1 #include <linux/module.h>
2 #include <linux/buffer_head.h>
4 #include <linux/pagemap.h>
5 #include <linux/highmem.h>
6 #include <linux/time.h>
7 #include <linux/init.h>
8 #include <linux/string.h>
9 #include <linux/smp_lock.h>
10 #include <linux/backing-dev.h>
11 #include <linux/mpage.h>
12 #include <linux/swap.h>
13 #include <linux/writeback.h>
14 #include <linux/statfs.h>
17 #include "transaction.h"
18 #include "btrfs_inode.h"
21 void btrfs_fsinfo_release(struct kobject *obj)
23 struct btrfs_fs_info *fsinfo = container_of(obj,
24 struct btrfs_fs_info, kobj);
28 struct kobj_type btrfs_fsinfo_ktype = {
29 .release = btrfs_fsinfo_release,
32 struct btrfs_iget_args {
34 struct btrfs_root *root;
37 decl_subsys(btrfs, &btrfs_fsinfo_ktype, NULL);
39 #define BTRFS_SUPER_MAGIC 0x9123682E
41 static struct inode_operations btrfs_dir_inode_operations;
42 static struct inode_operations btrfs_dir_ro_inode_operations;
43 static struct super_operations btrfs_super_ops;
44 static struct file_operations btrfs_dir_file_operations;
45 static struct inode_operations btrfs_file_inode_operations;
46 static struct address_space_operations btrfs_aops;
47 static struct file_operations btrfs_file_operations;
49 static void btrfs_read_locked_inode(struct inode *inode)
51 struct btrfs_path *path;
52 struct btrfs_inode_item *inode_item;
53 struct btrfs_root *root = BTRFS_I(inode)->root;
54 struct btrfs_key location;
57 path = btrfs_alloc_path();
59 btrfs_init_path(path);
60 mutex_lock(&root->fs_info->fs_mutex);
62 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
63 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
65 btrfs_free_path(path);
68 inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
70 struct btrfs_inode_item);
72 inode->i_mode = btrfs_inode_mode(inode_item);
73 inode->i_nlink = btrfs_inode_nlink(inode_item);
74 inode->i_uid = btrfs_inode_uid(inode_item);
75 inode->i_gid = btrfs_inode_gid(inode_item);
76 inode->i_size = btrfs_inode_size(inode_item);
77 inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
78 inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
79 inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
80 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
81 inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
82 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
83 inode->i_blocks = btrfs_inode_nblocks(inode_item);
84 inode->i_generation = btrfs_inode_generation(inode_item);
86 btrfs_free_path(path);
89 mutex_unlock(&root->fs_info->fs_mutex);
91 switch (inode->i_mode & S_IFMT) {
94 init_special_inode(inode, inode->i_mode,
95 btrfs_inode_rdev(inode_item));
99 inode->i_mapping->a_ops = &btrfs_aops;
100 inode->i_fop = &btrfs_file_operations;
101 inode->i_op = &btrfs_file_inode_operations;
104 inode->i_fop = &btrfs_dir_file_operations;
105 if (root == root->fs_info->tree_root)
106 inode->i_op = &btrfs_dir_ro_inode_operations;
108 inode->i_op = &btrfs_dir_inode_operations;
111 // inode->i_op = &page_symlink_inode_operations;
117 btrfs_release_path(root, path);
118 btrfs_free_path(path);
119 mutex_unlock(&root->fs_info->fs_mutex);
120 make_bad_inode(inode);
123 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
124 struct btrfs_root *root,
126 struct dentry *dentry)
128 struct btrfs_path *path;
129 const char *name = dentry->d_name.name;
130 int name_len = dentry->d_name.len;
133 struct btrfs_dir_item *di;
135 path = btrfs_alloc_path();
137 btrfs_init_path(path);
138 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
148 objectid = btrfs_disk_key_objectid(&di->location);
149 ret = btrfs_delete_one_dir_name(trans, root, path, di);
151 btrfs_release_path(root, path);
153 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
154 objectid, name, name_len, -1);
163 ret = btrfs_delete_one_dir_name(trans, root, path, di);
166 dentry->d_inode->i_ctime = dir->i_ctime;
168 btrfs_free_path(path);
170 inode_dec_link_count(dentry->d_inode);
171 dir->i_size -= name_len * 2;
172 mark_inode_dirty(dir);
177 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
179 struct btrfs_root *root;
180 struct btrfs_trans_handle *trans;
183 root = BTRFS_I(dir)->root;
184 mutex_lock(&root->fs_info->fs_mutex);
185 trans = btrfs_start_transaction(root, 1);
186 ret = btrfs_unlink_trans(trans, root, dir, dentry);
187 btrfs_end_transaction(trans, root);
188 mutex_unlock(&root->fs_info->fs_mutex);
192 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
194 struct inode *inode = dentry->d_inode;
197 struct btrfs_root *root = BTRFS_I(dir)->root;
198 struct btrfs_path *path;
199 struct btrfs_key key;
200 struct btrfs_trans_handle *trans;
201 struct btrfs_key found_key;
203 struct btrfs_leaf *leaf;
204 char *goodnames = "..";
206 path = btrfs_alloc_path();
208 btrfs_init_path(path);
209 mutex_lock(&root->fs_info->fs_mutex);
210 trans = btrfs_start_transaction(root, 1);
211 key.objectid = inode->i_ino;
212 key.offset = (u64)-1;
215 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
221 if (path->slots[0] == 0) {
226 leaf = btrfs_buffer_leaf(path->nodes[0]);
227 btrfs_disk_key_to_cpu(&found_key,
228 &leaf->items[path->slots[0]].key);
229 found_type = btrfs_key_type(&found_key);
230 if (found_key.objectid != inode->i_ino) {
234 if ((found_type != BTRFS_DIR_ITEM_KEY &&
235 found_type != BTRFS_DIR_INDEX_KEY) ||
236 (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
237 !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
241 ret = btrfs_del_item(trans, root, path);
244 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
246 btrfs_release_path(root, path);
249 btrfs_release_path(root, path);
251 /* now the directory is empty */
252 err = btrfs_unlink_trans(trans, root, dir, dentry);
257 btrfs_release_path(root, path);
258 btrfs_free_path(path);
259 mutex_unlock(&root->fs_info->fs_mutex);
260 ret = btrfs_end_transaction(trans, root);
266 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
267 struct btrfs_root *root,
270 struct btrfs_path *path;
275 path = btrfs_alloc_path();
277 btrfs_init_path(path);
278 ret = btrfs_lookup_inode(trans, root, path,
279 &BTRFS_I(inode)->location, -1);
281 ret = btrfs_del_item(trans, root, path);
283 btrfs_free_path(path);
287 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
288 struct btrfs_root *root,
292 struct btrfs_path *path;
293 struct btrfs_key key;
294 struct btrfs_disk_key *found_key;
295 struct btrfs_leaf *leaf;
296 struct btrfs_file_extent_item *fi = NULL;
297 u64 extent_start = 0;
298 u64 extent_num_blocks = 0;
301 path = btrfs_alloc_path();
303 /* FIXME, add redo link to tree so we don't leak on crash */
304 key.objectid = inode->i_ino;
305 key.offset = (u64)-1;
308 * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
311 btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
313 btrfs_init_path(path);
314 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
319 BUG_ON(path->slots[0] == 0);
322 leaf = btrfs_buffer_leaf(path->nodes[0]);
323 found_key = &leaf->items[path->slots[0]].key;
324 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
326 if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
327 btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
