trans = btrfs_start_transaction(root, 1);
4466
4467 /*
4468 * setattr is responsible for setting the ordered_data_close flag,
4469 * but that is only tested during the last file release. That
4470 * could happen well after the next commit, leaving a great big
4471 * window where new writes may get lost if someone chooses to write
4472 * to this file after truncating to zero
4473 *
4474 * The inode doesn't have any dirty data here, and so if we commit
4475 * this is a noop. If someone immediately starts writing to the inode
4476 * it is very likely we'll catch some of their writes in this
4477 * transaction, and the commit will find this file on the ordered
4478 * data list with good things to send down.
4479 *
4480 * This is a best effort solution, there is still a window where
4481 * using truncate to replace the contents of the file will
4482 * end up with a zero length file after a crash.
4483 */
4484 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
4485 btrfs_add_ordered_operation(trans, root, inode);
4486
4391 btrfs_set_trans_block_group(trans, inode); 4487 btrfs_set_trans_block_group(trans, inode);
4392 btrfs_i_size_write(inode, inode->i_size); 4488 btrfs_i_size_write(inode, inode->i_size);
4393 4489
@@ -4464,12 +4560,15 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
4464 ei->i_acl = BTRFS_ACL_NOT_CACHED; 4560 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4465 ei->i_default_acl = BTRFS_ACL_NOT_CACHED; 4561 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4466 INIT_LIST_HEAD(&ei->i_orphan); 4562 INIT_LIST_HEAD(&ei->i_orphan);
4563 INIT_LIST_HEAD(&ei->ordered_operations);
4467 return &ei->vfs_inode; 4564 return &ei->vfs_inode;
4468} 4565}
4469 4566
4470void btrfs_destroy_inode(struct inode *inode) 4567void btrfs_destroy_inode(struct inode *inode)
4471{ 4568{
4472 struct btrfs_ordered_extent *ordered; 4569 struct btrfs_ordered_extent *ordered;
4570 struct btrfs_root *root = BTRFS_I(inode)->root;
4571
4473 WARN_ON(!list_empty(&inode->i_dentry)); 4572 WARN_ON(!list_empty(&inode->i_dentry));
4474 WARN_ON(inode->i_data.nrpages); 4573 WARN_ON(inode->i_data.nrpages);
4475 4574
@@ -4480,13 +4579,24 @@ void btrfs_destroy_inode(struct inode *inode)
4480 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED) 4579 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4481 posix_acl_release(BTRFS_I(inode)->i_default_acl); 4580 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4482 4581
4483 spin_lock(&BTRFS_I(inode)->root->list_lock); 4582 /*
4583 * Make sure we're properly removed from the ordered operation
4584 * lists.
4585 */
4586 smp_mb();
4587 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
4588 spin_lock(&root->fs_info->ordered_extent_lock);
4589 list_del_init(&BTRFS_I(inode)->ordered_operations);
4590 spin_unlock(&root->fs_info->ordered_extent_lock);
4591 }
4592
4593 spin_lock(&root->list_lock);
4484 if (!list_empty(&BTRFS_I(inode)->i_orphan)) { 4594 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4485 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan" 4595 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4486 " list\n", inode->i_ino); 4596 " list\n", inode->i_ino);
4487 dump_stack(); 4597 dump_stack();
4488 } 4598 }
4489 spin_unlock(&BTRFS_I(inode)->root->list_lock); 4599 spin_unlock(&root->list_lock);
4490 4600
4491 while (1) { 4601 while (1) {
4492 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1); 4602 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
@@ -4611,8 +4721,36 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4611 if (ret) 4721 if (ret)
4612 goto out_unlock; 4722 goto out_unlock;
4613 4723
4724 /*
4725 * we're using rename to replace one file with another.
4726 * and the replacement file is large. Start IO on it now so
4727 * we don't add too much work to the end of the transaction
4728 */
4729 if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
4730 new_inode->i_size &&
4731 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
4732 filemap_flush(old_inode->i_mapping);
4733
4614 trans = btrfs_start_transaction(root, 1); 4734 trans = btrfs_start_transaction(root, 1);
4615 4735
4736 /*
4737 * make sure the inode gets flushed if it is replacing
4738 * something.
4739 */
4740 if (new_inode && new_inode->i_size &&
4741 old_inode && S_ISREG(old_inode->i_mode)) {
4742 btrfs_add_ordered_operation(trans, root, old_inode);
4743 }
4744
4745 /*
4746 * this is an ugly little race, but the rename is required to make
4747 * sure that if we crash, the inode is either at the old name
4748 * or the new one. pinning the log transaction lets us make sure
4749 * we don't allow a log commit to come in after we unlink the
4750 * name but before we add the new name back in.
4751 */
4752 btrfs_pin_log_trans(root);
4753
4616 btrfs_set_trans_block_group(trans, new_dir); 4754 btrfs_set_trans_block_group(trans, new_dir);
4617 4755
4618 btrfs_inc_nlink(old_dentry->d_inode); 4756 btrfs_inc_nlink(old_dentry->d_inode);
@@ -4620,6 +4758,9 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4620 new_dir->i_ctime = new_dir->i_mtime = ctime; 4758 new_dir->i_ctime = new_dir->i_mtime = ctime;
4621 old_inode->i_ctime = ctime; 4759 old_inode->i_ctime = ctime;
4622 4760
4761 if (old_dentry->d_parent != new_dentry->d_parent)
4762 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
4763
4623 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode, 4764 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4624 old_dentry->d_name.name, 4765 old_dentry->d_name.name,
4625 old_dentry->d_name.len); 4766 old_dentry->d_name.len);
@@ -4651,7 +4792,14 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4651 if (ret) 4792 if (ret)
4652 goto out_fail; 4793 goto out_fail;
4653 4794
4795 btrfs_log_new_name(trans, old_inode, old_dir,
4796 new_dentry->d_parent);
4654out_fail: 4797out_fail:
4798
4799 /* this btrfs_end_log_trans just allows the current
4800 * log-sub transaction to complete
4801 */
4802 btrfs_end_log_trans(root);
4655 btrfs_end_transaction_throttle(trans, root); 4803 btrfs_end_transaction_throttle(trans, root);
4656out_unlock: 4804out_unlock:
4657 return ret; 4805 return ret;
diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index 47b0a88c12a2..a5310c0f41e2 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -71,12 +71,13 @@ void btrfs_clear_lock_blocking(struct extent_buffer *eb)
71static int btrfs_spin_on_block(struct extent_buffer *eb) 71static int btrfs_spin_on_block(struct extent_buffer *eb)
72{ 72{
73 int i; 73 int i;
74
74 for (i = 0; i < 512; i++) { 75 for (i = 0; i < 512; i++) {
75 cpu_relax();
76 if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) 76 if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
77 return 1; 77 return 1;
78 if (need_resched()) 78 if (need_resched())
79 break; 79 break;
80 cpu_relax();
80 } 81 }
81 return 0; 82 return 0;
82} 83}
@@ -95,13 +96,15 @@ int btrfs_try_spin_lock(struct extent_buffer *eb)
95{ 96{
96 int i; 97 int i;
97 98
98 spin_nested(eb); 99 if (btrfs_spin_on_block(eb)) {
99 if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags)) 100 spin_nested(eb);
100 return 1; 101 if (!test_bit(EXTENT_BUFFER_BLOCKING, &eb->bflags))
101 spin_unlock(&eb->lock); 102 return 1;
102 103 spin_unlock(&eb->lock);
104 }
103 /* spin for a bit on the BLOCKING flag */ 105 /* spin for a bit on the BLOCKING flag */
104 for (i = 0; i < 2; i++) { 106 for (i = 0; i < 2; i++) {
107 cpu_relax();
105 if (!btrfs_spin_on_block(eb)) 108 if (!btrfs_spin_on_block(eb))
106 break; 109 break;
107 110
@@ -148,6 +151,9 @@ int btrfs_tree_lock(struct extent_buffer *eb)
148 DEFINE_WAIT(wait); 151 DEFINE_WAIT(wait);
149 wait.func = btrfs_wake_function; 152 wait.func = btrfs_wake_function;
150 153
154 if (!btrfs_spin_on_block(eb))
155 goto sleep;
156
151 while(1) { 157 while(1) {
152 spin_nested(eb); 158 spin_nested(eb);
153 159
@@ -165,9 +171,10 @@ int btrfs_tree_lock(struct extent_buffer *eb)
165 * spin for a bit, and if the blocking flag goes away, 171 * spin for a bit, and if the blocking flag goes away,
166 * loop around 172 * loop around
167 */ 173 */
174 cpu_relax();
168 if (btrfs_spin_on_block(eb)) 175 if (btrfs_spin_on_block(eb))
169 continue; 176 continue;
170 177sleep:
171 prepare_to_wait_exclusive(&eb->lock_wq, &wait, 178 prepare_to_wait_exclusive(&eb->lock_wq, &wait,
172 TASK_UNINTERRUPTIBLE); 179 TASK_UNINTERRUPTIBLE);
173 180
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 77c2411a5f0f..53c87b197d70 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -310,6 +310,16 @@ int btrfs_remove_ordered_extent(struct inode *inode,
310 310
311 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); 311 spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
312 list_del_init(&entry->root_extent_list); 312 list_del_init(&entry->root_extent_list);
313
314 /*
315 * we have no more ordered extents for this inode and
316 * no dirty pages. We can safely remove it from the
317 * list of ordered extents
318 */
319 if (RB_EMPTY_ROOT(&tree->tree) &&
320 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
321 list_del_init(&BTRFS_I(inode)->ordered_operations);
322 }
313 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock); 323 spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
314 324
315 mutex_unlock(&tree->mutex); 325 mutex_unlock(&tree->mutex);
@@ -370,6 +380,68 @@ int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only)
370} 380}
371 381
372/* 382/*
383 * this is used during transaction commit to write all the inodes
384 * added to the ordered operation list. These files must be fully on
385 * disk before the transaction commits.
