xref: /openbmc/linux/fs/btrfs/inode.c (revision 1832a6d5ee3b1af61001cadba9e10da9e91af4a4)
1 /*
2  * Copyright (C) 2007 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/buffer_head.h>
20 #include <linux/fs.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
23 #include <linux/time.h>
24 #include <linux/init.h>
25 #include <linux/string.h>
26 #include <linux/smp_lock.h>
27 #include <linux/backing-dev.h>
28 #include <linux/mpage.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/statfs.h>
32 #include <linux/compat.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/version.h>
35 #include <linux/xattr.h>
36 #include "ctree.h"
37 #include "disk-io.h"
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "ioctl.h"
41 #include "print-tree.h"
42 
43 struct btrfs_iget_args {
44 	u64 ino;
45 	struct btrfs_root *root;
46 };
47 
48 static struct inode_operations btrfs_dir_inode_operations;
49 static struct inode_operations btrfs_symlink_inode_operations;
50 static struct inode_operations btrfs_dir_ro_inode_operations;
51 static struct inode_operations btrfs_special_inode_operations;
52 static struct inode_operations btrfs_file_inode_operations;
53 static struct address_space_operations btrfs_aops;
54 static struct address_space_operations btrfs_symlink_aops;
55 static struct file_operations btrfs_dir_file_operations;
56 static struct extent_map_ops btrfs_extent_map_ops;
57 
58 static struct kmem_cache *btrfs_inode_cachep;
59 struct kmem_cache *btrfs_trans_handle_cachep;
60 struct kmem_cache *btrfs_transaction_cachep;
61 struct kmem_cache *btrfs_bit_radix_cachep;
62 struct kmem_cache *btrfs_path_cachep;
63 
64 #define S_SHIFT 12
65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
66 	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
67 	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
68 	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
69 	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
70 	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
71 	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
72 	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
73 };
74 
75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
76 			   int for_del)
77 {
78 	u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
79 	u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
80 	u64 thresh;
81 	int ret = 0;
82 
83 	if (for_del)
84 		thresh = (total * 90) / 100;
85 	else
86 		thresh = (total * 85) / 100;
87 
88 	spin_lock(&root->fs_info->delalloc_lock);
89 	if (used + root->fs_info->delalloc_bytes + num_required > thresh)
90 		ret = -ENOSPC;
91 	spin_unlock(&root->fs_info->delalloc_lock);
92 	return ret;
93 }
94 
95 static int cow_file_range(struct inode *inode, u64 start, u64 end)
96 {
97 	struct btrfs_root *root = BTRFS_I(inode)->root;
98 	struct btrfs_trans_handle *trans;
99 	u64 alloc_hint = 0;
100 	u64 num_bytes;
101 	u64 cur_alloc_size;
102 	u64 blocksize = root->sectorsize;
103 	struct btrfs_key ins;
104 	int ret;
105 
106 	trans = btrfs_start_transaction(root, 1);
107 	BUG_ON(!trans);
108 	btrfs_set_trans_block_group(trans, inode);
109 
110 	num_bytes = (end - start + blocksize) & ~(blocksize - 1);
111 	num_bytes = max(blocksize,  num_bytes);
112 	ret = btrfs_drop_extents(trans, root, inode,
113 				 start, start + num_bytes, start, &alloc_hint);
114 
115 	if (alloc_hint == EXTENT_MAP_INLINE)
116 		goto out;
117 
118 	while(num_bytes > 0) {
119 		cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
120 		ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
121 					 root->root_key.objectid,
122 					 trans->transid,
123 					 inode->i_ino, start, 0,
124 					 alloc_hint, (u64)-1, &ins, 1);
125 		if (ret) {
126 			WARN_ON(1);
127 			goto out;
128 		}
129 		ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
130 					       start, ins.objectid, ins.offset,
131 					       ins.offset);
132 		num_bytes -= cur_alloc_size;
133 		alloc_hint = ins.objectid + ins.offset;
134 		start += cur_alloc_size;
135 	}
136 out:
137 	btrfs_end_transaction(trans, root);
138 	return ret;
139 }
140 
141 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
142 {
143 	u64 extent_start;
144 	u64 extent_end;
145 	u64 bytenr;
146 	u64 cow_end;
147 	u64 loops = 0;
148 	struct btrfs_root *root = BTRFS_I(inode)->root;
149 	struct extent_buffer *leaf;
150 	int found_type;
151 	struct btrfs_path *path;
152 	struct btrfs_file_extent_item *item;
153 	int ret;
154 	int err;
155 	struct btrfs_key found_key;
156 
157 	path = btrfs_alloc_path();
158 	BUG_ON(!path);
159 again:
160 	ret = btrfs_lookup_file_extent(NULL, root, path,
161 				       inode->i_ino, start, 0);
162 	if (ret < 0) {
163 		btrfs_free_path(path);
164 		return ret;
165 	}
166 
167 	cow_end = end;
168 	if (ret != 0) {
169 		if (path->slots[0] == 0)
170 			goto not_found;
171 		path->slots[0]--;
172 	}
173 
174 	leaf = path->nodes[0];
175 	item = btrfs_item_ptr(leaf, path->slots[0],
176 			      struct btrfs_file_extent_item);
177 
178 	/* are we inside the extent that was found? */
179 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
180 	found_type = btrfs_key_type(&found_key);
181 	if (found_key.objectid != inode->i_ino ||
182 	    found_type != BTRFS_EXTENT_DATA_KEY) {
183 		goto not_found;
184 	}
185 
186 	found_type = btrfs_file_extent_type(leaf, item);
187 	extent_start = found_key.offset;
188 	if (found_type == BTRFS_FILE_EXTENT_REG) {
189 		extent_end = extent_start +
190 		       btrfs_file_extent_num_bytes(leaf, item);
191 		err = 0;
192 
193 		if (loops && start != extent_start)
194 			goto not_found;
195 
196 		if (start < extent_start || start >= extent_end)
197 			goto not_found;
198 
199 		cow_end = min(end, extent_end - 1);
200 		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
201 		if (bytenr == 0)
202 			goto not_found;
203 
204 		if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
205 			goto not_found;
206 		}
207 
208 		start = extent_end;
209 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
210 		goto not_found;
211 	}
212 loop:
213 	if (start > end) {
214 		btrfs_free_path(path);
215 		return 0;
216 	}
217 	btrfs_release_path(root, path);
218 	loops++;
219 	goto again;
220 
221 not_found:
222 	cow_file_range(inode, start, cow_end);
223 	start = cow_end + 1;
224 	goto loop;
225 }
226 
227 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
228 {
229 	struct btrfs_root *root = BTRFS_I(inode)->root;
230 	u64 num_bytes;
231 	int ret;
232 
233 	mutex_lock(&root->fs_info->fs_mutex);
234 	if (btrfs_test_opt(root, NODATACOW))
235 		ret = run_delalloc_nocow(inode, start, end);
236 	else
237 		ret = cow_file_range(inode, start, end);
238 
239 	spin_lock(&root->fs_info->delalloc_lock);
240 	num_bytes = end + 1 - start;
241 	if (root->fs_info->delalloc_bytes < num_bytes) {
242 		printk("delalloc accounting error total %llu sub %llu\n",
243 		       root->fs_info->delalloc_bytes, num_bytes);
244 	} else {
245 		root->fs_info->delalloc_bytes -= num_bytes;
246 	}
247 	spin_unlock(&root->fs_info->delalloc_lock);
248 
249 	mutex_unlock(&root->fs_info->fs_mutex);
250 	return ret;
251 }
252 
253 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
254 {
255 	struct inode *inode = page->mapping->host;
256 	struct btrfs_root *root = BTRFS_I(inode)->root;
257 	struct btrfs_trans_handle *trans;
258 	char *kaddr;
259 	int ret = 0;
260 	u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
261 	size_t offset = start - page_start;
262 
263 	if (btrfs_test_opt(root, NODATASUM))
264 		return 0;
265 
266 	mutex_lock(&root->fs_info->fs_mutex);
267 	trans = btrfs_start_transaction(root, 1);
268 	btrfs_set_trans_block_group(trans, inode);
269 	kaddr = kmap(page);
270 	btrfs_csum_file_block(trans, root, inode, inode->i_ino,
271 			      start, kaddr + offset, end - start + 1);
272 	kunmap(page);
273 	ret = btrfs_end_transaction(trans, root);
274 	BUG_ON(ret);
275 	mutex_unlock(&root->fs_info->fs_mutex);
276 	return ret;
277 }
278 
279 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
280 {
281 	int ret = 0;
282 	struct inode *inode = page->mapping->host;
283 	struct btrfs_root *root = BTRFS_I(inode)->root;
284 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
285 	struct btrfs_csum_item *item;
286 	struct btrfs_path *path = NULL;
287 	u32 csum;
288 
289 	if (btrfs_test_opt(root, NODATASUM))
290 		return 0;
291 
292 	mutex_lock(&root->fs_info->fs_mutex);
293 	path = btrfs_alloc_path();
294 	item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
295 	if (IS_ERR(item)) {
296 		ret = PTR_ERR(item);
297 		/* a csum that isn't present is a preallocated region. */
298 		if (ret == -ENOENT || ret == -EFBIG)
299 			ret = 0;
300 		csum = 0;
301 		goto out;
302 	}
303 	read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
304 			   BTRFS_CRC32_SIZE);
305 	set_state_private(em_tree, start, csum);
306 out:
307 	if (path)
308 		btrfs_free_path(path);
309 	mutex_unlock(&root->fs_info->fs_mutex);
310 	return ret;
311 }
312 
313 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
314 {
315 	size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
316 	struct inode *inode = page->mapping->host;
317 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
318 	char *kaddr;
319 	u64 private;
320 	int ret;
321 	struct btrfs_root *root = BTRFS_I(inode)->root;
322 	u32 csum = ~(u32)0;
323 	unsigned long flags;
324 
325 	if (btrfs_test_opt(root, NODATASUM))
326 		return 0;
327 
328 	ret = get_state_private(em_tree, start, &private);
329 	local_irq_save(flags);
330 	kaddr = kmap_atomic(page, KM_IRQ0);
331 	if (ret) {
332 		goto zeroit;
333 	}
334 	csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
335 	btrfs_csum_final(csum, (char *)&csum);
336 	if (csum != private) {
337 		goto zeroit;
338 	}
339 	kunmap_atomic(kaddr, KM_IRQ0);
340 	local_irq_restore(flags);
341 	return 0;
342 
343 zeroit:
344 	printk("btrfs csum failed ino %lu off %llu\n",