328 btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
330 if (btrfs_disk_key_offset(found_key) < inode->i_size)
333 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
334 fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
336 struct btrfs_file_extent_item);
337 if (btrfs_file_extent_type(fi) !=
338 BTRFS_FILE_EXTENT_INLINE) {
340 btrfs_file_extent_disk_blocknr(fi);
342 btrfs_file_extent_disk_num_blocks(fi);
343 /* FIXME blocksize != 4096 */
345 btrfs_file_extent_num_blocks(fi) << 3;
349 ret = btrfs_del_item(trans, root, path);
351 btrfs_release_path(root, path);
353 ret = btrfs_free_extent(trans, root, extent_start,
354 extent_num_blocks, 0);
360 btrfs_release_path(root, path);
361 btrfs_free_path(path);
365 static void btrfs_delete_inode(struct inode *inode)
367 struct btrfs_trans_handle *trans;
368 struct btrfs_root *root = BTRFS_I(inode)->root;
371 truncate_inode_pages(&inode->i_data, 0);
372 if (is_bad_inode(inode)) {
376 mutex_lock(&root->fs_info->fs_mutex);
377 trans = btrfs_start_transaction(root, 1);
378 if (S_ISREG(inode->i_mode)) {
379 ret = btrfs_truncate_in_trans(trans, root, inode);
382 btrfs_free_inode(trans, root, inode);
383 btrfs_end_transaction(trans, root);
384 mutex_unlock(&root->fs_info->fs_mutex);
390 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
391 struct btrfs_key *location)
393 const char *name = dentry->d_name.name;
394 int namelen = dentry->d_name.len;
395 struct btrfs_dir_item *di;
396 struct btrfs_path *path;
397 struct btrfs_root *root = BTRFS_I(dir)->root;
400 path = btrfs_alloc_path();
402 btrfs_init_path(path);
403 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
405 if (!di || IS_ERR(di)) {
406 location->objectid = 0;
410 btrfs_disk_key_to_cpu(location, &di->location);
412 btrfs_release_path(root, path);
413 btrfs_free_path(path);
417 int fixup_tree_root_location(struct btrfs_root *root,
418 struct btrfs_key *location,
419 struct btrfs_root **sub_root)
421 struct btrfs_path *path;
422 struct btrfs_root_item *ri;
424 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
426 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
429 path = btrfs_alloc_path();
431 mutex_lock(&root->fs_info->fs_mutex);
433 *sub_root = btrfs_read_fs_root(root->fs_info, location);
434 if (IS_ERR(*sub_root))
435 return PTR_ERR(*sub_root);
437 ri = &(*sub_root)->root_item;
438 location->objectid = btrfs_root_dirid(ri);
440 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
441 location->offset = 0;
443 btrfs_free_path(path);
444 mutex_unlock(&root->fs_info->fs_mutex);
448 int btrfs_init_locked_inode(struct inode *inode, void *p)
450 struct btrfs_iget_args *args = p;
451 inode->i_ino = args->ino;
452 BTRFS_I(inode)->root = args->root;
456 int btrfs_find_actor(struct inode *inode, void *opaque)
458 struct btrfs_iget_args *args = opaque;
459 return (args->ino == inode->i_ino &&
460 args->root == BTRFS_I(inode)->root);
463 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
464 struct btrfs_root *root)
467 struct btrfs_iget_args args;
471 inode = iget5_locked(s, objectid, btrfs_find_actor,
472 btrfs_init_locked_inode,
477 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
478 struct nameidata *nd)
480 struct inode * inode;
481 struct btrfs_inode *bi = BTRFS_I(dir);
482 struct btrfs_root *root = bi->root;
483 struct btrfs_root *sub_root = root;
484 struct btrfs_key location;
487 if (dentry->d_name.len > BTRFS_NAME_LEN)
488 return ERR_PTR(-ENAMETOOLONG);
489 mutex_lock(&root->fs_info->fs_mutex);
490 ret = btrfs_inode_by_name(dir, dentry, &location);
491 mutex_unlock(&root->fs_info->fs_mutex);
495 if (location.objectid) {
496 ret = fixup_tree_root_location(root, &location, &sub_root);
500 return ERR_PTR(-ENOENT);
501 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
504 return ERR_PTR(-EACCES);
505 if (inode->i_state & I_NEW) {
506 if (sub_root != root) {
507 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
509 sub_root->inode = inode;
511 BTRFS_I(inode)->root = sub_root;
512 memcpy(&BTRFS_I(inode)->location, &location,
514 btrfs_read_locked_inode(inode);
515 unlock_new_inode(inode);
518 return d_splice_alias(inode, dentry);
521 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
523 struct inode *inode = filp->f_path.dentry->d_inode;
524 struct btrfs_root *root = BTRFS_I(inode)->root;
525 struct btrfs_item *item;
526 struct btrfs_dir_item *di;
527 struct btrfs_key key;
528 struct btrfs_path *path;
531 struct btrfs_leaf *leaf;
534 unsigned char d_type = DT_UNKNOWN;
539 int key_type = BTRFS_DIR_INDEX_KEY;
541 /* FIXME, use a real flag for deciding about the key type */
542 if (root->fs_info->tree_root == root)
543 key_type = BTRFS_DIR_ITEM_KEY;
544 mutex_lock(&root->fs_info->fs_mutex);
545 key.objectid = inode->i_ino;
547 btrfs_set_key_type(&key, key_type);
548 key.offset = filp->f_pos;
549 path = btrfs_alloc_path();
550 btrfs_init_path(path);
551 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
556 leaf = btrfs_buffer_leaf(path->nodes[0]);
557 nritems = btrfs_header_nritems(&leaf->header);
558 slot = path->slots[0];
559 if (advance || slot >= nritems) {
560 if (slot >= nritems -1) {
561 ret = btrfs_next_leaf(root, path);
564 leaf = btrfs_buffer_leaf(path->nodes[0]);
565 nritems = btrfs_header_nritems(&leaf->header);
566 slot = path->slots[0];
573 item = leaf->items + slot;
574 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
576 if (btrfs_disk_key_type(&item->key) != key_type)
578 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
580 filp->f_pos = btrfs_disk_key_offset(&item->key);
582 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
584 di_total = btrfs_item_size(leaf->items + slot);
585 while(di_cur < di_total) {
586 over = filldir(dirent, (const char *)(di + 1),
587 btrfs_dir_name_len(di),
588 btrfs_disk_key_offset(&item->key),
589 btrfs_disk_key_objectid(&di->location),
593 di_len = btrfs_dir_name_len(di) + sizeof(*di);
595 di = (struct btrfs_dir_item *)((char *)di + di_len);
602 btrfs_release_path(root, path);
603 btrfs_free_path(path);
604 mutex_unlock(&root->fs_info->fs_mutex);
608 static void btrfs_put_super (struct super_block * sb)
610 struct btrfs_root *root = btrfs_sb(sb);
613 ret = close_ctree(root);
615 printk("close ctree returns %d\n", ret);
617 sb->s_fs_info = NULL;
620 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
622 struct inode * inode;
623 struct dentry * root_dentry;
624 struct btrfs_super_block *disk_super;
625 struct btrfs_root *tree_root;
626 struct btrfs_inode *bi;
628 sb->s_maxbytes = MAX_LFS_FILESIZE;
629 sb->s_magic = BTRFS_SUPER_MAGIC;
630 sb->s_op = &btrfs_super_ops;
633 tree_root = open_ctree(sb);
636 printk("btrfs: open_ctree failed\n");
639 sb->s_fs_info = tree_root;
640 disk_super = tree_root->fs_info->disk_super;
641 printk("read in super total blocks %Lu root %Lu\n",
642 btrfs_super_total_blocks(disk_super),
643 btrfs_super_root_dir(disk_super));
645 inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
648 bi->location.objectid = inode->i_ino;
649 bi->location.offset = 0;
650 bi->location.flags = 0;
651 bi->root = tree_root;
652 btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