386 *
387 * we have two modes here, one is to just start the IO via filemap_flush
388 * and the other is to wait for all the io. When we wait, we have an
389 * extra check to make sure the ordered operation list really is empty
390 * before we return
391 */
392int btrfs_run_ordered_operations(struct btrfs_root *root, int wait)
393{
394 struct btrfs_inode *btrfs_inode;
395 struct inode *inode;
396 struct list_head splice;
397
398 INIT_LIST_HEAD(&splice);
399
400 mutex_lock(&root->fs_info->ordered_operations_mutex);
401 spin_lock(&root->fs_info->ordered_extent_lock);
402again:
403 list_splice_init(&root->fs_info->ordered_operations, &splice);
404
405 while (!list_empty(&splice)) {
406 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
407 ordered_operations);
408
409 inode = &btrfs_inode->vfs_inode;
410
411 list_del_init(&btrfs_inode->ordered_operations);
412
413 /*
414 * the inode may be getting freed (in sys_unlink path).
415 */
416 inode = igrab(inode);
417
418 if (!wait && inode) {
419 list_add_tail(&BTRFS_I(inode)->ordered_operations,
420 &root->fs_info->ordered_operations);
421 }
422 spin_unlock(&root->fs_info->ordered_extent_lock);
423
424 if (inode) {
425 if (wait)
426 btrfs_wait_ordered_range(inode, 0, (u64)-1);
427 else
428 filemap_flush(inode->i_mapping);
429 iput(inode);
430 }
431
432 cond_resched();
433 spin_lock(&root->fs_info->ordered_extent_lock);
434 }
435 if (wait && !list_empty(&root->fs_info->ordered_operations))
436 goto again;
437
438 spin_unlock(&root->fs_info->ordered_extent_lock);
439 mutex_unlock(&root->fs_info->ordered_operations_mutex);
440
441 return 0;
442}
443
444/*
373 * Used to start IO or wait for a given ordered extent to finish. 445 * Used to start IO or wait for a given ordered extent to finish.
374 * 446 *
375 * If wait is one, this effectively waits on page writeback for all the pages 447 * If wait is one, this effectively waits on page writeback for all the pages
@@ -726,3 +798,49 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
726 798
727 return ret; 799 return ret;
728} 800}
801
802/*
803 * add a given inode to the list of inodes that must be fully on
804 * disk before a transaction commit finishes.
805 *
806 * This basically gives us the ext3 style data=ordered mode, and it is mostly
807 * used to make sure renamed files are fully on disk.
808 *
809 * It is a noop if the inode is already fully on disk.
810 *
811 * If trans is not null, we'll do a friendly check for a transaction that
812 * is already flushing things and force the IO down ourselves.
813 */
814int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 struct inode *inode)
817{
818 u64 last_mod;
819
820 last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
821
822 /*
823 * if this file hasn't been changed since the last transaction
824 * commit, we can safely return without doing anything
825 */
826 if (last_mod < root->fs_info->last_trans_committed)
827 return 0;
828
829 /*
830 * the transaction is already committing. Just start the IO and
831 * don't bother with all of this list nonsense
832 */
833 if (trans && root->fs_info->running_transaction->blocked) {
834 btrfs_wait_ordered_range(inode, 0, (u64)-1);
835 return 0;
836 }
837
838 spin_lock(&root->fs_info->ordered_extent_lock);
839 if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
840 list_add_tail(&BTRFS_I(inode)->ordered_operations,
841 &root->fs_info->ordered_operations);
842 }
843 spin_unlock(&root->fs_info->ordered_extent_lock);
844
845 return 0;
846}
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index ab66d5e8d6d6..3d31c8827b01 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -155,4 +155,8 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
155int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start, 155int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
156 loff_t end, int sync_mode); 156 loff_t end, int sync_mode);
157int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only); 157int btrfs_wait_ordered_extents(struct btrfs_root *root, int nocow_only);
158int btrfs_run_ordered_operations(struct btrfs_root *root, int wait);
159int btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
160 struct btrfs_root *root,
161 struct inode *inode);
158#endif 162#endif
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 4112d53d4f4d..664782c6a2df 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -65,6 +65,15 @@ static noinline int join_transaction(struct btrfs_root *root)
65 cur_trans->use_count = 1; 65 cur_trans->use_count = 1;
66 cur_trans->commit_done = 0; 66 cur_trans->commit_done = 0;
67 cur_trans->start_time = get_seconds(); 67 cur_trans->start_time = get_seconds();
68
69 cur_trans->delayed_refs.root.rb_node = NULL;
70 cur_trans->delayed_refs.num_entries = 0;
71 cur_trans->delayed_refs.num_heads_ready = 0;
72 cur_trans->delayed_refs.num_heads = 0;
73 cur_trans->delayed_refs.flushing = 0;
74 cur_trans->delayed_refs.run_delayed_start = 0;
75 spin_lock_init(&cur_trans->delayed_refs.lock);
76
68 INIT_LIST_HEAD(&cur_trans->pending_snapshots); 77 INIT_LIST_HEAD(&cur_trans->pending_snapshots);
69 list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 78 list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
70 extent_io_tree_init(&cur_trans->dirty_pages, 79 extent_io_tree_init(&cur_trans->dirty_pages,
@@ -182,6 +191,8 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
182 h->block_group = 0; 191 h->block_group = 0;
183 h->alloc_exclude_nr = 0; 192 h->alloc_exclude_nr = 0;
184 h->alloc_exclude_start = 0; 193 h->alloc_exclude_start = 0;
194 h->delayed_ref_updates = 0;
195
185 root->fs_info->running_transaction->use_count++; 196 root->fs_info->running_transaction->use_count++;
186 mutex_unlock(&root->fs_info->trans_mutex); 197 mutex_unlock(&root->fs_info->trans_mutex);
187 return h; 198 return h;
@@ -271,7 +282,6 @@ void btrfs_throttle(struct btrfs_root *root)
271 if (!root->fs_info->open_ioctl_trans) 282 if (!root->fs_info->open_ioctl_trans)
272 wait_current_trans(root); 283 wait_current_trans(root);
273 mutex_unlock(&root->fs_info->trans_mutex); 284 mutex_unlock(&root->fs_info->trans_mutex);
274
275 throttle_on_drops(root); 285 throttle_on_drops(root);
276} 286}
277 287
@@ -280,6 +290,27 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
280{ 290{
281 struct btrfs_transaction *cur_trans; 291 struct btrfs_transaction *cur_trans;
282 struct btrfs_fs_info *info = root->fs_info; 292 struct btrfs_fs_info *info = root->fs_info;
293 int count = 0;
294
295 while (count < 4) {
296 unsigned long cur = trans->delayed_ref_updates;
297 trans->delayed_ref_updates = 0;
298 if (cur &&
299 trans->transaction->delayed_refs.num_heads_ready > 64) {
300 trans->delayed_ref_updates = 0;