345 	       page->mapping->host->i_ino, (unsigned long long)start);
346 	memset(kaddr + offset, 1, end - start + 1);
347 	flush_dcache_page(page);
348 	kunmap_atomic(kaddr, KM_IRQ0);
349 	local_irq_restore(flags);
350 	return 0;
351 }
352 
353 void btrfs_read_locked_inode(struct inode *inode)
354 {
355 	struct btrfs_path *path;
356 	struct extent_buffer *leaf;
357 	struct btrfs_inode_item *inode_item;
358 	struct btrfs_inode_timespec *tspec;
359 	struct btrfs_root *root = BTRFS_I(inode)->root;
360 	struct btrfs_key location;
361 	u64 alloc_group_block;
362 	u32 rdev;
363 	int ret;
364 
365 	path = btrfs_alloc_path();
366 	BUG_ON(!path);
367 	mutex_lock(&root->fs_info->fs_mutex);
368 
369 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
370 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
371 	if (ret)
372 		goto make_bad;
373 
374 	leaf = path->nodes[0];
375 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
376 				    struct btrfs_inode_item);
377 
378 	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
379 	inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
380 	inode->i_uid = btrfs_inode_uid(leaf, inode_item);
381 	inode->i_gid = btrfs_inode_gid(leaf, inode_item);
382 	inode->i_size = btrfs_inode_size(leaf, inode_item);
383 
384 	tspec = btrfs_inode_atime(inode_item);
385 	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
386 	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
387 
388 	tspec = btrfs_inode_mtime(inode_item);
389 	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
390 	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
391 
392 	tspec = btrfs_inode_ctime(inode_item);
393 	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
394 	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
395 
396 	inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
397 	inode->i_generation = btrfs_inode_generation(leaf, inode_item);
398 	inode->i_rdev = 0;
399 	rdev = btrfs_inode_rdev(leaf, inode_item);
400 
401 	alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
402 	BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
403 						       alloc_group_block);
404 
405 	btrfs_free_path(path);
406 	inode_item = NULL;
407 
408 	mutex_unlock(&root->fs_info->fs_mutex);
409 
410 	switch (inode->i_mode & S_IFMT) {
411 	case S_IFREG:
412 		inode->i_mapping->a_ops = &btrfs_aops;
413 		BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
414 		inode->i_fop = &btrfs_file_operations;
415 		inode->i_op = &btrfs_file_inode_operations;
416 		break;
417 	case S_IFDIR:
418 		inode->i_fop = &btrfs_dir_file_operations;
419 		if (root == root->fs_info->tree_root)
420 			inode->i_op = &btrfs_dir_ro_inode_operations;
421 		else
422 			inode->i_op = &btrfs_dir_inode_operations;
423 		break;
424 	case S_IFLNK:
425 		inode->i_op = &btrfs_symlink_inode_operations;
426 		inode->i_mapping->a_ops = &btrfs_symlink_aops;
427 		break;
428 	default:
429 		init_special_inode(inode, inode->i_mode, rdev);
430 		break;
431 	}
432 	return;
433 
434 make_bad:
435 	btrfs_release_path(root, path);
436 	btrfs_free_path(path);
437 	mutex_unlock(&root->fs_info->fs_mutex);
438 	make_bad_inode(inode);
439 }
440 
441 static void fill_inode_item(struct extent_buffer *leaf,
442 			    struct btrfs_inode_item *item,
443 			    struct inode *inode)
444 {
445 	btrfs_set_inode_uid(leaf, item, inode->i_uid);
446 	btrfs_set_inode_gid(leaf, item, inode->i_gid);
447 	btrfs_set_inode_size(leaf, item, inode->i_size);
448 	btrfs_set_inode_mode(leaf, item, inode->i_mode);
449 	btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
450 
451 	btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
452 			       inode->i_atime.tv_sec);
453 	btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
454 				inode->i_atime.tv_nsec);
455 
456 	btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
457 			       inode->i_mtime.tv_sec);
458 	btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
459 				inode->i_mtime.tv_nsec);
460 
461 	btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
462 			       inode->i_ctime.tv_sec);
463 	btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
464 				inode->i_ctime.tv_nsec);
465 
466 	btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
467 	btrfs_set_inode_generation(leaf, item, inode->i_generation);
468 	btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
469 	btrfs_set_inode_block_group(leaf, item,
470 				    BTRFS_I(inode)->block_group->key.objectid);
471 }
472 
473 int btrfs_update_inode(struct btrfs_trans_handle *trans,
474 			      struct btrfs_root *root,
475 			      struct inode *inode)
476 {
477 	struct btrfs_inode_item *inode_item;
478 	struct btrfs_path *path;
479 	struct extent_buffer *leaf;
480 	int ret;
481 
482 	path = btrfs_alloc_path();
483 	BUG_ON(!path);
484 	ret = btrfs_lookup_inode(trans, root, path,
485 				 &BTRFS_I(inode)->location, 1);
486 	if (ret) {
487 		if (ret > 0)
488 			ret = -ENOENT;
489 		goto failed;
490 	}
491 
492 	leaf = path->nodes[0];
493 	inode_item = btrfs_item_ptr(leaf, path->slots[0],
494 				  struct btrfs_inode_item);
495 
496 	fill_inode_item(leaf, inode_item, inode);
497 	btrfs_mark_buffer_dirty(leaf);
498 	btrfs_set_inode_last_trans(trans, inode);
499 	ret = 0;
500 failed:
501 	btrfs_release_path(root, path);
502 	btrfs_free_path(path);
503 	return ret;
504 }
505 
506 
507 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
508 			      struct btrfs_root *root,
509 			      struct inode *dir,
510 			      struct dentry *dentry)
511 {
512 	struct btrfs_path *path;
513 	const char *name = dentry->d_name.name;
514 	int name_len = dentry->d_name.len;
515 	int ret = 0;
516 	struct extent_buffer *leaf;
517 	struct btrfs_dir_item *di;
518 	struct btrfs_key key;
519 
520 	path = btrfs_alloc_path();
521 	if (!path) {
522 		ret = -ENOMEM;
523 		goto err;
524 	}
525 
526 	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
527 				    name, name_len, -1);
528 	if (IS_ERR(di)) {
529 		ret = PTR_ERR(di);
530 		goto err;
531 	}
532 	if (!di) {
533 		ret = -ENOENT;
534 		goto err;
535 	}
536 	leaf = path->nodes[0];
537 	btrfs_dir_item_key_to_cpu(leaf, di, &key);
538 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
539 	if (ret)
540 		goto err;
541 	btrfs_release_path(root, path);
542 
543 	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
544 					 key.objectid, name, name_len, -1);
545 	if (IS_ERR(di)) {
546 		ret = PTR_ERR(di);
547 		goto err;
548 	}
549 	if (!di) {
550 		ret = -ENOENT;
551 		goto err;
552 	}
553 	ret = btrfs_delete_one_dir_name(trans, root, path, di);
554 
555 	dentry->d_inode->i_ctime = dir->i_ctime;
556 	ret = btrfs_del_inode_ref(trans, root, name, name_len,
557 				  dentry->d_inode->i_ino,
558 				  dentry->d_parent->d_inode->i_ino);
559 	if (ret) {
560 		printk("failed to delete reference to %.*s, "
561 		       "inode %lu parent %lu\n", name_len, name,
562 		       dentry->d_inode->i_ino,
563 		       dentry->d_parent->d_inode->i_ino);
564 	}
565 err:
566 	btrfs_free_path(path);
567 	if (!ret) {
568 		dir->i_size -= name_len * 2;
569 		dir->i_mtime = dir->i_ctime = CURRENT_TIME;
570 		btrfs_update_inode(trans, root, dir);
571 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
572 		dentry->d_inode->i_nlink--;
573 #else
574 		drop_nlink(dentry->d_inode);
575 #endif
576 		ret = btrfs_update_inode(trans, root, dentry->d_inode);
577 		dir->i_sb->s_dirt = 1;
578 	}
579 	return ret;
580 }
581 
582 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
583 {
584 	struct btrfs_root *root;
585 	struct btrfs_trans_handle *trans;
586 	int ret;
587 	unsigned long nr = 0;
588 
589 	root = BTRFS_I(dir)->root;
590 	mutex_lock(&root->fs_info->fs_mutex);
591 
592 	ret = btrfs_check_free_space(root, 1, 1);
593 	if (ret)
594 		goto fail;
595 
596 	trans = btrfs_start_transaction(root, 1);
597 
598 	btrfs_set_trans_block_group(trans, dir);
599 	ret = btrfs_unlink_trans(trans, root, dir, dentry);
600 	nr = trans->blocks_used;
601 
602 	btrfs_end_transaction(trans, root);
603 fail:
604 	mutex_unlock(&root->fs_info->fs_mutex);
605 	btrfs_btree_balance_dirty(root, nr);
606 	return ret;
607 }
608 
609 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
610 {
611 	struct inode *inode = dentry->d_inode;
612 	int err = 0;
613 	int ret;
614 	struct btrfs_root *root = BTRFS_I(dir)->root;
615 	struct btrfs_trans_handle *trans;
616 	unsigned long nr = 0;
617 
618 	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
619 		return -ENOTEMPTY;
620 
621 	mutex_lock(&root->fs_info->fs_mutex);
622 	ret = btrfs_check_free_space(root, 1, 1);
623 	if (ret)
624 		goto fail;
625 
626 	trans = btrfs_start_transaction(root, 1);
627 	btrfs_set_trans_block_group(trans, dir);
628 
629 	/* now the directory is empty */
630 	err = btrfs_unlink_trans(trans, root, dir, dentry);
631 	if (!err) {
632 		inode->i_size = 0;
633 	}
634 
635 	nr = trans->blocks_used;
636 	ret = btrfs_end_transaction(trans, root);
637 fail:
638 	mutex_unlock(&root->fs_info->fs_mutex);
639 	btrfs_btree_balance_dirty(root, nr);
640 
641 	if (ret && !err)
642 		err = ret;
643 	return err;
644 }
645 
646 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
647 			    struct btrfs_root *root,
648 			    struct inode *inode)
649 {
650 	struct btrfs_path *path;
651 	int ret;
652 
653 	clear_inode(inode);
654 
655 	path = btrfs_alloc_path();
656 	BUG_ON(!path);
657 	ret = btrfs_lookup_inode(trans, root, path,
658 				 &BTRFS_I(inode)->location, -1);
659 	if (ret > 0)
660 		ret = -ENOENT;
661 	if (!ret)
662 		ret = btrfs_del_item(trans, root, path);
663 	btrfs_free_path(path);
664 	return ret;
665 }
666 
667 /*
668  * this can truncate away extent items, csum items and directory items.
669  * It starts at a high offset and removes keys until it can't find
670  * any higher than i_size.
671  *
672  * csum items that cross the new i_size are truncated to the new size
673  * as well.