656 if (inode->i_state & I_NEW) {
657 btrfs_read_locked_inode(inode);
658 unlock_new_inode(inode);
661 root_dentry = d_alloc_root(inode);
666 sb->s_root = root_dentry;
671 static void fill_inode_item(struct btrfs_inode_item *item,
674 btrfs_set_inode_uid(item, inode->i_uid);
675 btrfs_set_inode_gid(item, inode->i_gid);
676 btrfs_set_inode_size(item, inode->i_size);
677 btrfs_set_inode_mode(item, inode->i_mode);
678 btrfs_set_inode_nlink(item, inode->i_nlink);
679 btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
680 btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
681 btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
682 btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
683 btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
684 btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
685 btrfs_set_inode_nblocks(item, inode->i_blocks);
686 btrfs_set_inode_generation(item, inode->i_generation);
689 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root,
693 struct btrfs_inode_item *inode_item;
694 struct btrfs_path *path;
697 path = btrfs_alloc_path();
699 btrfs_init_path(path);
700 ret = btrfs_lookup_inode(trans, root, path,
701 &BTRFS_I(inode)->location, 1);
708 inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
710 struct btrfs_inode_item);
712 fill_inode_item(inode_item, inode);
713 btrfs_mark_buffer_dirty(path->nodes[0]);
716 btrfs_release_path(root, path);
717 btrfs_free_path(path);
721 static int btrfs_write_inode(struct inode *inode, int wait)
723 struct btrfs_root *root = BTRFS_I(inode)->root;
724 struct btrfs_trans_handle *trans;
727 mutex_lock(&root->fs_info->fs_mutex);
728 trans = btrfs_start_transaction(root, 1);
729 ret = btrfs_update_inode(trans, root, inode);
731 btrfs_commit_transaction(trans, root);
733 btrfs_end_transaction(trans, root);
734 mutex_unlock(&root->fs_info->fs_mutex);
738 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
739 struct btrfs_root *root,
740 u64 objectid, int mode)
743 struct btrfs_inode_item inode_item;
744 struct btrfs_key *location;
747 inode = new_inode(root->fs_info->sb);
749 return ERR_PTR(-ENOMEM);
751 BTRFS_I(inode)->root = root;
753 inode->i_uid = current->fsuid;
754 inode->i_gid = current->fsgid;
755 inode->i_mode = mode;
756 inode->i_ino = objectid;
758 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
759 fill_inode_item(&inode_item, inode);
760 location = &BTRFS_I(inode)->location;
761 location->objectid = objectid;
763 location->offset = 0;
764 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
766 ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
769 insert_inode_hash(inode);
773 static int btrfs_add_link(struct btrfs_trans_handle *trans,
774 struct dentry *dentry, struct inode *inode)
777 struct btrfs_key key;
778 struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
779 key.objectid = inode->i_ino;
781 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
784 ret = btrfs_insert_dir_item(trans, root,
785 dentry->d_name.name, dentry->d_name.len,
786 dentry->d_parent->d_inode->i_ino,
789 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
790 ret = btrfs_update_inode(trans, root,
791 dentry->d_parent->d_inode);
796 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
797 struct dentry *dentry, struct inode *inode)
799 int err = btrfs_add_link(trans, dentry, inode);
801 d_instantiate(dentry, inode);
809 static int btrfs_create(struct inode *dir, struct dentry *dentry,
810 int mode, struct nameidata *nd)
812 struct btrfs_trans_handle *trans;
813 struct btrfs_root *root = BTRFS_I(dir)->root;
819 mutex_lock(&root->fs_info->fs_mutex);
820 trans = btrfs_start_transaction(root, 1);
822 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
828 inode = btrfs_new_inode(trans, root, objectid, mode);
829 err = PTR_ERR(inode);
832 // FIXME mark the inode dirty
833 err = btrfs_add_nondir(trans, dentry, inode);
837 inode->i_mapping->a_ops = &btrfs_aops;
838 inode->i_fop = &btrfs_file_operations;
839 inode->i_op = &btrfs_file_inode_operations;
841 dir->i_sb->s_dirt = 1;
843 btrfs_end_transaction(trans, root);
844 mutex_unlock(&root->fs_info->fs_mutex);
847 inode_dec_link_count(inode);
853 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
854 struct btrfs_root *root,
855 u64 objectid, u64 dirid)
859 struct btrfs_key key;
864 key.objectid = objectid;
867 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
869 ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
873 key.objectid = dirid;
874 ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
882 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
885 struct btrfs_trans_handle *trans;
886 struct btrfs_root *root = BTRFS_I(dir)->root;
891 mutex_lock(&root->fs_info->fs_mutex);
892 trans = btrfs_start_transaction(root, 1);
894 err = PTR_ERR(trans);
898 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
904 inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
906 err = PTR_ERR(inode);
910 inode->i_op = &btrfs_dir_inode_operations;
911 inode->i_fop = &btrfs_dir_file_operations;
913 err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
918 err = btrfs_update_inode(trans, root, inode);
921 err = btrfs_add_link(trans, dentry, inode);
924 d_instantiate(dentry, inode);
928 btrfs_end_transaction(trans, root);
930 mutex_unlock(&root->fs_info->fs_mutex);
936 static int btrfs_sync_file(struct file *file,
937 struct dentry *dentry, int datasync)
939 struct inode *inode = dentry->d_inode;
940 struct btrfs_root *root = BTRFS_I(inode)->root;
942 struct btrfs_trans_handle *trans;
944 mutex_lock(&root->fs_info->fs_mutex);
945 trans = btrfs_start_transaction(root, 1);
950 ret = btrfs_commit_transaction(trans, root);
951 mutex_unlock(&root->fs_info->fs_mutex);
953 return ret > 0 ? EIO : ret;
956 static int btrfs_sync_fs(struct super_block *sb, int wait)
958 struct btrfs_trans_handle *trans;
959 struct btrfs_root *root;
965 filemap_flush(root->fs_info->btree_inode->i_mapping);
968 filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
969 mutex_lock(&root->fs_info->fs_mutex);
970 trans = btrfs_start_transaction(root, 1);
971 ret = btrfs_commit_transaction(trans, root);
974 printk("btrfs sync_fs\n");
975 mutex_unlock(&root->fs_info->fs_mutex);
979 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
980 struct buffer_head *result, int create)
985 u64 extent_start = 0;
987 u64 objectid = inode->i_ino;
989 struct btrfs_path *path;
990 struct btrfs_root *root = BTRFS_I(inode)->root;
991 struct btrfs_file_extent_item *item;
992 struct btrfs_leaf *leaf;
993 struct btrfs_disk_key *found_key;
995 path = btrfs_alloc_path();
997 btrfs_init_path(path);
1002 ret = btrfs_lookup_file_extent(NULL, root, path,
1004 iblock << inode->i_blkbits, 0);
1011 if (path->slots[0] == 0) {
1012 btrfs_release_path(root, path);
1018 item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1019 struct btrfs_file_extent_item);
1020 leaf = btrfs_buffer_leaf(path->nodes[0]);
1021 blocknr = btrfs_file_extent_disk_blocknr(item);
1022 blocknr += btrfs_file_extent_offset(item);
1024 /* are we inside the extent that was found? */
1025 found_key = &leaf->items[path->slots[0]].key;