301
302 /*
303 * do a full flush if the transaction is trying
304 * to close
305 */
306 if (trans->transaction->delayed_refs.flushing)
307 cur = 0;
308 btrfs_run_delayed_refs(trans, root, cur);
309 } else {
310 break;
311 }
312 count++;
313 }
283 314
284 mutex_lock(&info->trans_mutex); 315 mutex_lock(&info->trans_mutex);
285 cur_trans = info->running_transaction; 316 cur_trans = info->running_transaction;
@@ -424,9 +455,10 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
424 u64 old_root_bytenr; 455 u64 old_root_bytenr;
425 struct btrfs_root *tree_root = root->fs_info->tree_root; 456 struct btrfs_root *tree_root = root->fs_info->tree_root;
426 457
427 btrfs_extent_post_op(trans, root);
428 btrfs_write_dirty_block_groups(trans, root); 458 btrfs_write_dirty_block_groups(trans, root);
429 btrfs_extent_post_op(trans, root); 459
460 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
461 BUG_ON(ret);
430 462
431 while (1) { 463 while (1) {
432 old_root_bytenr = btrfs_root_bytenr(&root->root_item); 464 old_root_bytenr = btrfs_root_bytenr(&root->root_item);
@@ -438,14 +470,14 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
438 btrfs_header_level(root->node)); 470 btrfs_header_level(root->node));
439 btrfs_set_root_generation(&root->root_item, trans->transid); 471 btrfs_set_root_generation(&root->root_item, trans->transid);
440 472
441 btrfs_extent_post_op(trans, root);
442
443 ret = btrfs_update_root(trans, tree_root, 473 ret = btrfs_update_root(trans, tree_root,
444 &root->root_key, 474 &root->root_key,
445 &root->root_item); 475 &root->root_item);
446 BUG_ON(ret); 476 BUG_ON(ret);
447 btrfs_write_dirty_block_groups(trans, root); 477 btrfs_write_dirty_block_groups(trans, root);
448 btrfs_extent_post_op(trans, root); 478
479 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
480 BUG_ON(ret);
449 } 481 }
450 return 0; 482 return 0;
451} 483}
@@ -459,15 +491,18 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
459 struct btrfs_fs_info *fs_info = root->fs_info; 491 struct btrfs_fs_info *fs_info = root->fs_info;
460 struct list_head *next; 492 struct list_head *next;
461 struct extent_buffer *eb; 493 struct extent_buffer *eb;
494 int ret;
462 495
463 btrfs_extent_post_op(trans, fs_info->tree_root); 496 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
497 BUG_ON(ret);
464 498
465 eb = btrfs_lock_root_node(fs_info->tree_root); 499 eb = btrfs_lock_root_node(fs_info->tree_root);
466 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb, 0); 500 btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb);
467 btrfs_tree_unlock(eb); 501 btrfs_tree_unlock(eb);
468 free_extent_buffer(eb); 502 free_extent_buffer(eb);
469 503
470 btrfs_extent_post_op(trans, fs_info->tree_root); 504 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
505 BUG_ON(ret);
471 506
472 while (!list_empty(&fs_info->dirty_cowonly_roots)) { 507 while (!list_empty(&fs_info->dirty_cowonly_roots)) {
473 next = fs_info->dirty_cowonly_roots.next; 508 next = fs_info->dirty_cowonly_roots.next;
@@ -475,6 +510,9 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
475 root = list_entry(next, struct btrfs_root, dirty_list); 510 root = list_entry(next, struct btrfs_root, dirty_list);
476 511
477 update_cowonly_root(trans, root); 512 update_cowonly_root(trans, root);
513
514 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
515 BUG_ON(ret);
478 } 516 }
479 return 0; 517 return 0;
480} 518}
@@ -635,6 +673,31 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
635} 673}
636 674
637/* 675/*
676 * when dropping snapshots, we generate a ton of delayed refs, and it makes
677 * sense not to join the transaction while it is trying to flush the current
678 * queue of delayed refs out.
679 *
680 * This is used by the drop snapshot code only
681 */
682static noinline int wait_transaction_pre_flush(struct btrfs_fs_info *info)
683{
684 DEFINE_WAIT(wait);
685
686 mutex_lock(&info->trans_mutex);
687 while (info->running_transaction &&
688 info->running_transaction->delayed_refs.flushing) {
689 prepare_to_wait(&info->transaction_wait, &wait,
690 TASK_UNINTERRUPTIBLE);
691 mutex_unlock(&info->trans_mutex);
692 schedule();
693 mutex_lock(&info->trans_mutex);
694 finish_wait(&info->transaction_wait, &wait);
695 }
696 mutex_unlock(&info->trans_mutex);
697 return 0;
698}
699
700/*
638 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on 701 * Given a list of roots that need to be deleted, call btrfs_drop_snapshot on
639 * all of them 702 * all of them
640 */ 703 */
@@ -661,7 +724,22 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
661 atomic_inc(&root->fs_info->throttles); 724 atomic_inc(&root->fs_info->throttles);
662 725
663 while (1) { 726 while (1) {
727 /*
728 * we don't want to jump in and create a bunch of
729 * delayed refs if the transaction is starting to close
730 */
731 wait_transaction_pre_flush(tree_root->fs_info);
664 trans = btrfs_start_transaction(tree_root, 1); 732 trans = btrfs_start_transaction(tree_root, 1);
733
734 /*
735 * we've joined a transaction, make sure it isn't
736 * closing right now
737 */
738 if (trans->transaction->delayed_refs.flushing) {
739 btrfs_end_transaction(trans, tree_root);
740 continue;
741 }
742
665 mutex_lock(&root->fs_info->drop_mutex); 743 mutex_lock(&root->fs_info->drop_mutex);
666 ret = btrfs_drop_snapshot(trans, dirty->root); 744 ret = btrfs_drop_snapshot(trans, dirty->root);
667 if (ret != -EAGAIN) 745 if (ret != -EAGAIN)
@@ -766,7 +844,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
766 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 844 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
767 845
768 old = btrfs_lock_root_node(root); 846 old = btrfs_lock_root_node(root);
769 btrfs_cow_block(trans, root, old, NULL, 0, &old, 0); 847 btrfs_cow_block(trans, root, old, NULL, 0, &old);
770 848
771 btrfs_copy_root(trans, root, old, &tmp, objectid); 849 btrfs_copy_root(trans, root, old, &tmp, objectid);
772 btrfs_tree_unlock(old); 850 btrfs_tree_unlock(old);
@@ -894,12 +972,31 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
894 struct extent_io_tree *pinned_copy; 972 struct extent_io_tree *pinned_copy;
895 DEFINE_WAIT(wait); 973 DEFINE_WAIT(wait);
896 int ret; 974 int ret;
975 int should_grow = 0;
976 unsigned long now = get_seconds();
977
978 btrfs_run_ordered_operations(root, 0);
979
980 /* make a pass through all the delayed refs we have so far
981 * any runnings procs may add more while we are here
982 */
983 ret = btrfs_run_delayed_refs(trans, root, 0);
984 BUG_ON(ret);
985
986 cur_trans = trans->transaction;
987 /*
988 * set the flushing flag so procs in this transaction have to
989 * start sending their work down.