674  */
675 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
676 				   struct btrfs_root *root,
677 				   struct inode *inode)
678 {
679 	int ret;
680 	struct btrfs_path *path;
681 	struct btrfs_key key;
682 	struct btrfs_key found_key;
683 	u32 found_type;
684 	struct extent_buffer *leaf;
685 	struct btrfs_file_extent_item *fi;
686 	u64 extent_start = 0;
687 	u64 extent_num_bytes = 0;
688 	u64 item_end = 0;
689 	u64 root_gen = 0;
690 	u64 root_owner = 0;
691 	int found_extent;
692 	int del_item;
693 	int extent_type = -1;
694 
695 	btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
696 	path = btrfs_alloc_path();
697 	path->reada = -1;
698 	BUG_ON(!path);
699 
700 	/* FIXME, add redo link to tree so we don't leak on crash */
701 	key.objectid = inode->i_ino;
702 	key.offset = (u64)-1;
703 	key.type = (u8)-1;
704 
705 	while(1) {
706 		btrfs_init_path(path);
707 		fi = NULL;
708 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
709 		if (ret < 0) {
710 			goto error;
711 		}
712 		if (ret > 0) {
713 			BUG_ON(path->slots[0] == 0);
714 			path->slots[0]--;
715 		}
716 		leaf = path->nodes[0];
717 		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
718 		found_type = btrfs_key_type(&found_key);
719 
720 		if (found_key.objectid != inode->i_ino)
721 			break;
722 
723 		if (found_type != BTRFS_CSUM_ITEM_KEY &&
724 		    found_type != BTRFS_DIR_ITEM_KEY &&
725 		    found_type != BTRFS_DIR_INDEX_KEY &&
726 		    found_type != BTRFS_EXTENT_DATA_KEY)
727 			break;
728 
729 		item_end = found_key.offset;
730 		if (found_type == BTRFS_EXTENT_DATA_KEY) {
731 			fi = btrfs_item_ptr(leaf, path->slots[0],
732 					    struct btrfs_file_extent_item);
733 			extent_type = btrfs_file_extent_type(leaf, fi);
734 			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
735 				item_end +=
736 				    btrfs_file_extent_num_bytes(leaf, fi);
737 			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
738 				struct btrfs_item *item = btrfs_item_nr(leaf,
739 							        path->slots[0]);
740 				item_end += btrfs_file_extent_inline_len(leaf,
741 									 item);
742 			}
743 			item_end--;
744 		}
745 		if (found_type == BTRFS_CSUM_ITEM_KEY) {
746 			ret = btrfs_csum_truncate(trans, root, path,
747 						  inode->i_size);
748 			BUG_ON(ret);
749 		}
750 		if (item_end < inode->i_size) {
751 			if (found_type == BTRFS_DIR_ITEM_KEY) {
752 				found_type = BTRFS_INODE_ITEM_KEY;
753 			} else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
754 				found_type = BTRFS_CSUM_ITEM_KEY;
755 			} else if (found_type) {
756 				found_type--;
757 			} else {
758 				break;
759 			}
760 			btrfs_set_key_type(&key, found_type);
761 			btrfs_release_path(root, path);
762 			continue;
763 		}
764 		if (found_key.offset >= inode->i_size)
765 			del_item = 1;
766 		else
767 			del_item = 0;
768 		found_extent = 0;
769 
770 		/* FIXME, shrink the extent if the ref count is only 1 */
771 		if (found_type != BTRFS_EXTENT_DATA_KEY)
772 			goto delete;
773 
774 		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
775 			u64 num_dec;
776 			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
777 			if (!del_item) {
778 				u64 orig_num_bytes =
779 					btrfs_file_extent_num_bytes(leaf, fi);
780 				extent_num_bytes = inode->i_size -
781 					found_key.offset + root->sectorsize - 1;
782 				btrfs_set_file_extent_num_bytes(leaf, fi,
783 							 extent_num_bytes);
784 				num_dec = (orig_num_bytes -
785 					   extent_num_bytes) >> 9;
786 				if (extent_start != 0) {
787 					inode->i_blocks -= num_dec;
788 				}
789 				btrfs_mark_buffer_dirty(leaf);
790 			} else {
791 				extent_num_bytes =
792 					btrfs_file_extent_disk_num_bytes(leaf,
793 									 fi);
794 				/* FIXME blocksize != 4096 */
795 				num_dec = btrfs_file_extent_num_bytes(leaf,
796 								       fi) >> 9;
797 				if (extent_start != 0) {
798 					found_extent = 1;
799 					inode->i_blocks -= num_dec;
800 				}
801 				root_gen = btrfs_header_generation(leaf);
802 				root_owner = btrfs_header_owner(leaf);
803 			}
804 		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
805 			   !del_item) {
806 			u32 newsize = inode->i_size - found_key.offset;
807 			newsize = btrfs_file_extent_calc_inline_size(newsize);
808 			ret = btrfs_truncate_item(trans, root, path,
809 						  newsize, 1);
810 			BUG_ON(ret);
811 		}
812 delete:
813 		if (del_item) {
814 			ret = btrfs_del_item(trans, root, path);
815 			if (ret)
816 				goto error;
817 		} else {
818 			break;
819 		}
820 		btrfs_release_path(root, path);
821 		if (found_extent) {
822 			ret = btrfs_free_extent(trans, root, extent_start,
823 						extent_num_bytes,
824 						root_owner,
825 						root_gen, inode->i_ino,
826 						found_key.offset, 0);
827 			BUG_ON(ret);
828 		}
829 	}
830 	ret = 0;
831 error:
832 	btrfs_release_path(root, path);
833 	btrfs_free_path(path);
834 	inode->i_sb->s_dirt = 1;
835 	return ret;
836 }
837 
838 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
839 			      size_t zero_start)
840 {
841 	char *kaddr;
842 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
843 	struct btrfs_root *root = BTRFS_I(inode)->root;
844 	u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
845 	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
846 	u64 existing_delalloc;
847 	u64 delalloc_start;
848 	int ret = 0;
849 
850 	WARN_ON(!PageLocked(page));
851 	set_page_extent_mapped(page);
852 
853 	lock_extent(em_tree, page_start, page_end, GFP_NOFS);
854 	delalloc_start = page_start;
855 	existing_delalloc = count_range_bits(&BTRFS_I(inode)->extent_tree,
856 					     &delalloc_start, page_end,
857 					     PAGE_CACHE_SIZE, EXTENT_DELALLOC);
858 	set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
859 			    page_end, GFP_NOFS);
860 
861 	spin_lock(&root->fs_info->delalloc_lock);
862 	root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
863 	spin_unlock(&root->fs_info->delalloc_lock);
864 
865 	if (zero_start != PAGE_CACHE_SIZE) {
866 		kaddr = kmap(page);
867 		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
868 		flush_dcache_page(page);
869 		kunmap(page);
870 	}
871 	set_page_dirty(page);
872 	unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
873 
874 	return ret;
875 }
876 
877 /*
878  * taken from block_truncate_page, but does cow as it zeros out
879  * any bytes left in the last page in the file.
880  */
881 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
882 {
883 	struct inode *inode = mapping->host;
884 	struct btrfs_root *root = BTRFS_I(inode)->root;
885 	u32 blocksize = root->sectorsize;
886 	pgoff_t index = from >> PAGE_CACHE_SHIFT;
887 	unsigned offset = from & (PAGE_CACHE_SIZE-1);
888 	struct page *page;
889 	int ret = 0;
890 	u64 page_start;
891 
892 	if ((offset & (blocksize - 1)) == 0)
893 		goto out;
894 
895 	down_read(&root->snap_sem);
896 	ret = -ENOMEM;
897 	page = grab_cache_page(mapping, index);
898 	if (!page)
899 		goto out;
900 	if (!PageUptodate(page)) {
901 		ret = btrfs_readpage(NULL, page);
902 		lock_page(page);
903 		if (!PageUptodate(page)) {
904 			ret = -EIO;
905 			goto out;
906 		}
907 	}
908 	page_start = (u64)page->index << PAGE_CACHE_SHIFT;
909 
910 	ret = btrfs_cow_one_page(inode, page, offset);
911 
912 	unlock_page(page);
913 	page_cache_release(page);
914 	up_read(&BTRFS_I(inode)->root->snap_sem);
915 out:
916 	return ret;
917 }
918 
919 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
920 {
921 	struct inode *inode = dentry->d_inode;
922 	int err;
923 
924 	err = inode_change_ok(inode, attr);
925 	if (err)
926 		return err;
927 
928 	if (S_ISREG(inode->i_mode) &&
929 	    attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
930 		struct btrfs_trans_handle *trans;
931 		struct btrfs_root *root = BTRFS_I(inode)->root;
932 		struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
933 
934 		u64 mask = root->sectorsize - 1;
935 		u64 pos = (inode->i_size + mask) & ~mask;
936 		u64 block_end = attr->ia_size | mask;
937 		u64 hole_size;
938 		u64 alloc_hint = 0;
939 
940 		if (attr->ia_size <= pos)
941 			goto out;
942 
943 		mutex_lock(&root->fs_info->fs_mutex);
944 		err = btrfs_check_free_space(root, 1, 0);
945 		mutex_unlock(&root->fs_info->fs_mutex);
946 		if (err)
947 			goto fail;
948 
949 		btrfs_truncate_page(inode->i_mapping, inode->i_size);
950 
951 		lock_extent(em_tree, pos, block_end, GFP_NOFS);
952 		hole_size = (attr->ia_size - pos + mask) & ~mask;
953 
954 		mutex_lock(&root->fs_info->fs_mutex);
955 		trans = btrfs_start_transaction(root, 1);
956 		btrfs_set_trans_block_group(trans, inode);
957 		err = btrfs_drop_extents(trans, root, inode,
958 					 pos, pos + hole_size, pos,
959 					 &alloc_hint);
960 
961 		if (alloc_hint != EXTENT_MAP_INLINE) {
962 			err = btrfs_insert_file_extent(trans, root,
963 						       inode->i_ino,
964 						       pos, 0, 0, hole_size);
965 		}
966 		btrfs_end_transaction(trans, root);
967 		mutex_unlock(&root->fs_info->fs_mutex);
968 		unlock_extent(em_tree, pos, block_end, GFP_NOFS);
969 		if (err)
970 			return err;
971 	}
972 out:
973 	err = inode_setattr(inode, attr);
974 fail:
975 	return err;
976 }
977 void btrfs_delete_inode(struct inode *inode)
978 {
979 	struct btrfs_trans_handle *trans;
980 	struct btrfs_root *root = BTRFS_I(inode)->root;
981 	unsigned long nr;
982 	int ret;
983 
984 	truncate_inode_pages(&inode->i_data, 0);
985 	if (is_bad_inode(inode)) {
986 		goto no_delete;
987 	}
988 
989 	inode->i_size = 0;
990 	mutex_lock(&root->fs_info->fs_mutex);
991 	trans = btrfs_start_transaction(root, 1);
992 
993 	btrfs_set_trans_block_group(trans, inode);
994 	ret = btrfs_truncate_in_trans(trans, root, inode);
995 	if (ret)
996 		goto no_delete_lock;
997 	ret = btrfs_delete_xattrs(trans, root, inode);
998 	if (ret)
999 		goto no_delete_lock;
1000 	ret = btrfs_free_inode(trans, root, inode);
1001 	if (ret)
1002 		goto no_delete_lock;
1003 	nr = trans->blocks_used;
1004 
1005 	btrfs_end_transaction(trans, root);
1006 	mutex_unlock(&root->fs_info->fs_mutex);
1007 	btrfs_btree_balance_dirty(root, nr);
1008 	return;
1009 
1010 no_delete_lock:
1011 	nr = trans->blocks_used;
1012 	btrfs_end_transaction(trans, root);
1013 	mutex_unlock(&root->fs_info->fs_mutex);
1014 	btrfs_btree_balance_dirty(root, nr);
1015 no_delete:
1016 	clear_inode(inode);
1017 }
1018 
1019 /*
1020  * this returns the key found in the dir entry in the location pointer.
1021  * If no dir entries were found, location->objectid is 0.
1022  */
1023 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1024 			       struct btrfs_key *location)
1025 {
1026 	const char *name = dentry->d_name.name;
1027 	int namelen = dentry->d_name.len;
1028 	struct btrfs_dir_item *di;
1029 	struct btrfs_path *path;
1030 	struct btrfs_root *root = BTRFS_I(dir)->root;
1031 	int ret = 0;
1032 
1033 	if (namelen == 1 && strcmp(name, ".") == 0) {
1034 		location->objectid = dir->i_ino;
1035 		location->type = BTRFS_INODE_ITEM_KEY;
1036 		location->offset = 0;
1037 		return 0;
1038 	}
1039 	path = btrfs_alloc_path();
1040 	BUG_ON(!path);
1041 
1042 	if (namelen == 2 && strcmp(name, "..") == 0) {
1043 		struct btrfs_key key;
1044 		struct extent_buffer *leaf;
1045 		u32 nritems;
1046 		int slot;
1047 
1048 		key.objectid = dir->i_ino;
1049 		btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1050 		key.offset = 0;
1051 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1052 		BUG_ON(ret == 0);
1053 		ret = 0;
1054 
1055 		leaf = path->nodes[0];
1056 		slot = path->slots[0];
1057 		nritems = btrfs_header_nritems(leaf);
1058 		if (slot >= nritems)
1059 			goto out_err;
1060 
1061 		btrfs_item_key_to_cpu(leaf, &key, slot);
1062 		if (key.objectid != dir->i_ino ||
1063 		    key.type != BTRFS_INODE_REF_KEY) {
1064 			goto out_err;
1065 		}
1066 		location->objectid = key.offset;
1067 		location->type = BTRFS_INODE_ITEM_KEY;
1068 		location->offset = 0;
1069 		goto out;
1070 	}
1071 
1072 	di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1073 				    namelen, 0);
1074 	if (IS_ERR(di))
1075 		ret = PTR_ERR(di);
1076 	if (!di || IS_ERR(di)) {
1077 		goto out_err;
1078 	}
1079 	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1080 out:
1081 	btrfs_free_path(path);
1082 	return ret;
1083 out_err:
1084 	location->objectid = 0;
1085 	goto out;
1086 }
1087 
1088 /*
1089  * when we hit a tree root in a directory, the btrfs part of the inode
1090  * needs to be changed to reflect the root directory of the tree root.  This
1091  * is kind of like crossing a mount point.