1026 found_type = btrfs_disk_key_type(found_key);
1027 if (btrfs_disk_key_objectid(found_key) != objectid ||
1028 found_type != BTRFS_EXTENT_DATA_KEY) {
1031 btrfs_release_path(root, path);
1034 found_type = btrfs_file_extent_type(item);
1035 extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1036 if (found_type == BTRFS_FILE_EXTENT_REG) {
1037 extent_start = extent_start >> inode->i_blkbits;
1038 extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1039 if (iblock >= extent_start && iblock < extent_end) {
1041 btrfs_map_bh_to_logical(root, result, blocknr +
1042 iblock - extent_start);
1045 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1049 size = btrfs_file_extent_inline_len(leaf->items +
1051 extent_end = (extent_start + size) >> inode->i_blkbits;
1052 extent_start >>= inode->i_blkbits;
1053 if (iblock < extent_start || iblock > extent_end) {
1056 ptr = btrfs_file_extent_inline_start(item);
1057 map = kmap(result->b_page);
1058 memcpy(map, ptr, size);
1059 memset(map + size, 0, PAGE_CACHE_SIZE - size);
1060 flush_dcache_page(result->b_page);
1061 kunmap(result->b_page);
1062 set_buffer_uptodate(result);
1063 SetPageChecked(result->b_page);
1064 btrfs_map_bh_to_logical(root, result, 0);
1067 btrfs_release_path(root, path);
1068 btrfs_free_path(path);
1072 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1073 struct buffer_head *result, int create)
1076 struct btrfs_root *root = BTRFS_I(inode)->root;
1077 mutex_lock(&root->fs_info->fs_mutex);
1078 err = btrfs_get_block_lock(inode, iblock, result, create);
1079 mutex_unlock(&root->fs_info->fs_mutex);
1083 static int btrfs_prepare_write(struct file *file, struct page *page,
1084 unsigned from, unsigned to)
1086 return nobh_prepare_write(page, from, to, btrfs_get_block);
1089 static void btrfs_write_super(struct super_block *sb)
1091 btrfs_sync_fs(sb, 1);
1094 static int btrfs_readpage(struct file *file, struct page *page)
1096 return mpage_readpage(page, btrfs_get_block);
1100 * While block_write_full_page is writing back the dirty buffers under
1101 * the page lock, whoever dirtied the buffers may decide to clean them
1102 * again at any time. We handle that by only looking at the buffer
1103 * state inside lock_buffer().
1105 * If block_write_full_page() is called for regular writeback
1106 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1107 * locked buffer. This only can happen if someone has written the buffer
1108 * directly, with submit_bh(). At the address_space level PageWriteback
1109 * prevents this contention from occurring.
1111 static int __btrfs_write_full_page(struct inode *inode, struct page *page,
1112 struct writeback_control *wbc)
1116 sector_t last_block;
1117 struct buffer_head *bh, *head;
1118 const unsigned blocksize = 1 << inode->i_blkbits;
1119 int nr_underway = 0;
1121 BUG_ON(!PageLocked(page));
1123 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1125 if (!page_has_buffers(page)) {
1126 create_empty_buffers(page, blocksize,
1127 (1 << BH_Dirty)|(1 << BH_Uptodate));
1131 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1132 * here, and the (potentially unmapped) buffers may become dirty at
1133 * any time. If a buffer becomes dirty here after we've inspected it
1134 * then we just miss that fact, and the page stays dirty.
1136 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1137 * handle that here by just cleaning them.
1140 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1141 head = page_buffers(page);
1145 * Get all the dirty buffers mapped to disk addresses and
1146 * handle any aliases from the underlying blockdev's mapping.
1149 if (block > last_block) {
1151 * mapped buffers outside i_size will occur, because
1152 * this page can be outside i_size when there is a
1153 * truncate in progress.
1156 * The buffer was zeroed by block_write_full_page()
1158 clear_buffer_dirty(bh);
1159 set_buffer_uptodate(bh);
1160 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1161 WARN_ON(bh->b_size != blocksize);
1162 err = btrfs_get_block(inode, block, bh, 0);
1165 if (buffer_new(bh)) {
1166 /* blockdev mappings never come here */
1167 clear_buffer_new(bh);
1168 unmap_underlying_metadata(bh->b_bdev,
1172 bh = bh->b_this_page;
1174 } while (bh != head);
1177 if (!buffer_mapped(bh))
1180 * If it's a fully non-blocking write attempt and we cannot
1181 * lock the buffer then redirty the page. Note that this can
1182 * potentially cause a busy-wait loop from pdflush and kswapd
1183 * activity, but those code paths have their own higher-level
1186 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1188 } else if (test_set_buffer_locked(bh)) {
1189 redirty_page_for_writepage(wbc, page);
1192 if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
1193 mark_buffer_async_write(bh);
1197 } while ((bh = bh->b_this_page) != head);
1200 * The page and its buffers are protected by PageWriteback(), so we can
1201 * drop the bh refcounts early.
1203 BUG_ON(PageWriteback(page));
1204 set_page_writeback(page);
1207 struct buffer_head *next = bh->b_this_page;
1208 if (buffer_async_write(bh)) {
1209 submit_bh(WRITE, bh);
1213 } while (bh != head);
1218 if (nr_underway == 0) {
1220 * The page was marked dirty, but the buffers were
1221 * clean. Someone wrote them back by hand with
1222 * ll_rw_block/submit_bh. A rare case.
1226 if (!buffer_uptodate(bh)) {
1230 bh = bh->b_this_page;
1231 } while (bh != head);
1233 SetPageUptodate(page);
1234 end_page_writeback(page);
1236 * The page and buffer_heads can be released at any time from
1239 wbc->pages_skipped++; /* We didn't write this page */
1245 * ENOSPC, or some other error. We may already have added some
1246 * blocks to the file, so we need to write these out to avoid
1247 * exposing stale data.
1248 * The page is currently locked and not marked for writeback
1251 /* Recovery: lock and submit the mapped buffers */
1253 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1255 mark_buffer_async_write(bh);
1258 * The buffer may have been set dirty during
1259 * attachment to a dirty page.
1261 clear_buffer_dirty(bh);
1263 } while ((bh = bh->b_this_page) != head);
1265 BUG_ON(PageWriteback(page));
1266 set_page_writeback(page);
1268 struct buffer_head *next = bh->b_this_page;
1269 if (buffer_async_write(bh)) {
1270 clear_buffer_dirty(bh);
1271 submit_bh(WRITE, bh);
1275 } while (bh != head);
1281 * The generic ->writepage function for buffer-backed address_spaces
1283 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1285 struct inode * const inode = page->mapping->host;
1286 loff_t i_size = i_size_read(inode);
1287 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1291 /* Is the page fully inside i_size? */
1292 if (page->index < end_index)
1293 return __btrfs_write_full_page(inode, page, wbc);
1295 /* Is the page fully outside i_size? (truncate in progress) */
1296 offset = i_size & (PAGE_CACHE_SIZE-1);
1297 if (page->index >= end_index+1 || !offset) {
1299 * The page may have dirty, unmapped buffers. For example,
1300 * they may have been added in ext3_writepage(). Make them
1301 * freeable here, so the page does not leak.