990 */
991 cur_trans->delayed_refs.flushing = 1;
992
993 ret = btrfs_run_delayed_refs(trans, root, 0);
994 BUG_ON(ret);
897 995
898 INIT_LIST_HEAD(&dirty_fs_roots);
899 mutex_lock(&root->fs_info->trans_mutex); 996 mutex_lock(&root->fs_info->trans_mutex);
900 if (trans->transaction->in_commit) { 997 INIT_LIST_HEAD(&dirty_fs_roots);
901 cur_trans = trans->transaction; 998 if (cur_trans->in_commit) {
902 trans->transaction->use_count++; 999 cur_trans->use_count++;
903 mutex_unlock(&root->fs_info->trans_mutex); 1000 mutex_unlock(&root->fs_info->trans_mutex);
904 btrfs_end_transaction(trans, root); 1001 btrfs_end_transaction(trans, root);
905 1002
@@ -922,7 +1019,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
922 1019
923 trans->transaction->in_commit = 1; 1020 trans->transaction->in_commit = 1;
924 trans->transaction->blocked = 1; 1021 trans->transaction->blocked = 1;
925 cur_trans = trans->transaction;
926 if (cur_trans->list.prev != &root->fs_info->trans_list) { 1022 if (cur_trans->list.prev != &root->fs_info->trans_list) {
927 prev_trans = list_entry(cur_trans->list.prev, 1023 prev_trans = list_entry(cur_trans->list.prev,
928 struct btrfs_transaction, list); 1024 struct btrfs_transaction, list);
@@ -937,6 +1033,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
937 } 1033 }
938 } 1034 }
939 1035
1036 if (now < cur_trans->start_time || now - cur_trans->start_time < 1)
1037 should_grow = 1;
1038
940 do { 1039 do {
941 int snap_pending = 0; 1040 int snap_pending = 0;
942 joined = cur_trans->num_joined; 1041 joined = cur_trans->num_joined;
@@ -949,7 +1048,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
949 1048
950 if (cur_trans->num_writers > 1) 1049 if (cur_trans->num_writers > 1)
951 timeout = MAX_SCHEDULE_TIMEOUT; 1050 timeout = MAX_SCHEDULE_TIMEOUT;
952 else 1051 else if (should_grow)
953 timeout = 1; 1052 timeout = 1;
954 1053
955 mutex_unlock(&root->fs_info->trans_mutex); 1054 mutex_unlock(&root->fs_info->trans_mutex);
@@ -959,16 +1058,30 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
959 BUG_ON(ret); 1058 BUG_ON(ret);
960 } 1059 }
961 1060
962 schedule_timeout(timeout); 1061 /*
1062 * rename don't use btrfs_join_transaction, so, once we
1063 * set the transaction to blocked above, we aren't going
1064 * to get any new ordered operations. We can safely run
1065 * it here and no for sure that nothing new will be added
1066 * to the list
1067 */
1068 btrfs_run_ordered_operations(root, 1);
1069
1070 smp_mb();
1071 if (cur_trans->num_writers > 1 || should_grow)
1072 schedule_timeout(timeout);
963 1073
964 mutex_lock(&root->fs_info->trans_mutex); 1074 mutex_lock(&root->fs_info->trans_mutex);
965 finish_wait(&cur_trans->writer_wait, &wait); 1075 finish_wait(&cur_trans->writer_wait, &wait);
966 } while (cur_trans->num_writers > 1 || 1076 } while (cur_trans->num_writers > 1 ||
967 (cur_trans->num_joined != joined)); 1077 (should_grow && cur_trans->num_joined != joined));
968 1078
969 ret = create_pending_snapshots(trans, root->fs_info); 1079 ret = create_pending_snapshots(trans, root->fs_info);
970 BUG_ON(ret); 1080 BUG_ON(ret);
971 1081
1082 ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
1083 BUG_ON(ret);
1084
972 WARN_ON(cur_trans != trans->transaction); 1085 WARN_ON(cur_trans != trans->transaction);
973 1086
974 /* btrfs_commit_tree_roots is responsible for getting the 1087 /* btrfs_commit_tree_roots is responsible for getting the
@@ -1032,6 +1145,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1032 btrfs_copy_pinned(root, pinned_copy); 1145 btrfs_copy_pinned(root, pinned_copy);
1033 1146
1034 trans->transaction->blocked = 0; 1147 trans->transaction->blocked = 0;
1148
1035 wake_up(&root->fs_info->transaction_throttle); 1149 wake_up(&root->fs_info->transaction_throttle);
1036 wake_up(&root->fs_info->transaction_wait); 1150 wake_up(&root->fs_info->transaction_wait);
1037 1151
@@ -1058,6 +1172,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
1058 mutex_lock(&root->fs_info->trans_mutex); 1172 mutex_lock(&root->fs_info->trans_mutex);
1059 1173
1060 cur_trans->commit_done = 1; 1174 cur_trans->commit_done = 1;
1175
1061 root->fs_info->last_trans_committed = cur_trans->transid; 1176 root->fs_info->last_trans_committed = cur_trans->transid;
1062 wake_up(&cur_trans->commit_wait); 1177 wake_up(&cur_trans->commit_wait);
1063 1178
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index ea292117f882..94f5bde2b58d 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -19,10 +19,16 @@
19#ifndef __BTRFS_TRANSACTION__ 19#ifndef __BTRFS_TRANSACTION__
20#define __BTRFS_TRANSACTION__ 20#define __BTRFS_TRANSACTION__
21#include "btrfs_inode.h" 21#include "btrfs_inode.h"
22#include "delayed-ref.h"
22 23
23struct btrfs_transaction { 24struct btrfs_transaction {
24 u64 transid; 25 u64 transid;
26 /*
27 * total writers in this transaction, it must be zero before the
28 * transaction can end
29 */
25 unsigned long num_writers; 30 unsigned long num_writers;
31
26 unsigned long num_joined; 32 unsigned long num_joined;
27 int in_commit; 33 int in_commit;
28 int use_count; 34 int use_count;
@@ -34,6 +40,7 @@ struct btrfs_transaction {
34 wait_queue_head_t writer_wait; 40 wait_queue_head_t writer_wait;
35 wait_queue_head_t commit_wait; 41 wait_queue_head_t commit_wait;
36 struct list_head pending_snapshots; 42 struct list_head pending_snapshots;
43 struct btrfs_delayed_ref_root delayed_refs;
37}; 44};
38 45
39struct btrfs_trans_handle { 46struct btrfs_trans_handle {
@@ -44,6 +51,7 @@ struct btrfs_trans_handle {
44 u64 block_group; 51 u64 block_group;
45 u64 alloc_exclude_start; 52 u64 alloc_exclude_start;
46 u64 alloc_exclude_nr; 53 u64 alloc_exclude_nr;
54 unsigned long delayed_ref_updates;
47}; 55};
48 56
49struct btrfs_pending_snapshot { 57struct btrfs_pending_snapshot {
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index 98d25fa4570e..b10eacdb1620 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -124,8 +124,6 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
124 } 124 }
125 125
126 btrfs_release_path(root, path); 126 btrfs_release_path(root, path);
127 if (is_extent)
128 btrfs_extent_post_op(trans, root);
129out: 127out:
130 if (path) 128 if (path)
131 btrfs_free_path(path); 129 btrfs_free_path(path);
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index 9c462fbd60fa..fc9b87a7975b 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -35,6 +35,49 @@
35#define LOG_INODE_EXISTS 1 35#define LOG_INODE_EXISTS 1
36 36
37/* 37/*
38 * directory trouble cases
39 *
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
44 *
45 * mkdir foo/some_dir
46 * normal commit
47 * rename foo/some_dir foo2/some_dir
48 * mkdir foo/some_dir
49 * fsync foo/some_dir/some_file
50 *
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
54 *
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
57 *
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
61 *
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
64 *
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
67 *
68 * mkdir f1/foo
69 * normal commit
70 * rm -rf f1/foo
71 * fsync(f1)
72 *
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
77 * ugly details.