1092  */
1093 static int fixup_tree_root_location(struct btrfs_root *root,
1094 			     struct btrfs_key *location,
1095 			     struct btrfs_root **sub_root,
1096 			     struct dentry *dentry)
1097 {
1098 	struct btrfs_path *path;
1099 	struct btrfs_root_item *ri;
1100 
1101 	if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1102 		return 0;
1103 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1104 		return 0;
1105 
1106 	path = btrfs_alloc_path();
1107 	BUG_ON(!path);
1108 	mutex_lock(&root->fs_info->fs_mutex);
1109 
1110 	*sub_root = btrfs_read_fs_root(root->fs_info, location,
1111 					dentry->d_name.name,
1112 					dentry->d_name.len);
1113 	if (IS_ERR(*sub_root))
1114 		return PTR_ERR(*sub_root);
1115 
1116 	ri = &(*sub_root)->root_item;
1117 	location->objectid = btrfs_root_dirid(ri);
1118 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1119 	location->offset = 0;
1120 
1121 	btrfs_free_path(path);
1122 	mutex_unlock(&root->fs_info->fs_mutex);
1123 	return 0;
1124 }
1125 
1126 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1127 {
1128 	struct btrfs_iget_args *args = p;
1129 	inode->i_ino = args->ino;
1130 	BTRFS_I(inode)->root = args->root;
1131 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1132 			     inode->i_mapping, GFP_NOFS);
1133 	return 0;
1134 }
1135 
1136 static int btrfs_find_actor(struct inode *inode, void *opaque)
1137 {
1138 	struct btrfs_iget_args *args = opaque;
1139 	return (args->ino == inode->i_ino &&
1140 		args->root == BTRFS_I(inode)->root);
1141 }
1142 
1143 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1144 				struct btrfs_root *root)
1145 {
1146 	struct inode *inode;
1147 	struct btrfs_iget_args args;
1148 	args.ino = objectid;
1149 	args.root = root;
1150 
1151 	inode = iget5_locked(s, objectid, btrfs_find_actor,
1152 			     btrfs_init_locked_inode,
1153 			     (void *)&args);
1154 	return inode;
1155 }
1156 
1157 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1158 				   struct nameidata *nd)
1159 {
1160 	struct inode * inode;
1161 	struct btrfs_inode *bi = BTRFS_I(dir);
1162 	struct btrfs_root *root = bi->root;
1163 	struct btrfs_root *sub_root = root;
1164 	struct btrfs_key location;
1165 	int ret;
1166 
1167 	if (dentry->d_name.len > BTRFS_NAME_LEN)
1168 		return ERR_PTR(-ENAMETOOLONG);
1169 
1170 	mutex_lock(&root->fs_info->fs_mutex);
1171 	ret = btrfs_inode_by_name(dir, dentry, &location);
1172 	mutex_unlock(&root->fs_info->fs_mutex);
1173 
1174 	if (ret < 0)
1175 		return ERR_PTR(ret);
1176 
1177 	inode = NULL;
1178 	if (location.objectid) {
1179 		ret = fixup_tree_root_location(root, &location, &sub_root,
1180 						dentry);
1181 		if (ret < 0)
1182 			return ERR_PTR(ret);
1183 		if (ret > 0)
1184 			return ERR_PTR(-ENOENT);
1185 		inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1186 					  sub_root);
1187 		if (!inode)
1188 			return ERR_PTR(-EACCES);
1189 		if (inode->i_state & I_NEW) {
1190 			/* the inode and parent dir are two different roots */
1191 			if (sub_root != root) {
1192 				igrab(inode);
1193 				sub_root->inode = inode;
1194 			}
1195 			BTRFS_I(inode)->root = sub_root;
1196 			memcpy(&BTRFS_I(inode)->location, &location,
1197 			       sizeof(location));
1198 			btrfs_read_locked_inode(inode);
1199 			unlock_new_inode(inode);
1200 		}
1201 	}
1202 	return d_splice_alias(inode, dentry);
1203 }
1204 
1205 static unsigned char btrfs_filetype_table[] = {
1206 	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1207 };
1208 
1209 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1210 {
1211 	struct inode *inode = filp->f_dentry->d_inode;
1212 	struct btrfs_root *root = BTRFS_I(inode)->root;
1213 	struct btrfs_item *item;
1214 	struct btrfs_dir_item *di;
1215 	struct btrfs_key key;
1216 	struct btrfs_key found_key;
1217 	struct btrfs_path *path;
1218 	int ret;
1219 	u32 nritems;
1220 	struct extent_buffer *leaf;
1221 	int slot;
1222 	int advance;
1223 	unsigned char d_type;
1224 	int over = 0;
1225 	u32 di_cur;
1226 	u32 di_total;
1227 	u32 di_len;
1228 	int key_type = BTRFS_DIR_INDEX_KEY;
1229 	char tmp_name[32];
1230 	char *name_ptr;
1231 	int name_len;
1232 
1233 	/* FIXME, use a real flag for deciding about the key type */
1234 	if (root->fs_info->tree_root == root)
1235 		key_type = BTRFS_DIR_ITEM_KEY;
1236 
1237 	/* special case for "." */
1238 	if (filp->f_pos == 0) {
1239 		over = filldir(dirent, ".", 1,
1240 			       1, inode->i_ino,
1241 			       DT_DIR);
1242 		if (over)
1243 			return 0;
1244 		filp->f_pos = 1;
1245 	}
1246 
1247 	mutex_lock(&root->fs_info->fs_mutex);
1248 	key.objectid = inode->i_ino;
1249 	path = btrfs_alloc_path();
1250 	path->reada = 2;
1251 
1252 	/* special case for .., just use the back ref */
1253 	if (filp->f_pos == 1) {
1254 		btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1255 		key.offset = 0;
1256 		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1257 		BUG_ON(ret == 0);
1258 		leaf = path->nodes[0];
1259 		slot = path->slots[0];
1260 		nritems = btrfs_header_nritems(leaf);
1261 		if (slot >= nritems) {
1262 			btrfs_release_path(root, path);
1263 			goto read_dir_items;
1264 		}
1265 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1266 		btrfs_release_path(root, path);
1267 		if (found_key.objectid != key.objectid ||
1268 		    found_key.type != BTRFS_INODE_REF_KEY)
1269 			goto read_dir_items;
1270 		over = filldir(dirent, "..", 2,
1271 			       2, found_key.offset, DT_DIR);
1272 		if (over)
1273 			goto nopos;
1274 		filp->f_pos = 2;
1275 	}
1276 
1277 read_dir_items:
1278 	btrfs_set_key_type(&key, key_type);
1279 	key.offset = filp->f_pos;
1280 
1281 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1282 	if (ret < 0)
1283 		goto err;
1284 	advance = 0;
1285 	while(1) {
1286 		leaf = path->nodes[0];
1287 		nritems = btrfs_header_nritems(leaf);
1288 		slot = path->slots[0];
1289 		if (advance || slot >= nritems) {
1290 			if (slot >= nritems -1) {
1291 				ret = btrfs_next_leaf(root, path);
1292 				if (ret)
1293 					break;
1294 				leaf = path->nodes[0];
1295 				nritems = btrfs_header_nritems(leaf);
1296 				slot = path->slots[0];
1297 			} else {
1298 				slot++;
1299 				path->slots[0]++;
1300 			}
1301 		}
1302 		advance = 1;
1303 		item = btrfs_item_nr(leaf, slot);
1304 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1305 
1306 		if (found_key.objectid != key.objectid)
1307 			break;
1308 		if (btrfs_key_type(&found_key) != key_type)
1309 			break;
1310 		if (found_key.offset < filp->f_pos)
1311 			continue;
1312 
1313 		filp->f_pos = found_key.offset;
1314 		advance = 1;
1315 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1316 		di_cur = 0;
1317 		di_total = btrfs_item_size(leaf, item);
1318 		while(di_cur < di_total) {
1319 			struct btrfs_key location;
1320 
1321 			name_len = btrfs_dir_name_len(leaf, di);
1322 			if (name_len < 32) {
1323 				name_ptr = tmp_name;
1324 			} else {
1325 				name_ptr = kmalloc(name_len, GFP_NOFS);
1326 				BUG_ON(!name_ptr);
1327 			}
1328 			read_extent_buffer(leaf, name_ptr,
1329 					   (unsigned long)(di + 1), name_len);
1330 
1331 			d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1332 			btrfs_dir_item_key_to_cpu(leaf, di, &location);
1333 
1334 			over = filldir(dirent, name_ptr, name_len,
1335 				       found_key.offset,
1336 				       location.objectid,
1337 				       d_type);
1338 
1339 			if (name_ptr != tmp_name)
1340 				kfree(name_ptr);
1341 
1342 			if (over)
1343 				goto nopos;
1344 			di_len = btrfs_dir_name_len(leaf, di) +
1345 				btrfs_dir_data_len(leaf, di) +sizeof(*di);
1346 			di_cur += di_len;
1347 			di = (struct btrfs_dir_item *)((char *)di + di_len);
1348 		}
1349 	}
1350 	filp->f_pos++;
1351 nopos:
1352 	ret = 0;
1353 err:
1354 	btrfs_release_path(root, path);
1355 	btrfs_free_path(path);
1356 	mutex_unlock(&root->fs_info->fs_mutex);
1357 	return ret;
1358 }
1359 
1360 int btrfs_write_inode(struct inode *inode, int wait)
1361 {
1362 	struct btrfs_root *root = BTRFS_I(inode)->root;
1363 	struct btrfs_trans_handle *trans;
1364 	int ret = 0;
1365 
1366 	if (wait) {
1367 		mutex_lock(&root->fs_info->fs_mutex);
1368 		trans = btrfs_start_transaction(root, 1);
1369 		btrfs_set_trans_block_group(trans, inode);
1370 		ret = btrfs_commit_transaction(trans, root);
1371 		mutex_unlock(&root->fs_info->fs_mutex);
1372 	}
1373 	return ret;
1374 }
1375 
1376 /*
1377  * This is somewhat expensive, updating the tree every time the
1378  * inode changes.  But, it is most likely to find the inode in cache.
1379  * FIXME, needs more benchmarking...there are no reasons other than performance
1380  * to keep or drop this code.