1303 block_invalidatepage(page, 0);
1305 return 0; /* don't care */
1309 * The page straddles i_size. It must be zeroed out on each and every
1310 * writepage invokation because it may be mmapped. "A file is mapped
1311 * in multiples of the page size. For a file that is not a multiple of
1312 * the page size, the remaining memory is zeroed when mapped, and
1313 * writes to that region are not written out to the file."
1315 kaddr = kmap_atomic(page, KM_USER0);
1316 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1317 flush_dcache_page(page);
1318 kunmap_atomic(kaddr, KM_USER0);
1319 return __btrfs_write_full_page(inode, page, wbc);
1322 static void btrfs_truncate(struct inode *inode)
1324 struct btrfs_root *root = BTRFS_I(inode)->root;
1326 struct btrfs_trans_handle *trans;
1328 if (!S_ISREG(inode->i_mode))
1330 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1333 nobh_truncate_page(inode->i_mapping, inode->i_size);
1335 /* FIXME, add redo link to tree so we don't leak on crash */
1336 mutex_lock(&root->fs_info->fs_mutex);
1337 trans = btrfs_start_transaction(root, 1);
1338 ret = btrfs_truncate_in_trans(trans, root, inode);
1340 ret = btrfs_end_transaction(trans, root);
1342 mutex_unlock(&root->fs_info->fs_mutex);
1343 mark_inode_dirty(inode);
1347 * Make sure any changes to nobh_commit_write() are reflected in
1348 * nobh_truncate_page(), since it doesn't call commit_write().
1350 static int btrfs_commit_write(struct file *file, struct page *page,
1351 unsigned from, unsigned to)
1353 struct inode *inode = page->mapping->host;
1354 struct buffer_head *bh;
1355 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1357 SetPageUptodate(page);
1358 bh = page_buffers(page);
1359 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1360 set_page_dirty(page);
1362 if (pos > inode->i_size) {
1363 i_size_write(inode, pos);
1364 mark_inode_dirty(inode);
1369 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1370 struct page **prepared_pages,
1371 const char __user * buf)
1373 long page_fault = 0;
1375 int offset = pos & (PAGE_CACHE_SIZE - 1);
1377 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1378 size_t count = min_t(size_t,
1379 PAGE_CACHE_SIZE - offset, write_bytes);
1380 struct page *page = prepared_pages[i];
1381 fault_in_pages_readable(buf, count);
1383 /* Copy data from userspace to the current page */
1385 page_fault = __copy_from_user(page_address(page) + offset,
1387 /* Flush processor's dcache for this page */
1388 flush_dcache_page(page);
1391 write_bytes -= count;
1396 return page_fault ? -EFAULT : 0;
1399 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1402 for (i = 0; i < num_pages; i++) {
1405 unlock_page(pages[i]);
1406 mark_page_accessed(pages[i]);
1407 page_cache_release(pages[i]);
1410 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1411 struct btrfs_root *root,
1413 struct page **pages,
1423 struct inode *inode = file->f_path.dentry->d_inode;
1424 struct buffer_head *bh;
1425 struct btrfs_file_extent_item *ei;
1427 for (i = 0; i < num_pages; i++) {
1428 offset = pos & (PAGE_CACHE_SIZE -1);
1429 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1430 /* FIXME, one block at a time */
1432 mutex_lock(&root->fs_info->fs_mutex);
1433 trans = btrfs_start_transaction(root, 1);
1435 bh = page_buffers(pages[i]);
1436 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
1437 struct btrfs_key key;
1438 struct btrfs_path *path;
1442 path = btrfs_alloc_path();
1444 key.objectid = inode->i_ino;
1445 key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
1447 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
1448 BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
1450 btrfs_file_extent_calc_inline_size(write_bytes);
1451 ret = btrfs_insert_empty_item(trans, root, path, &key,
1454 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1455 path->slots[0], struct btrfs_file_extent_item);
1456 btrfs_set_file_extent_generation(ei, trans->transid);
1457 btrfs_set_file_extent_type(ei,
1458 BTRFS_FILE_EXTENT_INLINE);
1459 ptr = btrfs_file_extent_inline_start(ei);
1460 memcpy(ptr, bh->b_data, offset + write_bytes);
1461 mark_buffer_dirty(path->nodes[0]);
1462 btrfs_free_path(path);
1464 btrfs_csum_file_block(trans, root, inode->i_ino,
1465 pages[i]->index << PAGE_CACHE_SHIFT,
1466 kmap(pages[i]), PAGE_CACHE_SIZE);
1469 SetPageChecked(pages[i]);
1470 ret = btrfs_end_transaction(trans, root);
1472 mutex_unlock(&root->fs_info->fs_mutex);
1474 ret = btrfs_commit_write(file, pages[i], offset,
1475 offset + this_write);
1481 WARN_ON(this_write > write_bytes);
1482 write_bytes -= this_write;
1488 static int drop_extents(struct btrfs_trans_handle *trans,
1489 struct btrfs_root *root,
1490 struct inode *inode,
1494 struct btrfs_key key;
1495 struct btrfs_leaf *leaf;
1497 struct btrfs_file_extent_item *extent;
1500 struct btrfs_file_extent_item old;
1501 struct btrfs_path *path;
1502 u64 search_start = start;
1508 path = btrfs_alloc_path();
1512 btrfs_release_path(root, path);
1513 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1518 if (path->slots[0] == 0) {
1529 leaf = btrfs_buffer_leaf(path->nodes[0]);
1530 slot = path->slots[0];
1531 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
1532 if (key.offset >= end || key.objectid != inode->i_ino) {
1536 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
1540 extent = btrfs_item_ptr(leaf, slot,
1541 struct btrfs_file_extent_item);
1542 found_type = btrfs_file_extent_type(extent);
1543 if (found_type == BTRFS_FILE_EXTENT_REG) {
1544 extent_end = key.offset +
1545 (btrfs_file_extent_num_blocks(extent) <<
1548 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1550 extent_end = key.offset +
1551 btrfs_file_extent_inline_len(leaf->items + slot);
1554 if (!found_extent && !found_inline) {
1559 if (search_start >= extent_end) {
1564 search_start = extent_end;
1566 if (end < extent_end && end >= key.offset) {
1568 memcpy(&old, extent, sizeof(old));
1569 ret = btrfs_inc_extent_ref(trans, root,
1570 btrfs_file_extent_disk_blocknr(&old),
1571 btrfs_file_extent_disk_num_blocks(&old));
1574 WARN_ON(found_inline);
1578 if (start > key.offset) {
1581 /* truncate existing extent */
1583 WARN_ON(start & (root->blocksize - 1));
1585 new_num = (start - key.offset) >>
1587 old_num = btrfs_file_extent_num_blocks(extent);
1588 inode->i_blocks -= (old_num - new_num) << 3;
1589 btrfs_set_file_extent_num_blocks(extent,
1591 mark_buffer_dirty(path->nodes[0]);
1595 ret = btrfs_truncate_item(trans, root, path,
1596 start - key.offset);
1602 u64 disk_blocknr = 0;
1603 u64 disk_num_blocks = 0;
1604 u64 extent_num_blocks = 0;
1607 btrfs_file_extent_disk_blocknr(extent);
1609 btrfs_file_extent_disk_num_blocks(extent);
1611 btrfs_file_extent_num_blocks(extent);
1613 ret = btrfs_del_item(trans, root, path);
1615 btrfs_release_path(root, path);
1618 btrfs_file_extent_num_blocks(extent) << 3;
1619 ret = btrfs_free_extent(trans, root,
1621 disk_num_blocks, 0);
1625 if (!bookend && search_start >= end) {
1632 if (bookend && found_extent) {
1633 /* create bookend */
1634 struct btrfs_key ins;
1635 ins.objectid = inode->i_ino;
1638 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
1640 btrfs_release_path(root, path);
1641 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1644 extent = btrfs_item_ptr(