78 */
79
80/*
38 * stages for the tree walking. The first 81 * stages for the tree walking. The first
39 * stage (0) is to only pin down the blocks we find 82 * stage (0) is to only pin down the blocks we find
40 * the second stage (1) is to make sure that all the inodes 83 * the second stage (1) is to make sure that all the inodes
@@ -47,12 +90,17 @@
47#define LOG_WALK_REPLAY_INODES 1 90#define LOG_WALK_REPLAY_INODES 1
48#define LOG_WALK_REPLAY_ALL 2 91#define LOG_WALK_REPLAY_ALL 2
49 92
50static int __btrfs_log_inode(struct btrfs_trans_handle *trans, 93static int btrfs_log_inode(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root, struct inode *inode, 94 struct btrfs_root *root, struct inode *inode,
52 int inode_only); 95 int inode_only);
53static int link_to_fixup_dir(struct btrfs_trans_handle *trans, 96static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
54 struct btrfs_root *root, 97 struct btrfs_root *root,
55 struct btrfs_path *path, u64 objectid); 98 struct btrfs_path *path, u64 objectid);
99static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root,
101 struct btrfs_root *log,
102 struct btrfs_path *path,
103 u64 dirid, int del_all);
56 104
57/* 105/*
58 * tree logging is a special write ahead log used to make sure that 106 * tree logging is a special write ahead log used to make sure that
@@ -133,10 +181,25 @@ static int join_running_log_trans(struct btrfs_root *root)
133} 181}
134 182
135/* 183/*
184 * This either makes the current running log transaction wait
185 * until you call btrfs_end_log_trans() or it makes any future
186 * log transactions wait until you call btrfs_end_log_trans()
187 */
188int btrfs_pin_log_trans(struct btrfs_root *root)
189{
190 int ret = -ENOENT;
191
192 mutex_lock(&root->log_mutex);
193 atomic_inc(&root->log_writers);
194 mutex_unlock(&root->log_mutex);
195 return ret;
196}
197
198/*
136 * indicate we're done making changes to the log tree 199 * indicate we're done making changes to the log tree
137 * and wake up anyone waiting to do a sync 200 * and wake up anyone waiting to do a sync
138 */ 201 */
139static int end_log_trans(struct btrfs_root *root) 202int btrfs_end_log_trans(struct btrfs_root *root)
140{ 203{
141 if (atomic_dec_and_test(&root->log_writers)) { 204 if (atomic_dec_and_test(&root->log_writers)) {
142 smp_mb(); 205 smp_mb();
@@ -203,7 +266,6 @@ static int process_one_buffer(struct btrfs_root *log,
203 mutex_lock(&log->fs_info->pinned_mutex); 266 mutex_lock(&log->fs_info->pinned_mutex);
204 btrfs_update_pinned_extents(log->fs_info->extent_root, 267 btrfs_update_pinned_extents(log->fs_info->extent_root,
205 eb->start, eb->len, 1); 268 eb->start, eb->len, 1);
206 mutex_unlock(&log->fs_info->pinned_mutex);
207 } 269 }
208 270
209 if (btrfs_buffer_uptodate(eb, gen)) { 271 if (btrfs_buffer_uptodate(eb, gen)) {
@@ -603,6 +665,7 @@ static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
603 665
604 ret = link_to_fixup_dir(trans, root, path, location.objectid); 666 ret = link_to_fixup_dir(trans, root, path, location.objectid);
605 BUG_ON(ret); 667 BUG_ON(ret);
668
606 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len); 669 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
607 BUG_ON(ret); 670 BUG_ON(ret);
608 kfree(name); 671 kfree(name);
@@ -804,6 +867,7 @@ conflict_again:
804 victim_name_len)) { 867 victim_name_len)) {
805 btrfs_inc_nlink(inode); 868 btrfs_inc_nlink(inode);
806 btrfs_release_path(root, path); 869 btrfs_release_path(root, path);
870
807 ret = btrfs_unlink_inode(trans, root, dir, 871 ret = btrfs_unlink_inode(trans, root, dir,
808 inode, victim_name, 872 inode, victim_name,
809 victim_name_len); 873 victim_name_len);
@@ -922,13 +986,20 @@ static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
922 key.offset--; 986 key.offset--;
923 btrfs_release_path(root, path); 987 btrfs_release_path(root, path);
924 } 988 }
925 btrfs_free_path(path); 989 btrfs_release_path(root, path);
926 if (nlink != inode->i_nlink) { 990 if (nlink != inode->i_nlink) {
927 inode->i_nlink = nlink; 991 inode->i_nlink = nlink;
928 btrfs_update_inode(trans, root, inode); 992 btrfs_update_inode(trans, root, inode);
929 } 993 }
930 BTRFS_I(inode)->index_cnt = (u64)-1; 994 BTRFS_I(inode)->index_cnt = (u64)-1;
931 995
996 if (inode->i_nlink == 0 && S_ISDIR(inode->i_mode)) {
997 ret = replay_dir_deletes(trans, root, NULL, path,
998 inode->i_ino, 1);
999 BUG_ON(ret);
1000 }
1001 btrfs_free_path(path);
1002
932 return 0; 1003 return 0;
933} 1004}
934 1005
@@ -971,9 +1042,12 @@ static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
971 1042
972 iput(inode); 1043 iput(inode);
973 1044
974 if (key.offset == 0) 1045 /*
975 break; 1046 * fixup on a directory may create new entries,
976 key.offset--; 1047 * make sure we always look for the highset possible
1048 * offset
1049 */
1050 key.offset = (u64)-1;
977 } 1051 }
978 btrfs_release_path(root, path); 1052 btrfs_release_path(root, path);
979 return 0; 1053 return 0;
@@ -1313,11 +1387,11 @@ again:
1313 read_extent_buffer(eb, name, (unsigned long)(di + 1), 1387 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1314 name_len); 1388 name_len);
1315 log_di = NULL; 1389 log_di = NULL;
1316 if (dir_key->type == BTRFS_DIR_ITEM_KEY) { 1390 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1317 log_di = btrfs_lookup_dir_item(trans, log, log_path, 1391 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1318 dir_key->objectid, 1392 dir_key->objectid,
1319 name, name_len, 0); 1393 name, name_len, 0);
1320 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) { 1394 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1321 log_di = btrfs_lookup_dir_index_item(trans, log, 1395 log_di = btrfs_lookup_dir_index_item(trans, log,
1322 log_path, 1396 log_path,
1323 dir_key->objectid, 1397 dir_key->objectid,
@@ -1378,7 +1452,7 @@ static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1378 struct btrfs_root *root, 1452 struct btrfs_root *root,
1379 struct btrfs_root *log, 1453 struct btrfs_root *log,
1380 struct btrfs_path *path, 1454 struct btrfs_path *path,
1381 u64 dirid) 1455 u64 dirid, int del_all)
1382{ 1456{
1383 u64 range_start; 1457 u64 range_start;
1384 u64 range_end; 1458 u64 range_end;
@@ -1408,10 +1482,14 @@ again:
1408 range_start = 0; 1482 range_start = 0;
1409 range_end = 0; 1483 range_end = 0;
1410 while (1) { 1484 while (1) {
1411 ret = find_dir_range(log, path, dirid, key_type, 1485 if (del_all)
1412 &range_start, &range_end); 1486 range_end = (u64)-1;
1413 if (ret != 0) 1487 else {
1414 break; 1488 ret = find_dir_range(log, path, dirid, key_type,
1489 &range_start, &range_end);
1490 if (ret != 0)
1491 break;
1492 }
1415 1493
1416 dir_key.offset = range_start; 1494 dir_key.offset = range_start;
1417 while (1) { 1495 while (1) {
@@ -1437,7 +1515,8 @@ again:
1437 break; 1515 break;
1438 1516
1439 ret = check_item_in_log(trans, root, log, path, 1517 ret = check_item_in_log(trans, root, log, path,
1440 log_path, dir, &found_key); 1518 log_path, dir,
1519 &found_key);
1441 BUG_ON(ret); 1520 BUG_ON(ret);
1442 if (found_key.offset == (u64)-1) 1521 if (found_key.offset == (u64)-1)
1443 break; 1522 break;
@@ -1514,7 +1593,7 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1514 mode = btrfs_inode_mode(eb, inode_item); 1593 mode = btrfs_inode_mode(eb, inode_item);
1515 if (S_ISDIR(mode)) { 1594 if (S_ISDIR(mode)) {
1516 ret = replay_dir_deletes(wc->trans, 1595 ret = replay_dir_deletes(wc->trans,
1517 root, log, path, key.objectid); 1596 root, log, path, key.objectid, 0);
1518 BUG_ON(ret); 1597 BUG_ON(ret);
1519 } 1598 }
1520 ret = overwrite_item(wc->trans, root, path, 1599 ret = overwrite_item(wc->trans, root, path,