1381  */
1382 void btrfs_dirty_inode(struct inode *inode)
1383 {
1384 	struct btrfs_root *root = BTRFS_I(inode)->root;
1385 	struct btrfs_trans_handle *trans;
1386 
1387 	mutex_lock(&root->fs_info->fs_mutex);
1388 	trans = btrfs_start_transaction(root, 1);
1389 	btrfs_set_trans_block_group(trans, inode);
1390 	btrfs_update_inode(trans, root, inode);
1391 	btrfs_end_transaction(trans, root);
1392 	mutex_unlock(&root->fs_info->fs_mutex);
1393 }
1394 
1395 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1396 				     struct btrfs_root *root,
1397 				     u64 objectid,
1398 				     struct btrfs_block_group_cache *group,
1399 				     int mode)
1400 {
1401 	struct inode *inode;
1402 	struct btrfs_inode_item *inode_item;
1403 	struct btrfs_key *location;
1404 	struct btrfs_path *path;
1405 	int ret;
1406 	int owner;
1407 
1408 	path = btrfs_alloc_path();
1409 	BUG_ON(!path);
1410 
1411 	inode = new_inode(root->fs_info->sb);
1412 	if (!inode)
1413 		return ERR_PTR(-ENOMEM);
1414 
1415 	extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1416 			     inode->i_mapping, GFP_NOFS);
1417 	BTRFS_I(inode)->root = root;
1418 
1419 	if (mode & S_IFDIR)
1420 		owner = 0;
1421 	else
1422 		owner = 1;
1423 	group = btrfs_find_block_group(root, group, 0, 0, owner);
1424 	BTRFS_I(inode)->block_group = group;
1425 
1426 	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1427 	if (ret)
1428 		goto fail;
1429 
1430 	inode->i_uid = current->fsuid;
1431 	inode->i_gid = current->fsgid;
1432 	inode->i_mode = mode;
1433 	inode->i_ino = objectid;
1434 	inode->i_blocks = 0;
1435 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1436 	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1437 				  struct btrfs_inode_item);
1438 	fill_inode_item(path->nodes[0], inode_item, inode);
1439 	btrfs_mark_buffer_dirty(path->nodes[0]);
1440 	btrfs_free_path(path);
1441 
1442 	location = &BTRFS_I(inode)->location;
1443 	location->objectid = objectid;
1444 	location->offset = 0;
1445 	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1446 
1447 	insert_inode_hash(inode);
1448 	return inode;
1449 fail:
1450 	btrfs_free_path(path);
1451 	return ERR_PTR(ret);
1452 }
1453 
1454 static inline u8 btrfs_inode_type(struct inode *inode)
1455 {
1456 	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1457 }
1458 
1459 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1460 			    struct dentry *dentry, struct inode *inode)
1461 {
1462 	int ret;
1463 	struct btrfs_key key;
1464 	struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1465 	struct inode *parent_inode;
1466 
1467 	key.objectid = inode->i_ino;
1468 	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1469 	key.offset = 0;
1470 
1471 	ret = btrfs_insert_dir_item(trans, root,
1472 				    dentry->d_name.name, dentry->d_name.len,
1473 				    dentry->d_parent->d_inode->i_ino,
1474 				    &key, btrfs_inode_type(inode));
1475 	if (ret == 0) {
1476 		ret = btrfs_insert_inode_ref(trans, root,
1477 				     dentry->d_name.name,
1478 				     dentry->d_name.len,
1479 				     inode->i_ino,
1480 				     dentry->d_parent->d_inode->i_ino);
1481 		parent_inode = dentry->d_parent->d_inode;
1482 		parent_inode->i_size += dentry->d_name.len * 2;
1483 		parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1484 		ret = btrfs_update_inode(trans, root,
1485 					 dentry->d_parent->d_inode);
1486 	}
1487 	return ret;
1488 }
1489 
1490 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1491 			    struct dentry *dentry, struct inode *inode)
1492 {
1493 	int err = btrfs_add_link(trans, dentry, inode);
1494 	if (!err) {
1495 		d_instantiate(dentry, inode);
1496 		return 0;
1497 	}
1498 	if (err > 0)
1499 		err = -EEXIST;
1500 	return err;
1501 }
1502 
1503 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1504 			int mode, dev_t rdev)
1505 {
1506 	struct btrfs_trans_handle *trans;
1507 	struct btrfs_root *root = BTRFS_I(dir)->root;
1508 	struct inode *inode = NULL;
1509 	int err;
1510 	int drop_inode = 0;
1511 	u64 objectid;
1512 	unsigned long nr = 0;
1513 
1514 	if (!new_valid_dev(rdev))
1515 		return -EINVAL;
1516 
1517 	mutex_lock(&root->fs_info->fs_mutex);
1518 	err = btrfs_check_free_space(root, 1, 0);
1519 	if (err)
1520 		goto fail;
1521 
1522 	trans = btrfs_start_transaction(root, 1);
1523 	btrfs_set_trans_block_group(trans, dir);
1524 
1525 	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1526 	if (err) {
1527 		err = -ENOSPC;
1528 		goto out_unlock;
1529 	}
1530 
1531 	inode = btrfs_new_inode(trans, root, objectid,
1532 				BTRFS_I(dir)->block_group, mode);
1533 	err = PTR_ERR(inode);
1534 	if (IS_ERR(inode))
1535 		goto out_unlock;
1536 
1537 	btrfs_set_trans_block_group(trans, inode);
1538 	err = btrfs_add_nondir(trans, dentry, inode);
1539 	if (err)
1540 		drop_inode = 1;
1541 	else {
1542 		inode->i_op = &btrfs_special_inode_operations;
1543 		init_special_inode(inode, inode->i_mode, rdev);
1544 		btrfs_update_inode(trans, root, inode);
1545 	}
1546 	dir->i_sb->s_dirt = 1;
1547 	btrfs_update_inode_block_group(trans, inode);
1548 	btrfs_update_inode_block_group(trans, dir);
1549 out_unlock:
1550 	nr = trans->blocks_used;
1551 	btrfs_end_transaction(trans, root);
1552 fail:
1553 	mutex_unlock(&root->fs_info->fs_mutex);
1554 
1555 	if (drop_inode) {
1556 		inode_dec_link_count(inode);
1557 		iput(inode);
1558 	}
1559 	btrfs_btree_balance_dirty(root, nr);
1560 	return err;
1561 }
1562 
1563 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1564 			int mode, struct nameidata *nd)
1565 {
1566 	struct btrfs_trans_handle *trans;
1567 	struct btrfs_root *root = BTRFS_I(dir)->root;
1568 	struct inode *inode = NULL;
1569 	int err;
1570 	int drop_inode = 0;
1571 	unsigned long nr = 0;
1572 	u64 objectid;
1573 
1574 	mutex_lock(&root->fs_info->fs_mutex);
1575 	err = btrfs_check_free_space(root, 1, 0);
1576 	if (err)
1577 		goto fail;
1578 	trans = btrfs_start_transaction(root, 1);
1579 	btrfs_set_trans_block_group(trans, dir);
1580 
1581 	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1582 	if (err) {
1583 		err = -ENOSPC;
1584 		goto out_unlock;
1585 	}
1586 
1587 	inode = btrfs_new_inode(trans, root, objectid,
1588 				BTRFS_I(dir)->block_group, mode);
1589 	err = PTR_ERR(inode);
1590 	if (IS_ERR(inode))
1591 		goto out_unlock;
1592 
1593 	btrfs_set_trans_block_group(trans, inode);
1594 	err = btrfs_add_nondir(trans, dentry, inode);
1595 	if (err)
1596 		drop_inode = 1;
1597 	else {
1598 		inode->i_mapping->a_ops = &btrfs_aops;
1599 		inode->i_fop = &btrfs_file_operations;
1600 		inode->i_op = &btrfs_file_inode_operations;
1601 		extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1602 				     inode->i_mapping, GFP_NOFS);
1603 		BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1604 	}
1605 	dir->i_sb->s_dirt = 1;
1606 	btrfs_update_inode_block_group(trans, inode);
1607 	btrfs_update_inode_block_group(trans, dir);
1608 out_unlock:
1609 	nr = trans->blocks_used;
1610 	btrfs_end_transaction(trans, root);
1611 fail:
1612 	mutex_unlock(&root->fs_info->fs_mutex);
1613 
1614 	if (drop_inode) {
1615 		inode_dec_link_count(inode);
1616 		iput(inode);
1617 	}
1618 	btrfs_btree_balance_dirty(root, nr);
1619 	return err;
1620 }
1621 
1622 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1623 		      struct dentry *dentry)
1624 {
1625 	struct btrfs_trans_handle *trans;
1626 	struct btrfs_root *root = BTRFS_I(dir)->root;
1627 	struct inode *inode = old_dentry->d_inode;
1628 	unsigned long nr = 0;
1629 	int err;
1630 	int drop_inode = 0;
1631 
1632 	if (inode->i_nlink == 0)
1633 		return -ENOENT;
1634 
1635 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1636 	inode->i_nlink++;
1637 #else
1638 	inc_nlink(inode);
1639 #endif
1640 	mutex_lock(&root->fs_info->fs_mutex);
1641 	err = btrfs_check_free_space(root, 1, 0);
1642 	if (err)
1643 		goto fail;
1644 	trans = btrfs_start_transaction(root, 1);
1645 
1646 	btrfs_set_trans_block_group(trans, dir);
1647 	atomic_inc(&inode->i_count);
1648 	err = btrfs_add_nondir(trans, dentry, inode);
1649 
1650 	if (err)
1651 		drop_inode = 1;
1652 
1653 	dir->i_sb->s_dirt = 1;
1654 	btrfs_update_inode_block_group(trans, dir);
1655 	err = btrfs_update_inode(trans, root, inode);
1656 
1657 	if (err)
1658 		drop_inode = 1;
1659 
1660 	nr = trans->blocks_used;
1661 	btrfs_end_transaction(trans, root);
1662 fail:
1663 	mutex_unlock(&root->fs_info->fs_mutex);
1664 
1665 	if (drop_inode) {
1666 		inode_dec_link_count(inode);
1667 		iput(inode);
1668 	}
1669 	btrfs_btree_balance_dirty(root, nr);
1670 	return err;
1671 }
1672 
1673 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1674 {
1675 	struct inode *inode;
1676 	struct btrfs_trans_handle *trans;
1677 	struct btrfs_root *root = BTRFS_I(dir)->root;
1678 	int err = 0;
1679 	int drop_on_err = 0;
1680 	u64 objectid;
1681 	unsigned long nr = 1;
1682 
1683 	mutex_lock(&root->fs_info->fs_mutex);
1684 	err = btrfs_check_free_space(root, 1, 0);
1685 	if (err)
1686 		goto out_unlock;
1687 
1688 	trans = btrfs_start_transaction(root, 1);
1689 	btrfs_set_trans_block_group(trans, dir);
1690 
1691 	if (IS_ERR(trans)) {
1692 		err = PTR_ERR(trans);
1693 		goto out_unlock;
1694 	}
1695 
1696 	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1697 	if (err) {
1698 		err = -ENOSPC;
1699 		goto out_unlock;
1700 	}
1701 
1702 	inode = btrfs_new_inode(trans, root, objectid,
1703 				BTRFS_I(dir)->block_group, S_IFDIR | mode);
1704 	if (IS_ERR(inode)) {
1705 		err = PTR_ERR(inode);
1706 		goto out_fail;
1707 	}
1708 
1709 	drop_on_err = 1;
1710 	inode->i_op = &btrfs_dir_inode_operations;
1711 	inode->i_fop = &btrfs_dir_file_operations;
1712 	btrfs_set_trans_block_group(trans, inode);
1713 
1714 	inode->i_size = 0;
1715 	err = btrfs_update_inode(trans, root, inode);
1716 	if (err)
1717 		goto out_fail;
1718 
1719 	err = btrfs_add_link(trans, dentry, inode);
1720 	if (err)
1721 		goto out_fail;
1722 
1723 	d_instantiate(dentry, inode);
1724 	drop_on_err = 0;
1725 	dir->i_sb->s_dirt = 1;
1726 	btrfs_update_inode_block_group(trans, inode);
1727 	btrfs_update_inode_block_group(trans, dir);
1728 
1729 out_fail:
1730 	nr = trans->blocks_used;
1731 	btrfs_end_transaction(trans, root);
1732 
1733 out_unlock:
1734 	mutex_unlock(&root->fs_info->fs_mutex);
1735 	if (drop_on_err)
1736 		iput(inode);
1737 	btrfs_btree_balance_dirty(root, nr);
1738 	return err;
1739 }
1740 
1741 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1742 				    size_t page_offset, u64 start, u64 end,
1743 				    int create)
1744 {
1745 	int ret;
1746 	int err = 0;
1747 	u64 bytenr;
1748 	u64 extent_start = 0;
1749 	u64 extent_end = 0;
1750 	u64 objectid = inode->i_ino;
1751 	u32 found_type;
1752 	int failed_insert = 0;
1753 	struct btrfs_path *path;
1754 	struct btrfs_root *root = BTRFS_I(inode)->root;
1755 	struct btrfs_file_extent_item *item;
1756 	struct extent_buffer *leaf;
1757 	struct btrfs_key found_key;
1758 	struct extent_map *em = NULL;
1759 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1760 	struct btrfs_trans_handle *trans = NULL;
1761 
1762 	path = btrfs_alloc_path();
1763 	BUG_ON(!path);
1764 	mutex_lock(&root->fs_info->fs_mutex);