1645 btrfs_buffer_leaf(path->nodes[0]),
1647 struct btrfs_file_extent_item);
1648 btrfs_set_file_extent_disk_blocknr(extent,
1649 btrfs_file_extent_disk_blocknr(&old));
1650 btrfs_set_file_extent_disk_num_blocks(extent,
1651 btrfs_file_extent_disk_num_blocks(&old));
1653 btrfs_set_file_extent_offset(extent,
1654 btrfs_file_extent_offset(&old) +
1655 ((end - key.offset) >> inode->i_blkbits));
1656 WARN_ON(btrfs_file_extent_num_blocks(&old) <
1657 (end - key.offset) >> inode->i_blkbits);
1658 btrfs_set_file_extent_num_blocks(extent,
1659 btrfs_file_extent_num_blocks(&old) -
1660 ((end - key.offset) >> inode->i_blkbits));
1662 btrfs_set_file_extent_type(extent,
1663 BTRFS_FILE_EXTENT_REG);
1664 btrfs_set_file_extent_generation(extent,
1665 btrfs_file_extent_generation(&old));
1666 btrfs_mark_buffer_dirty(path->nodes[0]);
1668 btrfs_file_extent_num_blocks(extent) << 3;
1674 btrfs_free_path(path);
1678 static int prepare_pages(struct btrfs_root *root,
1680 struct page **pages,
1683 unsigned long first_index,
1684 unsigned long last_index,
1686 u64 alloc_extent_start)
1689 unsigned long index = pos >> PAGE_CACHE_SHIFT;
1690 struct inode *inode = file->f_path.dentry->d_inode;
1694 struct buffer_head *bh;
1695 struct buffer_head *head;
1696 loff_t isize = i_size_read(inode);
1698 memset(pages, 0, num_pages * sizeof(struct page *));
1700 for (i = 0; i < num_pages; i++) {
1701 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1704 goto failed_release;
1706 offset = pos & (PAGE_CACHE_SIZE -1);
1707 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1708 create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
1709 (1 << BH_Uptodate));
1710 head = page_buffers(pages[i]);
1713 err = btrfs_map_bh_to_logical(root, bh,
1714 alloc_extent_start);
1717 goto failed_truncate;
1718 bh = bh->b_this_page;
1719 if (alloc_extent_start)
1720 alloc_extent_start++;
1721 } while (bh != head);
1723 WARN_ON(this_write > write_bytes);
1724 write_bytes -= this_write;
1729 btrfs_drop_pages(pages, num_pages);
1733 btrfs_drop_pages(pages, num_pages);
1735 vmtruncate(inode, isize);
1739 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1740 size_t count, loff_t *ppos)
1743 size_t num_written = 0;
1746 struct inode *inode = file->f_path.dentry->d_inode;
1747 struct btrfs_root *root = BTRFS_I(inode)->root;
1748 struct page *pages[8];
1749 struct page *pinned[2] = { NULL, NULL };
1750 unsigned long first_index;
1751 unsigned long last_index;
1754 u64 alloc_extent_start;
1755 struct btrfs_trans_handle *trans;
1756 struct btrfs_key ins;
1758 if (file->f_flags & O_DIRECT)
1761 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1762 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1763 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1768 err = remove_suid(file->f_path.dentry);
1771 file_update_time(file);
1773 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1774 num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
1777 mutex_lock(&inode->i_mutex);
1778 first_index = pos >> PAGE_CACHE_SHIFT;
1779 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1781 if ((first_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1782 (pos & (PAGE_CACHE_SIZE - 1))) {
1783 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1784 if (!PageUptodate(pinned[0])) {
1785 ret = mpage_readpage(pinned[0], btrfs_get_block);
1788 unlock_page(pinned[0]);
1791 if (first_index != last_index &&
1792 (last_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1793 (count & (PAGE_CACHE_SIZE - 1))) {
1794 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1795 if (!PageUptodate(pinned[1])) {
1796 ret = mpage_readpage(pinned[1], btrfs_get_block);
1799 unlock_page(pinned[1]);
1803 mutex_lock(&root->fs_info->fs_mutex);
1804 trans = btrfs_start_transaction(root, 1);
1807 mutex_unlock(&root->fs_info->fs_mutex);
1810 /* FIXME blocksize != 4096 */
1811 inode->i_blocks += num_blocks << 3;
1812 if (start_pos < inode->i_size) {
1813 /* FIXME blocksize != pagesize */
1814 ret = drop_extents(trans, root, inode,
1816 (pos + count + root->blocksize -1) &
1817 ~((u64)root->blocksize - 1));
1820 if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
1821 pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1822 ret = btrfs_alloc_extent(trans, root, num_blocks, 1,
1825 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1826 start_pos, ins.objectid, ins.offset);
1833 alloc_extent_start = ins.objectid;
1834 ret = btrfs_end_transaction(trans, root);
1835 mutex_unlock(&root->fs_info->fs_mutex);
1838 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1839 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1840 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1843 memset(pages, 0, sizeof(pages));
1844 ret = prepare_pages(root, file, pages, num_pages,
1845 pos, first_index, last_index,
1846 write_bytes, alloc_extent_start);
1849 /* FIXME blocks != pagesize */
1850 if (alloc_extent_start)
1851 alloc_extent_start += num_pages;
1852 ret = btrfs_copy_from_user(pos, num_pages,
1853 write_bytes, pages, buf);
1856 ret = dirty_and_release_pages(NULL, root, file, pages,
1857 num_pages, pos, write_bytes);
1859 btrfs_drop_pages(pages, num_pages);
1862 count -= write_bytes;
1864 num_written += write_bytes;
1866 balance_dirty_pages_ratelimited(inode->i_mapping);
1870 mutex_unlock(&inode->i_mutex);
1873 page_cache_release(pinned[0]);
1875 page_cache_release(pinned[1]);
1877 current->backing_dev_info = NULL;
1878 mark_inode_dirty(inode);
1879 return num_written ? num_written : err;
1882 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1883 unsigned long offset, unsigned long size)
1886 unsigned long left, count = desc->count;
1887 struct inode *inode = page->mapping->host;
1892 if (!PageChecked(page)) {
1893 /* FIXME, do it per block */
1894 struct btrfs_root *root = BTRFS_I(inode)->root;
1896 int ret = btrfs_csum_verify_file_block(root,
1897 page->mapping->host->i_ino,
1898 page->index << PAGE_CACHE_SHIFT,
1899 kmap(page), PAGE_CACHE_SIZE);
1901 printk("failed to verify ino %lu page %lu\n",
1902 page->mapping->host->i_ino,
1904 memset(page_address(page), 0, PAGE_CACHE_SIZE);
1906 SetPageChecked(page);
1910 * Faults on the destination of a read are common, so do it before
1913 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1914 kaddr = kmap_atomic(page, KM_USER0);
1915 left = __copy_to_user_inatomic(desc->arg.buf,
1916 kaddr + offset, size);
1917 kunmap_atomic(kaddr, KM_USER0);
1922 /* Do it the slow way */
1924 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1929 desc->error = -EFAULT;
1932 desc->count = count - size;
1933 desc->written += size;
1934 desc->arg.buf += size;
1939 * btrfs_file_aio_read - filesystem read routine
1940 * @iocb: kernel I/O control block
1941 * @iov: io vector request
1942 * @nr_segs: number of segments in the iovec
1943 * @pos: current file position
1945 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1946 unsigned long nr_segs, loff_t pos)
1948 struct file *filp = iocb->ki_filp;
1952 loff_t *ppos = &iocb->ki_pos;
1955 for (seg = 0; seg < nr_segs; seg++) {
1956 const struct iovec *iv = &iov[seg];
1959 * If any segment has a negative length, or the cumulative
1960 * length ever wraps negative then return -EINVAL.