@@ -1533,6 +1612,17 @@ static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1533 root, inode, inode->i_size, 1612 root, inode, inode->i_size,
1534 BTRFS_EXTENT_DATA_KEY); 1613 BTRFS_EXTENT_DATA_KEY);
1535 BUG_ON(ret); 1614 BUG_ON(ret);
1615
1616 /* if the nlink count is zero here, the iput
1617 * will free the inode. We bump it to make
1618 * sure it doesn't get freed until the link
1619 * count fixup is done
1620 */
1621 if (inode->i_nlink == 0) {
1622 btrfs_inc_nlink(inode);
1623 btrfs_update_inode(wc->trans,
1624 root, inode);
1625 }
1536 iput(inode); 1626 iput(inode);
1537 } 1627 }
1538 ret = link_to_fixup_dir(wc->trans, root, 1628 ret = link_to_fixup_dir(wc->trans, root,
@@ -1840,7 +1930,8 @@ static int update_log_root(struct btrfs_trans_handle *trans,
1840 return ret; 1930 return ret;
1841} 1931}
1842 1932
1843static int wait_log_commit(struct btrfs_root *root, unsigned long transid) 1933static int wait_log_commit(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root, unsigned long transid)
1844{ 1935{
1845 DEFINE_WAIT(wait); 1936 DEFINE_WAIT(wait);
1846 int index = transid % 2; 1937 int index = transid % 2;
@@ -1854,9 +1945,12 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid)
1854 prepare_to_wait(&root->log_commit_wait[index], 1945 prepare_to_wait(&root->log_commit_wait[index],
1855 &wait, TASK_UNINTERRUPTIBLE); 1946 &wait, TASK_UNINTERRUPTIBLE);
1856 mutex_unlock(&root->log_mutex); 1947 mutex_unlock(&root->log_mutex);
1857 if (root->log_transid < transid + 2 && 1948
1949 if (root->fs_info->last_trans_log_full_commit !=
1950 trans->transid && root->log_transid < transid + 2 &&
1858 atomic_read(&root->log_commit[index])) 1951 atomic_read(&root->log_commit[index]))
1859 schedule(); 1952 schedule();
1953
1860 finish_wait(&root->log_commit_wait[index], &wait); 1954 finish_wait(&root->log_commit_wait[index], &wait);
1861 mutex_lock(&root->log_mutex); 1955 mutex_lock(&root->log_mutex);
1862 } while (root->log_transid < transid + 2 && 1956 } while (root->log_transid < transid + 2 &&
@@ -1864,14 +1958,16 @@ static int wait_log_commit(struct btrfs_root *root, unsigned long transid)
1864 return 0; 1958 return 0;
1865} 1959}
1866 1960
1867static int wait_for_writer(struct btrfs_root *root) 1961static int wait_for_writer(struct btrfs_trans_handle *trans,
1962 struct btrfs_root *root)
1868{ 1963{
1869 DEFINE_WAIT(wait); 1964 DEFINE_WAIT(wait);
1870 while (atomic_read(&root->log_writers)) { 1965 while (atomic_read(&root->log_writers)) {
1871 prepare_to_wait(&root->log_writer_wait, 1966 prepare_to_wait(&root->log_writer_wait,
1872 &wait, TASK_UNINTERRUPTIBLE); 1967 &wait, TASK_UNINTERRUPTIBLE);
1873 mutex_unlock(&root->log_mutex); 1968 mutex_unlock(&root->log_mutex);
1874 if (atomic_read(&root->log_writers)) 1969 if (root->fs_info->last_trans_log_full_commit !=
1970 trans->transid && atomic_read(&root->log_writers))
1875 schedule(); 1971 schedule();
1876 mutex_lock(&root->log_mutex); 1972 mutex_lock(&root->log_mutex);
1877 finish_wait(&root->log_writer_wait, &wait); 1973 finish_wait(&root->log_writer_wait, &wait);
@@ -1882,7 +1978,14 @@ static int wait_for_writer(struct btrfs_root *root)
1882/* 1978/*
1883 * btrfs_sync_log does sends a given tree log down to the disk and 1979 * btrfs_sync_log does sends a given tree log down to the disk and
1884 * updates the super blocks to record it. When this call is done, 1980 * updates the super blocks to record it. When this call is done,
1885 * you know that any inodes previously logged are safely on disk 1981 * you know that any inodes previously logged are safely on disk only
1982 * if it returns 0.
1983 *
1984 * Any other return value means you need to call btrfs_commit_transaction.
1985 * Some of the edge cases for fsyncing directories that have had unlinks
1986 * or renames done in the past mean that sometimes the only safe
1987 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1988 * that has happened.
1886 */ 1989 */
1887int btrfs_sync_log(struct btrfs_trans_handle *trans, 1990int btrfs_sync_log(struct btrfs_trans_handle *trans,
1888 struct btrfs_root *root) 1991 struct btrfs_root *root)
@@ -1896,7 +1999,7 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1896 mutex_lock(&root->log_mutex); 1999 mutex_lock(&root->log_mutex);
1897 index1 = root->log_transid % 2; 2000 index1 = root->log_transid % 2;
1898 if (atomic_read(&root->log_commit[index1])) { 2001 if (atomic_read(&root->log_commit[index1])) {
1899 wait_log_commit(root, root->log_transid); 2002 wait_log_commit(trans, root, root->log_transid);
1900 mutex_unlock(&root->log_mutex); 2003 mutex_unlock(&root->log_mutex);
1901 return 0; 2004 return 0;
1902 } 2005 }
@@ -1904,18 +2007,26 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1904 2007
1905 /* wait for previous tree log sync to complete */ 2008 /* wait for previous tree log sync to complete */
1906 if (atomic_read(&root->log_commit[(index1 + 1) % 2])) 2009 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1907 wait_log_commit(root, root->log_transid - 1); 2010 wait_log_commit(trans, root, root->log_transid - 1);
1908 2011
1909 while (1) { 2012 while (1) {
1910 unsigned long batch = root->log_batch; 2013 unsigned long batch = root->log_batch;
1911 mutex_unlock(&root->log_mutex); 2014 mutex_unlock(&root->log_mutex);
1912 schedule_timeout_uninterruptible(1); 2015 schedule_timeout_uninterruptible(1);
1913 mutex_lock(&root->log_mutex); 2016 mutex_lock(&root->log_mutex);
1914 wait_for_writer(root); 2017
2018 wait_for_writer(trans, root);
1915 if (batch == root->log_batch) 2019 if (batch == root->log_batch)
1916 break; 2020 break;
1917 } 2021 }
1918 2022
2023 /* bail out if we need to do a full commit */
2024 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2025 ret = -EAGAIN;
2026 mutex_unlock(&root->log_mutex);
2027 goto out;
2028 }
2029
1919 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages); 2030 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1920 BUG_ON(ret); 2031 BUG_ON(ret);
1921 2032
@@ -1951,16 +2062,29 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1951 2062
1952 index2 = log_root_tree->log_transid % 2; 2063 index2 = log_root_tree->log_transid % 2;
1953 if (atomic_read(&log_root_tree->log_commit[index2])) { 2064 if (atomic_read(&log_root_tree->log_commit[index2])) {
1954 wait_log_commit(log_root_tree, log_root_tree->log_transid); 2065 wait_log_commit(trans, log_root_tree,
2066 log_root_tree->log_transid);
1955 mutex_unlock(&log_root_tree->log_mutex); 2067 mutex_unlock(&log_root_tree->log_mutex);
1956 goto out; 2068 goto out;
1957 } 2069 }
1958 atomic_set(&log_root_tree->log_commit[index2], 1); 2070 atomic_set(&log_root_tree->log_commit[index2], 1);
1959 2071
1960 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) 2072 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
1961 wait_log_commit(log_root_tree, log_root_tree->log_transid - 1); 2073 wait_log_commit(trans, log_root_tree,
2074 log_root_tree->log_transid - 1);
2075 }
2076
2077 wait_for_writer(trans, log_root_tree);
1962 2078
1963 wait_for_writer(log_root_tree); 2079 /*
2080 * now that we've moved on to the tree of log tree roots,
2081 * check the full commit flag again
2082 */
2083 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2084 mutex_unlock(&log_root_tree->log_mutex);
2085 ret = -EAGAIN;
2086 goto out_wake_log_root;
2087 }
1964 2088
1965 ret = btrfs_write_and_wait_marked_extents(log_root_tree, 2089 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
1966 &log_root_tree->dirty_log_pages); 2090 &log_root_tree->dirty_log_pages);
@@ -1985,7 +2109,9 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
1985 * in and cause problems either. 2109 * in and cause problems either.