1765 
1766 again:
1767 	em = lookup_extent_mapping(em_tree, start, end);
1768 	if (em) {
1769 		goto out;
1770 	}
1771 	if (!em) {
1772 		em = alloc_extent_map(GFP_NOFS);
1773 		if (!em) {
1774 			err = -ENOMEM;
1775 			goto out;
1776 		}
1777 		em->start = EXTENT_MAP_HOLE;
1778 		em->end = EXTENT_MAP_HOLE;
1779 	}
1780 	em->bdev = inode->i_sb->s_bdev;
1781 	ret = btrfs_lookup_file_extent(trans, root, path,
1782 				       objectid, start, trans != NULL);
1783 	if (ret < 0) {
1784 		err = ret;
1785 		goto out;
1786 	}
1787 
1788 	if (ret != 0) {
1789 		if (path->slots[0] == 0)
1790 			goto not_found;
1791 		path->slots[0]--;
1792 	}
1793 
1794 	leaf = path->nodes[0];
1795 	item = btrfs_item_ptr(leaf, path->slots[0],
1796 			      struct btrfs_file_extent_item);
1797 	/* are we inside the extent that was found? */
1798 	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1799 	found_type = btrfs_key_type(&found_key);
1800 	if (found_key.objectid != objectid ||
1801 	    found_type != BTRFS_EXTENT_DATA_KEY) {
1802 		goto not_found;
1803 	}
1804 
1805 	found_type = btrfs_file_extent_type(leaf, item);
1806 	extent_start = found_key.offset;
1807 	if (found_type == BTRFS_FILE_EXTENT_REG) {
1808 		extent_end = extent_start +
1809 		       btrfs_file_extent_num_bytes(leaf, item);
1810 		err = 0;
1811 		if (start < extent_start || start >= extent_end) {
1812 			em->start = start;
1813 			if (start < extent_start) {
1814 				if (end < extent_start)
1815 					goto not_found;
1816 				em->end = extent_end - 1;
1817 			} else {
1818 				em->end = end;
1819 			}
1820 			goto not_found_em;
1821 		}
1822 		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1823 		if (bytenr == 0) {
1824 			em->start = extent_start;
1825 			em->end = extent_end - 1;
1826 			em->block_start = EXTENT_MAP_HOLE;
1827 			em->block_end = EXTENT_MAP_HOLE;
1828 			goto insert;
1829 		}
1830 		bytenr += btrfs_file_extent_offset(leaf, item);
1831 		em->block_start = bytenr;
1832 		em->block_end = em->block_start +
1833 			btrfs_file_extent_num_bytes(leaf, item) - 1;
1834 		em->start = extent_start;
1835 		em->end = extent_end - 1;
1836 		goto insert;
1837 	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1838 		unsigned long ptr;
1839 		char *map;
1840 		size_t size;
1841 		size_t extent_offset;
1842 		size_t copy_size;
1843 
1844 		size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1845 						    path->slots[0]));
1846 		extent_end = (extent_start + size - 1) |
1847 			((u64)root->sectorsize - 1);
1848 		if (start < extent_start || start >= extent_end) {
1849 			em->start = start;
1850 			if (start < extent_start) {
1851 				if (end < extent_start)
1852 					goto not_found;
1853 				em->end = extent_end;
1854 			} else {
1855 				em->end = end;
1856 			}
1857 			goto not_found_em;
1858 		}
1859 		em->block_start = EXTENT_MAP_INLINE;
1860 		em->block_end = EXTENT_MAP_INLINE;
1861 
1862 		if (!page) {
1863 			em->start = extent_start;
1864 			em->end = extent_start + size - 1;
1865 			goto out;
1866 		}
1867 
1868 		extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1869 			extent_start + page_offset;
1870 		copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1871 				size - extent_offset);
1872 		em->start = extent_start + extent_offset;
1873 		em->end = (em->start + copy_size -1) |
1874 			((u64)root->sectorsize -1);
1875 		map = kmap(page);
1876 		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1877 		if (create == 0 && !PageUptodate(page)) {
1878 			read_extent_buffer(leaf, map + page_offset, ptr,
1879 					   copy_size);
1880 			flush_dcache_page(page);
1881 		} else if (create && PageUptodate(page)) {
1882 			if (!trans) {
1883 				kunmap(page);
1884 				free_extent_map(em);
1885 				em = NULL;
1886 				btrfs_release_path(root, path);
1887 				trans = btrfs_start_transaction(root, 1);
1888 				goto again;
1889 			}
1890 			write_extent_buffer(leaf, map + page_offset, ptr,
1891 					    copy_size);
1892 			btrfs_mark_buffer_dirty(leaf);
1893 		}
1894 		kunmap(page);
1895 		set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1896 		goto insert;
1897 	} else {
1898 		printk("unkknown found_type %d\n", found_type);
1899 		WARN_ON(1);
1900 	}
1901 not_found:
1902 	em->start = start;
1903 	em->end = end;
1904 not_found_em:
1905 	em->block_start = EXTENT_MAP_HOLE;
1906 	em->block_end = EXTENT_MAP_HOLE;
1907 insert:
1908 	btrfs_release_path(root, path);
1909 	if (em->start > start || em->end < start) {
1910 		printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1911 		err = -EIO;
1912 		goto out;
1913 	}
1914 	ret = add_extent_mapping(em_tree, em);
1915 	if (ret == -EEXIST) {
1916 		free_extent_map(em);
1917 		em = NULL;
1918 		if (0 && failed_insert == 1) {
1919 			btrfs_drop_extent_cache(inode, start, end);
1920 		}
1921 		failed_insert++;
1922 		if (failed_insert > 5) {
1923 			printk("failing to insert %Lu %Lu\n", start, end);
1924 			err = -EIO;
1925 			goto out;
1926 		}
1927 		goto again;
1928 	}
1929 	err = 0;
1930 out:
1931 	btrfs_free_path(path);
1932 	if (trans) {
1933 		ret = btrfs_end_transaction(trans, root);
1934 		if (!err)
1935 			err = ret;
1936 	}
1937 	mutex_unlock(&root->fs_info->fs_mutex);
1938 	if (err) {
1939 		free_extent_map(em);
1940 		WARN_ON(1);
1941 		return ERR_PTR(err);
1942 	}
1943 	return em;
1944 }
1945 
1946 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1947 {
1948 	return extent_bmap(mapping, iblock, btrfs_get_extent);
1949 }
1950 
1951 static int btrfs_prepare_write(struct file *file, struct page *page,
1952 			       unsigned from, unsigned to)
1953 {
1954 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1955 	int err;
1956 
1957 	mutex_lock(&root->fs_info->fs_mutex);
1958 	err = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
1959 	mutex_lock(&root->fs_info->fs_mutex);
1960 	if (err)
1961 		return -ENOSPC;
1962 
1963 	return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1964 				    page->mapping->host, page, from, to,
1965 				    btrfs_get_extent);
1966 }
1967 
1968 int btrfs_readpage(struct file *file, struct page *page)
1969 {
1970 	struct extent_map_tree *tree;
1971 	tree = &BTRFS_I(page->mapping->host)->extent_tree;
1972 	return extent_read_full_page(tree, page, btrfs_get_extent);
1973 }
1974 
1975 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1976 {
1977 	struct extent_map_tree *tree;
1978 
1979 
1980 	if (current->flags & PF_MEMALLOC) {
1981 		redirty_page_for_writepage(wbc, page);
1982 		unlock_page(page);
1983 		return 0;
1984 	}
1985 	tree = &BTRFS_I(page->mapping->host)->extent_tree;
1986 	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1987 }
1988 
1989 static int btrfs_writepages(struct address_space *mapping,
1990 			    struct writeback_control *wbc)
1991 {
1992 	struct extent_map_tree *tree;
1993 	tree = &BTRFS_I(mapping->host)->extent_tree;
1994 	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
1995 }
1996 
1997 static int
1998 btrfs_readpages(struct file *file, struct address_space *mapping,
1999 		struct list_head *pages, unsigned nr_pages)
2000 {
2001 	struct extent_map_tree *tree;
2002 	tree = &BTRFS_I(mapping->host)->extent_tree;
2003 	return extent_readpages(tree, mapping, pages, nr_pages,
2004 				btrfs_get_extent);
2005 }
2006 
2007 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2008 {
2009 	struct extent_map_tree *tree;
2010 	int ret;
2011 
2012 	tree = &BTRFS_I(page->mapping->host)->extent_tree;
2013 	ret = try_release_extent_mapping(tree, page);
2014 	if (ret == 1) {
2015 		ClearPagePrivate(page);
2016 		set_page_private(page, 0);
2017 		page_cache_release(page);
2018 	}
2019 	return ret;
2020 }
2021 
2022 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2023 {
2024 	struct extent_map_tree *tree;
2025 
2026 	tree = &BTRFS_I(page->mapping->host)->extent_tree;
2027 	extent_invalidatepage(tree, page, offset);
2028 	btrfs_releasepage(page, GFP_NOFS);
2029 }
2030 
2031 /*
2032  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2033  * called from a page fault handler when a page is first dirtied. Hence we must
2034  * be careful to check for EOF conditions here. We set the page up correctly
2035  * for a written page which means we get ENOSPC checking when writing into
2036  * holes and correct delalloc and unwritten extent mapping on filesystems that
2037  * support these features.
2038  *
2039  * We are not allowed to take the i_mutex here so we have to play games to
2040  * protect against truncate races as the page could now be beyond EOF.  Because
2041  * vmtruncate() writes the inode size before removing pages, once we have the
2042  * page lock we can determine safely if the page is beyond EOF. If it is not
2043  * beyond EOF, then the page is guaranteed safe against truncation until we
2044  * unlock the page.
2045  */
2046 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2047 {
2048 	struct inode *inode = fdentry(vma->vm_file)->d_inode;
2049 	struct btrfs_root *root = BTRFS_I(inode)->root;
2050 	unsigned long end;
2051 	loff_t size;
2052 	int ret;
2053 	u64 page_start;
2054 
2055 	mutex_lock(&root->fs_info->fs_mutex);
2056 	ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2057 	mutex_lock(&root->fs_info->fs_mutex);
2058 	if (ret)
2059 		goto out;
2060 
2061 	ret = -EINVAL;
2062 
2063 	down_read(&BTRFS_I(inode)->root->snap_sem);
2064 	lock_page(page);
2065 	wait_on_page_writeback(page);
2066 	size = i_size_read(inode);
2067 	page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2068 
2069 	if ((page->mapping != inode->i_mapping) ||
2070 	    (page_start > size)) {
2071 		/* page got truncated out from underneath us */
2072 		goto out_unlock;
2073 	}
2074 
2075 	/* page is wholly or partially inside EOF */
2076 	if (page_start + PAGE_CACHE_SIZE > size)
2077 		end = size & ~PAGE_CACHE_MASK;
2078 	else
2079 		end = PAGE_CACHE_SIZE;
2080 
2081 	ret = btrfs_cow_one_page(inode, page, end);
2082 
2083 out_unlock:
2084 	up_read(&BTRFS_I(inode)->root->snap_sem);
2085 	unlock_page(page);
2086 out:
2087 	return ret;
2088 }
2089 
2090 static void btrfs_truncate(struct inode *inode)
2091 {
2092 	struct btrfs_root *root = BTRFS_I(inode)->root;
2093 	int ret;
2094 	struct btrfs_trans_handle *trans;
2095 	unsigned long nr;
2096 
2097 	if (!S_ISREG(inode->i_mode))
2098 		return;
2099 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2100 		return;
2101 
2102 	btrfs_truncate_page(inode->i_mapping, inode->i_size);
2103 
2104 	mutex_lock(&root->fs_info->fs_mutex);
2105 	trans = btrfs_start_transaction(root, 1);
2106 	btrfs_set_trans_block_group(trans, inode);
2107 
2108 	/* FIXME, add redo link to tree so we don't leak on crash */
2109 	ret = btrfs_truncate_in_trans(trans, root, inode);
2110 	btrfs_update_inode(trans, root, inode);
2111 	nr = trans->blocks_used;
2112 
2113 	ret = btrfs_end_transaction(trans, root);
2114 	BUG_ON(ret);
2115 	mutex_unlock(&root->fs_info->fs_mutex);
2116 	btrfs_btree_balance_dirty(root, nr);
2117 }
2118 
2119 int btrfs_commit_write(struct file *file, struct page *page,
2120 		       unsigned from, unsigned to)
2121 {
2122 	loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2123 	struct inode *inode = page->mapping->host;