1962 count += iv->iov_len;
1963 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1965 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1970 count -= iv->iov_len; /* This segment is no good */
1975 for (seg = 0; seg < nr_segs; seg++) {
1976 read_descriptor_t desc;
1979 desc.arg.buf = iov[seg].iov_base;
1980 desc.count = iov[seg].iov_len;
1981 if (desc.count == 0)
1984 do_generic_file_read(filp, ppos, &desc,
1986 retval += desc.written;
1988 retval = retval ?: desc.error;
1996 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1998 struct btrfs_trans_handle *trans;
1999 struct btrfs_key key;
2000 struct btrfs_root_item root_item;
2001 struct btrfs_inode_item *inode_item;
2002 struct buffer_head *subvol;
2003 struct btrfs_leaf *leaf;
2004 struct btrfs_root *new_root;
2005 struct inode *inode;
2008 u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2010 mutex_lock(&root->fs_info->fs_mutex);
2011 trans = btrfs_start_transaction(root, 1);
2014 subvol = btrfs_alloc_free_block(trans, root);
2015 leaf = btrfs_buffer_leaf(subvol);
2016 btrfs_set_header_nritems(&leaf->header, 0);
2017 btrfs_set_header_level(&leaf->header, 0);
2018 btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
2019 btrfs_set_header_generation(&leaf->header, trans->transid);
2020 memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
2021 sizeof(leaf->header.fsid));
2023 inode_item = &root_item.inode;
2024 memset(inode_item, 0, sizeof(*inode_item));
2025 btrfs_set_inode_generation(inode_item, 1);
2026 btrfs_set_inode_size(inode_item, 3);
2027 btrfs_set_inode_nlink(inode_item, 1);
2028 btrfs_set_inode_nblocks(inode_item, 1);
2029 btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
2031 btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
2032 btrfs_set_root_refs(&root_item, 1);
2034 mark_buffer_dirty(subvol);
2038 ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2042 btrfs_set_root_dirid(&root_item, new_dirid);
2044 key.objectid = objectid;
2047 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2048 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2053 * insert the directory item
2055 key.offset = (u64)-1;
2056 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2058 root->fs_info->sb->s_root->d_inode->i_ino,
2062 ret = btrfs_commit_transaction(trans, root);
2065 new_root = btrfs_read_fs_root(root->fs_info, &key);
2068 trans = btrfs_start_transaction(new_root, 1);
2071 inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
2072 inode->i_op = &btrfs_dir_inode_operations;
2073 inode->i_fop = &btrfs_dir_file_operations;
2075 ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
2080 ret = btrfs_update_inode(trans, new_root, inode);
2083 ret = btrfs_commit_transaction(trans, new_root);
2088 mutex_unlock(&root->fs_info->fs_mutex);
2092 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2094 struct btrfs_trans_handle *trans;
2095 struct btrfs_key key;
2096 struct btrfs_root_item new_root_item;
2100 if (!root->ref_cows)
2103 mutex_lock(&root->fs_info->fs_mutex);
2104 trans = btrfs_start_transaction(root, 1);
2107 ret = btrfs_update_inode(trans, root, root->inode);
2110 ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2114 memcpy(&new_root_item, &root->root_item,
2115 sizeof(new_root_item));
2117 key.objectid = objectid;
2120 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2121 btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2123 ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2128 * insert the directory item
2130 key.offset = (u64)-1;
2131 ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2133 root->fs_info->sb->s_root->d_inode->i_ino,
2138 ret = btrfs_inc_root_ref(trans, root);
2141 ret = btrfs_commit_transaction(trans, root);
2143 mutex_unlock(&root->fs_info->fs_mutex);
2147 static int add_disk(struct btrfs_root *root, char *name, int namelen)
2149 struct block_device *bdev;
2150 struct btrfs_path *path;
2151 struct super_block *sb = root->fs_info->sb;
2152 struct btrfs_root *dev_root = root->fs_info->dev_root;
2153 struct btrfs_trans_handle *trans;
2154 struct btrfs_device_item *dev_item;
2155 struct btrfs_key key;
2162 printk("adding disk %s\n", name);
2163 path = btrfs_alloc_path();
2166 num_blocks = btrfs_super_total_blocks(root->fs_info->disk_super);
2167 bdev = open_bdev_excl(name, O_RDWR, sb);
2169 ret = PTR_ERR(bdev);
2170 printk("open bdev excl failed ret %d\n", ret);
2173 set_blocksize(bdev, sb->s_blocksize);
2174 new_blocks = bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2175 key.objectid = num_blocks;
2176 key.offset = new_blocks;
2178 btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
2180 mutex_lock(&dev_root->fs_info->fs_mutex);
2181 trans = btrfs_start_transaction(dev_root, 1);
2182 item_size = sizeof(*dev_item) + namelen;
2183 printk("insert empty on %Lu %Lu %u size %d\n", num_blocks, new_blocks, key.flags, item_size);
2184 ret = btrfs_insert_empty_item(trans, dev_root, path, &key, item_size);
2186 printk("insert failed %d\n", ret);
2187 close_bdev_excl(bdev);
2192 dev_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2193 path->slots[0], struct btrfs_device_item);
2194 btrfs_set_device_pathlen(dev_item, namelen);
2195 memcpy(dev_item + 1, name, namelen);
2197 device_id = btrfs_super_last_device_id(root->fs_info->disk_super) + 1;
2198 btrfs_set_super_last_device_id(root->fs_info->disk_super, device_id);
2199 btrfs_set_device_id(dev_item, device_id);
2200 mark_buffer_dirty(path->nodes[0]);
2202 ret = btrfs_insert_dev_radix(root, bdev, device_id, num_blocks,
2206 btrfs_set_super_total_blocks(root->fs_info->disk_super,
2207 num_blocks + new_blocks);
2208 i_size_write(root->fs_info->btree_inode,
2209 (num_blocks + new_blocks) <<
2210 root->fs_info->btree_inode->i_blkbits);
2214 ret = btrfs_commit_transaction(trans, dev_root);
2216 mutex_unlock(&root->fs_info->fs_mutex);
2218 btrfs_free_path(path);
2223 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2224 cmd, unsigned long arg)
2226 struct btrfs_root *root = BTRFS_I(inode)->root;
2227 struct btrfs_ioctl_vol_args vol_args;
2229 struct btrfs_dir_item *di;
2231 struct btrfs_path *path;
2235 case BTRFS_IOC_SNAP_CREATE:
2236 if (copy_from_user(&vol_args,
2237 (struct btrfs_ioctl_vol_args __user *)arg,
2240 namelen = strlen(vol_args.name);
2241 if (namelen > BTRFS_VOL_NAME_MAX)
2243 path = btrfs_alloc_path();
2246 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2247 mutex_lock(&root->fs_info->fs_mutex);
2248 di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2250 vol_args.name, namelen, 0);
2251 mutex_unlock(&root->fs_info->fs_mutex);
2252 btrfs_free_path(path);
2253 if (di && !IS_ERR(di))
2256 if (root == root->fs_info->tree_root)
2257 ret = create_subvol(root, vol_args.name, namelen);
2259 ret = create_snapshot(root, vol_args.name, namelen);
2262 case BTRFS_IOC_ADD_DISK:
2263 if (copy_from_user(&vol_args,
2264 (struct btrfs_ioctl_vol_args __user *)arg,
2267 namelen = strlen(vol_args.name);
2268 if (namelen > BTRFS_VOL_NAME_MAX)
2270 vol_args.name[namelen] = '\0';
2271 ret = add_disk(root, vol_args.name, namelen);
2279 static struct kmem_cache *btrfs_inode_cachep;
2280 struct kmem_cache *btrfs_trans_handle_cachep;
2281 struct kmem_cache *btrfs_transaction_cachep;
2282 struct kmem_cache *btrfs_bit_radix_cachep;
2283 struct kmem_cache *btrfs_path_cachep;
2286 * Called inside transaction, so use GFP_NOFS
2288 static struct inode *btrfs_alloc_inode(struct super_block *sb)
2290 struct btrfs_inode *ei;
2292 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2295 return &ei->vfs_inode;
2298 static void btrfs_destroy_inode(struct inode *inode)
2300 WARN_ON(!list_empty(&inode->i_dentry));
2301 WARN_ON(inode->i_data.nrpages);
2303 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2306 static void init_once(void * foo, struct kmem_cache * cachep,
2307 unsigned long flags)
2309 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2311 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2312 SLAB_CTOR_CONSTRUCTOR) {
2313 inode_init_once(&ei->vfs_inode);
2317 static int init_inodecache(void)
2319 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
2320 sizeof(struct btrfs_inode),
2321 0, (SLAB_RECLAIM_ACCOUNT|
2324 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
2325 sizeof(struct btrfs_trans_handle),
2326 0, (SLAB_RECLAIM_ACCOUNT|
2329 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
2330 sizeof(struct btrfs_transaction),
2331 0, (SLAB_RECLAIM_ACCOUNT|
2334 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
2335 sizeof(struct btrfs_transaction),
2336 0, (SLAB_RECLAIM_ACCOUNT|
2339 btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
2341 0, (SLAB_RECLAIM_ACCOUNT|
2343 SLAB_DESTROY_BY_RCU),
2345 if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
2346 btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
2351 static void destroy_inodecache(void)
2353 kmem_cache_destroy(btrfs_inode_cachep);
2354 kmem_cache_destroy(btrfs_trans_handle_cachep);
2355 kmem_cache_destroy(btrfs_transaction_cachep);
2356 kmem_cache_destroy(btrfs_bit_radix_cachep);
2357 kmem_cache_destroy(btrfs_path_cachep);
2360 static int btrfs_get_sb(struct file_system_type *fs_type,
2361 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2363 return get_sb_bdev(fs_type, flags, dev_name, data,
2364 btrfs_fill_super, mnt);
2368 static int btrfs_getattr(struct vfsmount *mnt,
2369 struct dentry *dentry, struct kstat *stat)
2371 struct inode *inode = dentry->d_inode;
2372 generic_fillattr(inode, stat);
2373 stat->blksize = 256 * 1024;
2377 static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
2379 struct btrfs_root *root = btrfs_sb(dentry->d_sb);
2380 struct btrfs_super_block *disk_super = root->fs_info->disk_super;
2382 buf->f_namelen = BTRFS_NAME_LEN;
2383 buf->f_blocks = btrfs_super_total_blocks(disk_super);
2384 buf->f_bfree = buf->f_blocks - btrfs_super_blocks_used(disk_super);
2385 buf->f_bavail = buf->f_bfree;
2386 buf->f_bsize = dentry->d_sb->s_blocksize;
2387 buf->f_type = BTRFS_SUPER_MAGIC;
2390 static struct file_system_type btrfs_fs_type = {
2391 .owner = THIS_MODULE,
2393 .get_sb = btrfs_get_sb,
2394 .kill_sb = kill_block_super,
2395 .fs_flags = FS_REQUIRES_DEV,
2398 static struct super_operations btrfs_super_ops = {
2399 .delete_inode = btrfs_delete_inode,
2400 .put_super = btrfs_put_super,
2401 .read_inode = btrfs_read_locked_inode,
2402 .write_super = btrfs_write_super,
2403 .sync_fs = btrfs_sync_fs,
2404 .write_inode = btrfs_write_inode,
2405 .alloc_inode = btrfs_alloc_inode,
2406 .destroy_inode = btrfs_destroy_inode,
2407 .statfs = btrfs_statfs,
2410 static struct inode_operations btrfs_dir_inode_operations = {
2411 .lookup = btrfs_lookup,
2412 .create = btrfs_create,
2413 .unlink = btrfs_unlink,
2414 .mkdir = btrfs_mkdir,
2415 .rmdir = btrfs_rmdir,
2418 static struct inode_operations btrfs_dir_ro_inode_operations = {
2419 .lookup = btrfs_lookup,
2422 static struct file_operations btrfs_dir_file_operations = {
2423 .llseek = generic_file_llseek,
2424 .read = generic_read_dir,
2425 .readdir = btrfs_readdir,
2426 .ioctl = btrfs_ioctl,
2429 static struct address_space_operations btrfs_aops = {
2430 .readpage = btrfs_readpage,
2431 .writepage = btrfs_writepage,
2432 .sync_page = block_sync_page,
2433 .prepare_write = btrfs_prepare_write,
2434 .commit_write = btrfs_commit_write,
2437 static struct inode_operations btrfs_file_inode_operations = {
2438 .truncate = btrfs_truncate,
2439 .getattr = btrfs_getattr,
2442 static struct file_operations btrfs_file_operations = {
2443 .llseek = generic_file_llseek,
2444 .read = do_sync_read,
2445 .aio_read = btrfs_file_aio_read,
2446 .write = btrfs_file_write,
2447 .mmap = generic_file_mmap,
2448 .open = generic_file_open,
2449 .ioctl = btrfs_ioctl,
2450 .fsync = btrfs_sync_file,
2453 static int __init init_btrfs_fs(void)
2456 printk("btrfs loaded!\n");
2457 err = init_inodecache();
2460 kset_set_kset_s(&btrfs_subsys, fs_subsys);
2461 err = subsystem_register(&btrfs_subsys);
2464 return register_filesystem(&btrfs_fs_type);
2466 destroy_inodecache();
2470 static void __exit exit_btrfs_fs(void)
2472 destroy_inodecache();
2473 unregister_filesystem(&btrfs_fs_type);
2474 subsystem_unregister(&btrfs_subsys);
2475 printk("btrfs unloaded\n");
2478 module_init(init_btrfs_fs)
2479 module_exit(exit_btrfs_fs)
2481 MODULE_LICENSE("GPL");