1986 */ 2110 */
1987 write_ctree_super(trans, root->fs_info->tree_root, 2); 2111 write_ctree_super(trans, root->fs_info->tree_root, 2);
2112 ret = 0;
1988 2113
2114out_wake_log_root:
1989 atomic_set(&log_root_tree->log_commit[index2], 0); 2115 atomic_set(&log_root_tree->log_commit[index2], 0);
1990 smp_mb(); 2116 smp_mb();
1991 if (waitqueue_active(&log_root_tree->log_commit_wait[index2])) 2117 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
@@ -1998,7 +2124,8 @@ out:
1998 return 0; 2124 return 0;
1999} 2125}
2000 2126
2001/* * free all the extents used by the tree log. This should be called 2127/*
2128 * free all the extents used by the tree log. This should be called
2002 * at commit time of the full transaction 2129 * at commit time of the full transaction
2003 */ 2130 */
2004int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root) 2131int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
@@ -2132,7 +2259,7 @@ int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2132 2259
2133 btrfs_free_path(path); 2260 btrfs_free_path(path);
2134 mutex_unlock(&BTRFS_I(dir)->log_mutex); 2261 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2135 end_log_trans(root); 2262 btrfs_end_log_trans(root);
2136 2263
2137 return 0; 2264 return 0;
2138} 2265}
@@ -2159,7 +2286,7 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2159 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino, 2286 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2160 dirid, &index); 2287 dirid, &index);
2161 mutex_unlock(&BTRFS_I(inode)->log_mutex); 2288 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2162 end_log_trans(root); 2289 btrfs_end_log_trans(root);
2163 2290
2164 return ret; 2291 return ret;
2165} 2292}
@@ -2559,7 +2686,7 @@ static noinline int copy_items(struct btrfs_trans_handle *trans,
2559 * 2686 *
2560 * This handles both files and directories. 2687 * This handles both files and directories.
2561 */ 2688 */
2562static int __btrfs_log_inode(struct btrfs_trans_handle *trans, 2689static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2563 struct btrfs_root *root, struct inode *inode, 2690 struct btrfs_root *root, struct inode *inode,
2564 int inode_only) 2691 int inode_only)
2565{ 2692{
@@ -2585,28 +2712,17 @@ static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2585 min_key.offset = 0; 2712 min_key.offset = 0;
2586 2713
2587 max_key.objectid = inode->i_ino; 2714 max_key.objectid = inode->i_ino;
2715
2716 /* today the code can only do partial logging of directories */
2717 if (!S_ISDIR(inode->i_mode))
2718 inode_only = LOG_INODE_ALL;
2719
2588 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode)) 2720 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2589 max_key.type = BTRFS_XATTR_ITEM_KEY; 2721 max_key.type = BTRFS_XATTR_ITEM_KEY;
2590 else 2722 else
2591 max_key.type = (u8)-1; 2723 max_key.type = (u8)-1;
2592 max_key.offset = (u64)-1; 2724 max_key.offset = (u64)-1;
2593 2725
2594 /*
2595 * if this inode has already been logged and we're in inode_only
2596 * mode, we don't want to delete the things that have already
2597 * been written to the log.
2598 *
2599 * But, if the inode has been through an inode_only log,
2600 * the logged_trans field is not set. This allows us to catch
2601 * any new names for this inode in the backrefs by logging it
2602 * again
2603 */
2604 if (inode_only == LOG_INODE_EXISTS &&
2605 BTRFS_I(inode)->logged_trans == trans->transid) {
2606 btrfs_free_path(path);
2607 btrfs_free_path(dst_path);
2608 goto out;
2609 }
2610 mutex_lock(&BTRFS_I(inode)->log_mutex); 2726 mutex_lock(&BTRFS_I(inode)->log_mutex);
2611 2727
2612 /* 2728 /*
@@ -2693,7 +2809,6 @@ next_slot:
2693 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) { 2809 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2694 btrfs_release_path(root, path); 2810 btrfs_release_path(root, path);
2695 btrfs_release_path(log, dst_path); 2811 btrfs_release_path(log, dst_path);
2696 BTRFS_I(inode)->log_dirty_trans = 0;
2697 ret = log_directory_changes(trans, root, inode, path, dst_path); 2812 ret = log_directory_changes(trans, root, inode, path, dst_path);
2698 BUG_ON(ret); 2813 BUG_ON(ret);
2699 } 2814 }
@@ -2702,19 +2817,69 @@ next_slot:
2702 2817
2703 btrfs_free_path(path); 2818 btrfs_free_path(path);
2704 btrfs_free_path(dst_path); 2819 btrfs_free_path(dst_path);
2705out:
2706 return 0; 2820 return 0;
2707} 2821}
2708 2822
2709int btrfs_log_inode(struct btrfs_trans_handle *trans, 2823/*
2710 struct btrfs_root *root, struct inode *inode, 2824 * follow the dentry parent pointers up the chain and see if any
2711 int inode_only) 2825 * of the directories in it require a full commit before they can
2826 * be logged. Returns zero if nothing special needs to be done or 1 if
2827 * a full commit is required.
2828 */
2829static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2830 struct inode *inode,
2831 struct dentry *parent,
2832 struct super_block *sb,
2833 u64 last_committed)
2712{ 2834{
2713 int ret; 2835 int ret = 0;
2836 struct btrfs_root *root;
2714 2837
2715 start_log_trans(trans, root); 2838 /*
2716 ret = __btrfs_log_inode(trans, root, inode, inode_only); 2839 * for regular files, if its inode is already on disk, we don't
2717 end_log_trans(root); 2840 * have to worry about the parents at all. This is because
2841 * we can use the last_unlink_trans field to record renames
2842 * and other fun in this file.