2124 
2125 	btrfs_cow_one_page(inode, page, PAGE_CACHE_SIZE);
2126 
2127 	if (pos > inode->i_size) {
2128 		i_size_write(inode, pos);
2129 		mark_inode_dirty(inode);
2130 	}
2131 	return 0;
2132 }
2133 
2134 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
2135 {
2136 	struct btrfs_trans_handle *trans;
2137 	struct btrfs_key key;
2138 	struct btrfs_root_item root_item;
2139 	struct btrfs_inode_item *inode_item;
2140 	struct extent_buffer *leaf;
2141 	struct btrfs_root *new_root;
2142 	struct inode *inode;
2143 	struct inode *dir;
2144 	int ret;
2145 	int err;
2146 	u64 objectid;
2147 	u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2148 	unsigned long nr = 1;
2149 
2150 	mutex_lock(&root->fs_info->fs_mutex);
2151 	ret = btrfs_check_free_space(root, 1, 0);
2152 	if (ret)
2153 		goto fail_commit;
2154 
2155 	trans = btrfs_start_transaction(root, 1);
2156 	BUG_ON(!trans);
2157 
2158 	ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2159 				       0, &objectid);
2160 	if (ret)
2161 		goto fail;
2162 
2163 	leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2164 					objectid, trans->transid, 0, 0,
2165 					0, 0);
2166 	if (IS_ERR(leaf))
2167 		return PTR_ERR(leaf);
2168 
2169 	btrfs_set_header_nritems(leaf, 0);
2170 	btrfs_set_header_level(leaf, 0);
2171 	btrfs_set_header_bytenr(leaf, leaf->start);
2172 	btrfs_set_header_generation(leaf, trans->transid);
2173 	btrfs_set_header_owner(leaf, objectid);
2174 
2175 	write_extent_buffer(leaf, root->fs_info->fsid,
2176 			    (unsigned long)btrfs_header_fsid(leaf),
2177 			    BTRFS_FSID_SIZE);
2178 	btrfs_mark_buffer_dirty(leaf);
2179 
2180 	inode_item = &root_item.inode;
2181 	memset(inode_item, 0, sizeof(*inode_item));
2182 	inode_item->generation = cpu_to_le64(1);
2183 	inode_item->size = cpu_to_le64(3);
2184 	inode_item->nlink = cpu_to_le32(1);
2185 	inode_item->nblocks = cpu_to_le64(1);
2186 	inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2187 
2188 	btrfs_set_root_bytenr(&root_item, leaf->start);
2189 	btrfs_set_root_level(&root_item, 0);
2190 	btrfs_set_root_refs(&root_item, 1);
2191 	btrfs_set_root_used(&root_item, 0);
2192 
2193 	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2194 	root_item.drop_level = 0;
2195 
2196 	free_extent_buffer(leaf);
2197 	leaf = NULL;
2198 
2199 	btrfs_set_root_dirid(&root_item, new_dirid);
2200 
2201 	key.objectid = objectid;
2202 	key.offset = 1;
2203 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2204 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2205 				&root_item);
2206 	if (ret)
2207 		goto fail;
2208 
2209 	/*
2210 	 * insert the directory item
2211 	 */
2212 	key.offset = (u64)-1;
2213 	dir = root->fs_info->sb->s_root->d_inode;
2214 	ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2215 				    name, namelen, dir->i_ino, &key,
2216 				    BTRFS_FT_DIR);
2217 	if (ret)
2218 		goto fail;
2219 
2220 	ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2221 			     name, namelen, objectid,
2222 			     root->fs_info->sb->s_root->d_inode->i_ino);
2223 	if (ret)
2224 		goto fail;
2225 
2226 	ret = btrfs_commit_transaction(trans, root);
2227 	if (ret)
2228 		goto fail_commit;
2229 
2230 	new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2231 	BUG_ON(!new_root);
2232 
2233 	trans = btrfs_start_transaction(new_root, 1);
2234 	BUG_ON(!trans);
2235 
2236 	inode = btrfs_new_inode(trans, new_root, new_dirid,
2237 				BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2238 	if (IS_ERR(inode))
2239 		goto fail;
2240 	inode->i_op = &btrfs_dir_inode_operations;
2241 	inode->i_fop = &btrfs_dir_file_operations;
2242 	new_root->inode = inode;
2243 
2244 	ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2245 				     new_dirid);
2246 	inode->i_nlink = 1;
2247 	inode->i_size = 0;
2248 	ret = btrfs_update_inode(trans, new_root, inode);
2249 	if (ret)
2250 		goto fail;
2251 fail:
2252 	nr = trans->blocks_used;
2253 	err = btrfs_commit_transaction(trans, root);
2254 	if (err && !ret)
2255 		ret = err;
2256 fail_commit:
2257 	mutex_unlock(&root->fs_info->fs_mutex);
2258 	btrfs_btree_balance_dirty(root, nr);
2259 	return ret;
2260 }
2261 
2262 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2263 {
2264 	struct btrfs_trans_handle *trans;
2265 	struct btrfs_key key;
2266 	struct btrfs_root_item new_root_item;
2267 	struct extent_buffer *tmp;
2268 	int ret;
2269 	int err;
2270 	u64 objectid;
2271 	unsigned long nr = 0;
2272 
2273 	if (!root->ref_cows)
2274 		return -EINVAL;
2275 
2276 	down_write(&root->snap_sem);
2277 	freeze_bdev(root->fs_info->sb->s_bdev);
2278 	thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
2279 
2280 	mutex_lock(&root->fs_info->fs_mutex);
2281 	ret = btrfs_check_free_space(root, 1, 0);
2282 	if (ret)
2283 		goto fail_unlock;
2284 
2285 	trans = btrfs_start_transaction(root, 1);
2286 	BUG_ON(!trans);
2287 
2288 	ret = btrfs_update_inode(trans, root, root->inode);
2289 	if (ret)
2290 		goto fail;
2291 
2292 	ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2293 				       0, &objectid);
2294 	if (ret)
2295 		goto fail;
2296 
2297 	memcpy(&new_root_item, &root->root_item,
2298 	       sizeof(new_root_item));
2299 
2300 	key.objectid = objectid;
2301 	key.offset = 1;
2302 	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2303 
2304 	extent_buffer_get(root->node);
2305 	btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
2306 	free_extent_buffer(tmp);
2307 
2308 	btrfs_copy_root(trans, root, root->node, &tmp, objectid);
2309 
2310 	btrfs_set_root_bytenr(&new_root_item, tmp->start);
2311 	btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
2312 	ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2313 				&new_root_item);
2314 	free_extent_buffer(tmp);
2315 	if (ret)
2316 		goto fail;
2317 
2318 	/*
2319 	 * insert the directory item
2320 	 */
2321 	key.offset = (u64)-1;
2322 	ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2323 				    name, namelen,
2324 				    root->fs_info->sb->s_root->d_inode->i_ino,
2325 				    &key, BTRFS_FT_DIR);
2326 
2327 	if (ret)
2328 		goto fail;
2329 
2330 	ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2331 			     name, namelen, objectid,
2332 			     root->fs_info->sb->s_root->d_inode->i_ino);
2333 
2334 	if (ret)
2335 		goto fail;
2336 fail:
2337 	nr = trans->blocks_used;
2338 	err = btrfs_commit_transaction(trans, root);
2339 
2340 	if (err && !ret)
2341 		ret = err;
2342 fail_unlock:
2343 	mutex_unlock(&root->fs_info->fs_mutex);
2344 	up_write(&root->snap_sem);
2345 	btrfs_btree_balance_dirty(root, nr);
2346 	return ret;
2347 }
2348 
2349 static unsigned long force_ra(struct address_space *mapping,
2350 			      struct file_ra_state *ra, struct file *file,
2351 			      pgoff_t offset, pgoff_t last_index)
2352 {
2353 	pgoff_t req_size;
2354 
2355 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2356 	req_size = last_index - offset + 1;
2357 	offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2358 	return offset;
2359 #else
2360 	req_size = min(last_index - offset + 1, (pgoff_t)128);
2361 	page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2362 	return offset + req_size;
2363 #endif
2364 }
2365 
2366 int btrfs_defrag_file(struct file *file) {
2367 	struct inode *inode = fdentry(file)->d_inode;
2368 	struct btrfs_root *root = BTRFS_I(inode)->root;
2369 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2370 	struct page *page;
2371 	unsigned long last_index;
2372 	unsigned long ra_index = 0;
2373 	u64 page_start;
2374 	u64 page_end;
2375 	unsigned long i;
2376 	int ret;
2377 
2378 	mutex_lock(&root->fs_info->fs_mutex);
2379 	ret = btrfs_check_free_space(root, inode->i_size, 0);
2380 	mutex_unlock(&root->fs_info->fs_mutex);
2381 	if (ret)
2382 		return -ENOSPC;
2383 
2384 	mutex_lock(&inode->i_mutex);
2385 	last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2386 	for (i = 0; i <= last_index; i++) {
2387 		if (i == ra_index) {
2388 			ra_index = force_ra(inode->i_mapping, &file->f_ra,
2389 					    file, ra_index, last_index);
2390 		}
2391 		page = grab_cache_page(inode->i_mapping, i);
2392 		if (!page)
2393 			goto out_unlock;
2394 		if (!PageUptodate(page)) {
2395 			btrfs_readpage(NULL, page);
2396 			lock_page(page);
2397 			if (!PageUptodate(page)) {
2398 				unlock_page(page);
2399 				page_cache_release(page);
2400 				goto out_unlock;
2401 			}
2402 		}
2403 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2404 		page_end = page_start + PAGE_CACHE_SIZE - 1;
2405 
2406 		lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2407 		set_extent_delalloc(em_tree, page_start,
2408 				    page_end, GFP_NOFS);
2409 		unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2410 		set_page_dirty(page);
2411 		unlock_page(page);
2412 		page_cache_release(page);
2413 		balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2414 	}
2415 
2416 out_unlock:
2417 	mutex_unlock(&inode->i_mutex);
2418 	return 0;
2419 }
2420 
2421 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2422 {
2423 	struct btrfs_ioctl_vol_args *vol_args;
2424 	struct btrfs_dir_item *di;
2425 	struct btrfs_path *path;
2426 	u64 root_dirid;
2427 	int namelen;
2428 	int ret;
2429 
2430 	vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2431 
2432 	if (!vol_args)
2433 		return -ENOMEM;
2434 
2435 	if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2436 		ret = -EFAULT;
2437 		goto out;
2438 	}
2439 
2440 	namelen = strlen(vol_args->name);
2441 	if (namelen > BTRFS_VOL_NAME_MAX) {
2442 		ret = -EINVAL;
2443 		goto out;
2444 	}
2445 	if (strchr(vol_args->name, '/')) {
2446 		ret = -EINVAL;
2447 		goto out;
2448 	}
2449 
2450 	path = btrfs_alloc_path();
2451 	if (!path) {
2452 		ret = -ENOMEM;
2453 		goto out;
2454 	}
2455 
2456 	root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2457 	mutex_lock(&root->fs_info->fs_mutex);
2458 	di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2459 			    path, root_dirid,
2460 			    vol_args->name, namelen, 0);
2461 	mutex_unlock(&root->fs_info->fs_mutex);
2462 	btrfs_free_path(path);
2463 
2464 	if (di && !IS_ERR(di)) {
2465 		ret = -EEXIST;
2466 		goto out;
2467 	}
2468 
2469 	if (IS_ERR(di)) {
2470 		ret = PTR_ERR(di);
2471 		goto out;
2472 	}
2473 
2474 	if (root == root->fs_info->tree_root)
2475 		ret = create_subvol(root, vol_args->name, namelen);
2476 	else
2477 		ret = create_snapshot(root, vol_args->name, namelen);
2478 out:
2479 	kfree(vol_args);
2480 	return ret;
2481 }
2482 
2483 static int btrfs_ioctl_defrag(struct file *file)
2484 {
2485 	struct inode *inode = fdentry(file)->d_inode;
2486 	struct btrfs_root *root = BTRFS_I(inode)->root;
2487 
2488 	switch (inode->i_mode & S_IFMT) {
2489 	case S_IFDIR:
2490 		mutex_lock(&root->fs_info->fs_mutex);
2491 		btrfs_defrag_root(root, 0);
2492 		btrfs_defrag_root(root->fs_info->extent_root, 0);
2493 		mutex_unlock(&root->fs_info->fs_mutex);
2494 		break;
2495 	case S_IFREG:
2496 		btrfs_defrag_file(file);
2497 		break;
2498 	}
2499 
2500 	return 0;
2501 }
2502 
2503 long btrfs_ioctl(struct file *file, unsigned int
2504 		cmd, unsigned long arg)
2505 {
2506 	struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2507 
2508 	switch (cmd) {
2509 	case BTRFS_IOC_SNAP_CREATE:
2510 		return btrfs_ioctl_snap_create(root, (void __user *)arg);
2511 	case BTRFS_IOC_DEFRAG:
2512 		return btrfs_ioctl_defrag(file);
2513 	}
2514 
2515 	return -ENOTTY;
2516 }
2517 
2518 /*
2519  * Called inside transaction, so use GFP_NOFS
2520  */
2521 struct inode *btrfs_alloc_inode(struct super_block *sb)
2522 {
2523 	struct btrfs_inode *ei;
2524 
2525 	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2526 	if (!ei)
2527 		return NULL;
2528 	ei->last_trans = 0;
2529 	return &ei->vfs_inode;
2530 }
2531 
2532 void btrfs_destroy_inode(struct inode *inode)
2533 {
2534 	WARN_ON(!list_empty(&inode->i_dentry));
2535 	WARN_ON(inode->i_data.nrpages);
2536 
2537 	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2538 }
2539 
2540 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2541 static void init_once(struct kmem_cache * cachep, void *foo)
2542 #else
2543 static void init_once(void * foo, struct kmem_cache * cachep,
2544 		      unsigned long flags)
2545 #endif
2546 {
2547 	struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2548 
2549 	inode_init_once(&ei->vfs_inode);
2550 }
2551 
2552 void btrfs_destroy_cachep(void)
2553 {
2554 	if (btrfs_inode_cachep)
2555 		kmem_cache_destroy(btrfs_inode_cachep);
2556 	if (btrfs_trans_handle_cachep)
2557 		kmem_cache_destroy(btrfs_trans_handle_cachep);
2558 	if (btrfs_transaction_cachep)
2559 		kmem_cache_destroy(btrfs_transaction_cachep);
2560 	if (btrfs_bit_radix_cachep)
2561 		kmem_cache_destroy(btrfs_bit_radix_cachep);
2562 	if (btrfs_path_cachep)
2563 		kmem_cache_destroy(btrfs_path_cachep);
2564 }
2565 
2566 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2567 				       unsigned long extra_flags,
2568 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2569 				       void (*ctor)(struct kmem_cache *, void *)
2570 #else
2571 				       void (*ctor)(void *, struct kmem_cache *,
2572 						    unsigned long)
2573 #endif
2574 				     )
2575 {
2576 	return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2577 				 SLAB_MEM_SPREAD | extra_flags), ctor
2578 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2579 				 ,NULL
2580 #endif
2581 				);
2582 }
2583 
2584 int btrfs_init_cachep(void)
2585 {
2586 	btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2587 					  sizeof(struct btrfs_inode),
2588 					  0, init_once);
2589 	if (!btrfs_inode_cachep)
2590 		goto fail;
2591 	btrfs_trans_handle_cachep =
2592 			btrfs_cache_create("btrfs_trans_handle_cache",
2593 					   sizeof(struct btrfs_trans_handle),
2594 					   0, NULL);
2595 	if (!btrfs_trans_handle_cachep)
2596 		goto fail;
2597 	btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2598 					     sizeof(struct btrfs_transaction),
2599 					     0, NULL);
2600 	if (!btrfs_transaction_cachep)
2601 		goto fail;
2602 	btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2603 					 sizeof(struct btrfs_path),
2604 					 0, NULL);
2605 	if (!btrfs_path_cachep)
2606 		goto fail;
2607 	btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2608 					      SLAB_DESTROY_BY_RCU, NULL);
2609 	if (!btrfs_bit_radix_cachep)
2610 		goto fail;
2611 	return 0;
2612 fail:
2613 	btrfs_destroy_cachep();
2614 	return -ENOMEM;
2615 }
2616 
2617 static int btrfs_getattr(struct vfsmount *mnt,
2618 			 struct dentry *dentry, struct kstat *stat)
2619 {
2620 	struct inode *inode = dentry->d_inode;
2621 	generic_fillattr(inode, stat);
2622 	stat->blksize = 256 * 1024;
2623 	return 0;
2624 }
2625 
2626 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2627 			   struct inode * new_dir,struct dentry *new_dentry)
2628 {
2629 	struct btrfs_trans_handle *trans;
2630 	struct btrfs_root *root = BTRFS_I(old_dir)->root;
2631 	struct inode *new_inode = new_dentry->d_inode;
2632 	struct inode *old_inode = old_dentry->d_inode;
2633 	struct timespec ctime = CURRENT_TIME;
2634 	struct btrfs_path *path;
2635 	int ret;
2636 
2637 	if (S_ISDIR(old_inode->i_mode) && new_inode &&
2638 	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2639 		return -ENOTEMPTY;
2640 	}
2641 
2642 	mutex_lock(&root->fs_info->fs_mutex);
2643 	ret = btrfs_check_free_space(root, 1, 0);
2644 	if (ret)
2645 		goto out_unlock;
2646 
2647 	trans = btrfs_start_transaction(root, 1);
2648 
2649 	btrfs_set_trans_block_group(trans, new_dir);
2650 	path = btrfs_alloc_path();
2651 	if (!path) {
2652 		ret = -ENOMEM;
2653 		goto out_fail;
2654 	}
2655 
2656 	old_dentry->d_inode->i_nlink++;
2657 	old_dir->i_ctime = old_dir->i_mtime = ctime;
2658 	new_dir->i_ctime = new_dir->i_mtime = ctime;
2659 	old_inode->i_ctime = ctime;
2660 
2661 	ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2662 	if (ret)
2663 		goto out_fail;
2664 
2665 	if (new_inode) {
2666 		new_inode->i_ctime = CURRENT_TIME;
2667 		ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2668 		if (ret)
2669 			goto out_fail;
2670 	}
2671 	ret = btrfs_add_link(trans, new_dentry, old_inode);
2672 	if (ret)
2673 		goto out_fail;
2674 
2675 out_fail:
2676 	btrfs_free_path(path);
2677 	btrfs_end_transaction(trans, root);
2678 out_unlock:
2679 	mutex_unlock(&root->fs_info->fs_mutex);
2680 	return ret;
2681 }
2682 
2683 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2684 			 const char *symname)
2685 {
2686 	struct btrfs_trans_handle *trans;
2687 	struct btrfs_root *root = BTRFS_I(dir)->root;
2688 	struct btrfs_path *path;
2689 	struct btrfs_key key;
2690 	struct inode *inode = NULL;
2691 	int err;
2692 	int drop_inode = 0;
2693 	u64 objectid;
2694 	int name_len;
2695 	int datasize;
2696 	unsigned long ptr;
2697 	struct btrfs_file_extent_item *ei;
2698 	struct extent_buffer *leaf;
2699 	unsigned long nr = 0;
2700 
2701 	name_len = strlen(symname) + 1;
2702 	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2703 		return -ENAMETOOLONG;
2704 
2705 	mutex_lock(&root->fs_info->fs_mutex);
2706 	err = btrfs_check_free_space(root, 1, 0);
2707 	if (err)
2708 		goto out_fail;
2709 
2710 	trans = btrfs_start_transaction(root, 1);
2711 	btrfs_set_trans_block_group(trans, dir);
2712 
2713 	err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2714 	if (err) {
2715 		err = -ENOSPC;
2716 		goto out_unlock;
2717 	}
2718 
2719 	inode = btrfs_new_inode(trans, root, objectid,
2720 				BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2721 	err = PTR_ERR(inode);
2722 	if (IS_ERR(inode))
2723 		goto out_unlock;
2724 
2725 	btrfs_set_trans_block_group(trans, inode);
2726 	err = btrfs_add_nondir(trans, dentry, inode);
2727 	if (err)
2728 		drop_inode = 1;
2729 	else {
2730 		inode->i_mapping->a_ops = &btrfs_aops;
2731 		inode->i_fop = &btrfs_file_operations;
2732 		inode->i_op = &btrfs_file_inode_operations;
2733 		extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2734 				     inode->i_mapping, GFP_NOFS);
2735 		BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2736 	}
2737 	dir->i_sb->s_dirt = 1;
2738 	btrfs_update_inode_block_group(trans, inode);
2739 	btrfs_update_inode_block_group(trans, dir);
2740 	if (drop_inode)
2741 		goto out_unlock;
2742 
2743 	path = btrfs_alloc_path();
2744 	BUG_ON(!path);
2745 	key.objectid = inode->i_ino;
2746 	key.offset = 0;
2747 	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2748 	datasize = btrfs_file_extent_calc_inline_size(name_len);
2749 	err = btrfs_insert_empty_item(trans, root, path, &key,
2750 				      datasize);
2751 	if (err) {
2752 		drop_inode = 1;
2753 		goto out_unlock;
2754 	}
2755 	leaf = path->nodes[0];
2756 	ei = btrfs_item_ptr(leaf, path->slots[0],
2757 			    struct btrfs_file_extent_item);
2758 	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2759 	btrfs_set_file_extent_type(leaf, ei,
2760 				   BTRFS_FILE_EXTENT_INLINE);
2761 	ptr = btrfs_file_extent_inline_start(ei);
2762 	write_extent_buffer(leaf, symname, ptr, name_len);
2763 	btrfs_mark_buffer_dirty(leaf);
2764 	btrfs_free_path(path);
2765 
2766 	inode->i_op = &btrfs_symlink_inode_operations;
2767 	inode->i_mapping->a_ops = &btrfs_symlink_aops;
2768 	inode->i_size = name_len - 1;
2769 	err = btrfs_update_inode(trans, root, inode);
2770 	if (err)
2771 		drop_inode = 1;
2772 
2773 out_unlock:
2774 	nr = trans->blocks_used;
2775 	btrfs_end_transaction(trans, root);
2776 out_fail:
2777 	mutex_unlock(&root->fs_info->fs_mutex);
2778 	if (drop_inode) {
2779 		inode_dec_link_count(inode);
2780 		iput(inode);
2781 	}
2782 	btrfs_btree_balance_dirty(root, nr);
2783 	return err;
2784 }
2785 
2786 static struct inode_operations btrfs_dir_inode_operations = {
2787 	.lookup		= btrfs_lookup,
2788 	.create		= btrfs_create,
2789 	.unlink		= btrfs_unlink,
2790 	.link		= btrfs_link,
2791 	.mkdir		= btrfs_mkdir,
2792 	.rmdir		= btrfs_rmdir,
2793 	.rename		= btrfs_rename,
2794 	.symlink	= btrfs_symlink,
2795 	.setattr	= btrfs_setattr,
2796 	.mknod		= btrfs_mknod,
2797 	.setxattr	= generic_setxattr,
2798 	.getxattr	= generic_getxattr,
2799 	.listxattr	= btrfs_listxattr,
2800 	.removexattr	= generic_removexattr,
2801 };
2802 
2803 static struct inode_operations btrfs_dir_ro_inode_operations = {
2804 	.lookup		= btrfs_lookup,
2805 };
2806 
2807 static struct file_operations btrfs_dir_file_operations = {
2808 	.llseek		= generic_file_llseek,
2809 	.read		= generic_read_dir,
2810 	.readdir	= btrfs_readdir,
2811 	.unlocked_ioctl	= btrfs_ioctl,
2812 #ifdef CONFIG_COMPAT
2813 	.compat_ioctl	= btrfs_ioctl,
2814 #endif
2815 };
2816 
2817 static struct extent_map_ops btrfs_extent_map_ops = {
2818 	.fill_delalloc = run_delalloc_range,
2819 	.writepage_io_hook = btrfs_writepage_io_hook,
2820 	.readpage_io_hook = btrfs_readpage_io_hook,
2821 	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
2822 };
2823 
2824 static struct address_space_operations btrfs_aops = {
2825 	.readpage	= btrfs_readpage,
2826 	.writepage	= btrfs_writepage,
2827 	.writepages	= btrfs_writepages,
2828 	.readpages	= btrfs_readpages,
2829 	.sync_page	= block_sync_page,
2830 	.prepare_write	= btrfs_prepare_write,
2831 	.commit_write	= btrfs_commit_write,
2832 	.bmap		= btrfs_bmap,
2833 	.invalidatepage = btrfs_invalidatepage,
2834 	.releasepage	= btrfs_releasepage,
2835 	.set_page_dirty	= __set_page_dirty_nobuffers,
2836 };
2837 
2838 static struct address_space_operations btrfs_symlink_aops = {
2839 	.readpage	= btrfs_readpage,
2840 	.writepage	= btrfs_writepage,
2841 	.invalidatepage = btrfs_invalidatepage,
2842 	.releasepage	= btrfs_releasepage,
2843 };
2844 
2845 static struct inode_operations btrfs_file_inode_operations = {
2846 	.truncate	= btrfs_truncate,
2847 	.getattr	= btrfs_getattr,
2848 	.setattr	= btrfs_setattr,
2849 	.setxattr	= generic_setxattr,
2850 	.getxattr	= generic_getxattr,
2851 	.listxattr      = btrfs_listxattr,
2852 	.removexattr	= generic_removexattr,
2853 };
2854 
2855 static struct inode_operations btrfs_special_inode_operations = {
2856 	.getattr	= btrfs_getattr,
2857 	.setattr	= btrfs_setattr,
2858 };
2859 
2860 static struct inode_operations btrfs_symlink_inode_operations = {
2861 	.readlink	= generic_readlink,
2862 	.follow_link	= page_follow_link_light,
2863 	.put_link	= page_put_link,
2864 };
2865