2843 */
2844 if (S_ISREG(inode->i_mode) &&
2845 BTRFS_I(inode)->generation <= last_committed &&
2846 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2847 goto out;
2848
2849 if (!S_ISDIR(inode->i_mode)) {
2850 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2851 goto out;
2852 inode = parent->d_inode;
2853 }
2854
2855 while (1) {
2856 BTRFS_I(inode)->logged_trans = trans->transid;
2857 smp_mb();
2858
2859 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2860 root = BTRFS_I(inode)->root;
2861
2862 /*
2863 * make sure any commits to the log are forced
2864 * to be full commits
2865 */
2866 root->fs_info->last_trans_log_full_commit =
2867 trans->transid;
2868 ret = 1;
2869 break;
2870 }
2871
2872 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2873 break;
2874
2875 if (parent == sb->s_root)
2876 break;
2877
2878 parent = parent->d_parent;
2879 inode = parent->d_inode;
2880
2881 }
2882out:
2718 return ret; 2883 return ret;
2719} 2884}
2720 2885
@@ -2724,31 +2889,65 @@ int btrfs_log_inode(struct btrfs_trans_handle *trans,
2724 * only logging is done of any parent directories that are older than 2889 * only logging is done of any parent directories that are older than
2725 * the last committed transaction 2890 * the last committed transaction
2726 */ 2891 */
2727int btrfs_log_dentry(struct btrfs_trans_handle *trans, 2892int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2728 struct btrfs_root *root, struct dentry *dentry) 2893 struct btrfs_root *root, struct inode *inode,
2894 struct dentry *parent, int exists_only)
2729{ 2895{
2730 int inode_only = LOG_INODE_ALL; 2896 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
2731 struct super_block *sb; 2897 struct super_block *sb;
2732 int ret; 2898 int ret = 0;
2899 u64 last_committed = root->fs_info->last_trans_committed;
2900
2901 sb = inode->i_sb;
2902
2903 if (root->fs_info->last_trans_log_full_commit >
2904 root->fs_info->last_trans_committed) {
2905 ret = 1;
2906 goto end_no_trans;
2907 }
2908
2909 ret = check_parent_dirs_for_sync(trans, inode, parent,
2910 sb, last_committed);
2911 if (ret)
2912 goto end_no_trans;
2733 2913
2734 start_log_trans(trans, root); 2914 start_log_trans(trans, root);
2735 sb = dentry->d_inode->i_sb;
2736 while (1) {
2737 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2738 inode_only);
2739 BUG_ON(ret);
2740 inode_only = LOG_INODE_EXISTS;
2741 2915
2742 dentry = dentry->d_parent; 2916 ret = btrfs_log_inode(trans, root, inode, inode_only);
2743 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb) 2917 BUG_ON(ret);
2918
2919 /*
2920 * for regular files, if its inode is already on disk, we don't
2921 * have to worry about the parents at all. This is because
2922 * we can use the last_unlink_trans field to record renames
2923 * and other fun in this file.
2924 */
2925 if (S_ISREG(inode->i_mode) &&
2926 BTRFS_I(inode)->generation <= last_committed &&
2927 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2928 goto no_parent;
2929
2930 inode_only = LOG_INODE_EXISTS;
2931 while (1) {
2932 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2744 break; 2933 break;
2745 2934
2746 if (BTRFS_I(dentry->d_inode)->generation <= 2935 inode = parent->d_inode;
2747 root->fs_info->last_trans_committed) 2936 if (BTRFS_I(inode)->generation >
2937 root->fs_info->last_trans_committed) {
2938 ret = btrfs_log_inode(trans, root, inode, inode_only);
2939 BUG_ON(ret);
2940 }
2941 if (parent == sb->s_root)
2748 break; 2942 break;
2943
2944 parent = parent->d_parent;
2749 } 2945 }
2750 end_log_trans(root); 2946no_parent:
2751 return 0; 2947 ret = 0;
2948 btrfs_end_log_trans(root);
2949end_no_trans:
2950 return ret;
2752} 2951}
2753 2952
2754/* 2953/*
@@ -2760,12 +2959,8 @@ int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2760int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, 2959int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2761 struct btrfs_root *root, struct dentry *dentry) 2960 struct btrfs_root *root, struct dentry *dentry)
2762{ 2961{
2763 u64 gen; 2962 return btrfs_log_inode_parent(trans, root, dentry->d_inode,
2764 gen = root->fs_info->last_trans_new_blockgroup; 2963 dentry->d_parent, 0);
2765 if (gen > root->fs_info->last_trans_committed)
2766 return 1;
2767 else
2768 return btrfs_log_dentry(trans, root, dentry);
2769} 2964}
2770 2965
2771/* 2966/*
@@ -2884,3 +3079,94 @@ again:
2884 kfree(log_root_tree); 3079 kfree(log_root_tree);
2885 return 0; 3080 return 0;
2886} 3081}
3082
3083/*
3084 * there are some corner cases where we want to force a full
3085 * commit instead of allowing a directory to be logged.
3086 *
3087 * They revolve around files there were unlinked from the directory, and
3088 * this function updates the parent directory so that a full commit is
3089 * properly done if it is fsync'd later after the unlinks are done.
3090 */
3091void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3092 struct inode *dir, struct inode *inode,
3093 int for_rename)
3094{
3095 /*
3096 * when we're logging a file, if it hasn't been renamed
3097 * or unlinked, and its inode is fully committed on disk,
3098 * we don't have to worry about walking up the directory chain
3099 * to log its parents.
3100 *
3101 * So, we use the last_unlink_trans field to put this transid
3102 * into the file. When the file is logged we check it and
3103 * don't log the parents if the file is fully on disk.
3104 */
3105 if (S_ISREG(inode->i_mode))
3106 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3107
3108 /*
3109 * if this directory was already logged any new
3110 * names for this file/dir will get recorded
3111 */
3112 smp_mb();
3113 if (BTRFS_I(dir)->logged_trans == trans->transid)
3114 return;
3115
3116 /*
3117 * if the inode we're about to unlink was logged,
3118 * the log will be properly updated for any new names
3119 */
3120 if (BTRFS_I(inode)->logged_trans == trans->transid)
3121 return;
3122
3123 /*
3124 * when renaming files across directories, if the directory
3125 * there we're unlinking from gets fsync'd later on, there's
3126 * no way to find the destination directory later and fsync it
3127 * properly. So, we have to be conservative and force commits
3128 * so the new name gets discovered.
3129 */
3130 if (for_rename)
3131 goto record;
3132
3133 /* we can safely do the unlink without any special recording */
3134 return;
3135
3136record:
3137 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3138}
3139
3140/*
3141 * Call this after adding a new name for a file and it will properly
3142 * update the log to reflect the new name.
3143 *
3144 * It will return zero if all goes well, and it will return 1 if a
3145 * full transaction commit is required.
3146 */
3147int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3148 struct inode *inode, struct inode *old_dir,
3149 struct dentry *parent)
3150{
3151 struct btrfs_root * root = BTRFS_I(inode)->root;
3152
3153 /*
3154 * this will force the logging code to walk the dentry chain
3155 * up for the file
3156 */
3157 if (S_ISREG(inode->i_mode))
3158 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3159
3160 /*
3161 * if this inode hasn't been logged and directory we're renaming it
3162 * from hasn't been logged, we don't need to log it
3163 */
3164 if (BTRFS_I(inode)->logged_trans <=
3165 root->fs_info->last_trans_committed &&
3166 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3167 root->fs_info->last_trans_committed))
3168 return 0;
3169
3170 return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3171}
3172
diff --git a/fs/btrfs/tree-log.h b/fs/btrfs/tree-log.h
index b9409b32ed02..d09c7609e16b 100644
--- a/fs/btrfs/tree-log.h
+++ b/fs/btrfs/tree-log.h
@@ -22,14 +22,9 @@
22int btrfs_sync_log(struct btrfs_trans_handle *trans, 22int btrfs_sync_log(struct btrfs_trans_handle *trans,
23 struct btrfs_root *root); 23 struct btrfs_root *root);
24int btrfs_free_log(struct btrfs_trans_handle *trans, 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); 25int btrfs_recover_log_trees(struct btrfs_root *tree_root);
28int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans, 26int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
29 struct btrfs_root *root, struct dentry *dentry); 27 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, 28int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root, 29 struct btrfs_root *root,
35 const char *name, int name_len, 30 const char *name, int name_len,
@@ -38,4 +33,16 @@ int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
38 struct btrfs_root *root, 33 struct btrfs_root *root,
39 const char *name, int name_len, 34 const char *name, int name_len,
40 struct inode *inode, u64 dirid); 35 struct inode *inode, u64 dirid);
36int btrfs_join_running_log_trans(struct btrfs_root *root);
37int btrfs_end_log_trans(struct btrfs_root *root);
38int btrfs_pin_log_trans(struct btrfs_root *root);
39int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
40 struct btrfs_root *root, struct inode *inode,
41 struct dentry *parent, int exists_only);
42void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
43 struct inode *dir, struct inode *inode,
44 int for_rename);
45int btrfs_log_new_name(struct btrfs_trans_handle *trans,
46 struct inode *inode, struct inode *old_dir,
47 struct dentry *parent);
41#endif 48#endif
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-18 10:48:44 -0500