2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #include <linux/buffer_head.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/version.h>
36 #include "transaction.h"
37 #include "btrfs_inode.h"
39 #include "print-tree.h"
42 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
43 struct page **prepared_pages,
44 const char __user * buf)
48 int offset = pos & (PAGE_CACHE_SIZE - 1);
50 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
51 size_t count = min_t(size_t,
52 PAGE_CACHE_SIZE - offset, write_bytes);
53 struct page *page = prepared_pages[i];
54 fault_in_pages_readable(buf, count);
56 /* Copy data from userspace to the current page */
58 page_fault = __copy_from_user(page_address(page) + offset,
60 /* Flush processor's dcache for this page */
61 flush_dcache_page(page);
69 return page_fault ? -EFAULT : 0;
72 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
75 for (i = 0; i < num_pages; i++) {
78 unlock_page(pages[i]);
79 mark_page_accessed(pages[i]);
80 page_cache_release(pages[i]);
84 static int insert_inline_extent(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root, struct inode *inode,
86 u64 offset, ssize_t size,
87 struct page *page, size_t page_offset)
90 struct btrfs_path *path;
92 struct btrfs_file_extent_item *ei;
97 path = btrfs_alloc_path();
101 btrfs_set_trans_block_group(trans, inode);
103 key.objectid = inode->i_ino;
106 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
107 BUG_ON(size >= PAGE_CACHE_SIZE);
108 datasize = btrfs_file_extent_calc_inline_size(size);
110 ret = btrfs_insert_empty_item(trans, root, path, &key,
116 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
117 path->slots[0], struct btrfs_file_extent_item);
118 btrfs_set_file_extent_generation(ei, trans->transid);
119 btrfs_set_file_extent_type(ei,
120 BTRFS_FILE_EXTENT_INLINE);
121 ptr = btrfs_file_extent_inline_start(ei);
123 kaddr = kmap_atomic(page, KM_USER0);
124 btrfs_memcpy(root, path->nodes[0]->b_data,
125 ptr, kaddr + page_offset, size);
126 kunmap_atomic(kaddr, KM_USER0);
127 btrfs_mark_buffer_dirty(path->nodes[0]);
129 btrfs_free_path(path);
133 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
134 struct btrfs_root *root,
143 struct inode *inode = file->f_path.dentry->d_inode;
144 struct extent_map *em;
145 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
149 u64 end_of_last_block;
150 u64 end_pos = pos + write_bytes;
151 loff_t isize = i_size_read(inode);
153 em = alloc_extent_map(GFP_NOFS);
157 em->bdev = inode->i_sb->s_bdev;
159 start_pos = pos & ~((u64)root->blocksize - 1);
160 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
163 end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
164 lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
165 mutex_lock(&root->fs_info->fs_mutex);
166 trans = btrfs_start_transaction(root, 1);
171 btrfs_set_trans_block_group(trans, inode);
172 inode->i_blocks += num_blocks << 3;
175 if ((end_of_last_block & 4095) == 0) {
176 printk("strange end of last %Lu %lu %Lu\n", start_pos, write_bytes, end_of_last_block);
178 set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
180 /* FIXME...EIEIO, ENOSPC and more */
182 /* insert any holes we need to create */
183 if (inode->i_size < start_pos) {
184 u64 last_pos_in_file;
186 u64 mask = root->blocksize - 1;
187 last_pos_in_file = (isize + mask) & ~mask;
188 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
189 hole_size >>= inode->i_blkbits;
190 if (last_pos_in_file < start_pos) {
191 err = btrfs_insert_file_extent(trans, root,
201 * either allocate an extent for the new bytes or setup the key
202 * to show we are doing inline data in the extent
204 if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
205 pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
207 for (i = 0; i < num_pages; i++) {
208 struct page *p = pages[i];
212 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
213 last_end += PAGE_CACHE_SIZE - 1;
214 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
217 struct page *p = pages[0];
218 /* step one, delete the existing extents in this range */
219 /* FIXME blocksize != pagesize */
220 if (start_pos < inode->i_size) {
221 err = btrfs_drop_extents(trans, root, inode, start_pos,
222 (pos + write_bytes + root->blocksize -1) &
223 ~((u64)root->blocksize - 1), &hint_block);
228 err = insert_inline_extent(trans, root, inode, start_pos,
229 end_pos - start_pos, p, 0);
231 em->start = start_pos;
233 em->block_start = EXTENT_MAP_INLINE;
234 em->block_end = EXTENT_MAP_INLINE;
235 add_extent_mapping(em_tree, em);
237 if (end_pos > isize) {
238 i_size_write(inode, end_pos);
239 btrfs_update_inode(trans, root, inode);
242 err = btrfs_end_transaction(trans, root);
244 mutex_unlock(&root->fs_info->fs_mutex);
245 unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
250 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
252 struct extent_map *em;
253 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
256 em = lookup_extent_mapping(em_tree, start, end);
259 remove_extent_mapping(em_tree, em);
262 /* once for the tree*/
269 * this is very complex, but the basic idea is to drop all extents
270 * in the range start - end. hint_block is filled in with a block number
271 * that would be a good hint to the block allocator for this file.
273 * If an extent intersects the range but is not entirely inside the range
274 * it is either truncated or split. Anything entirely inside the range
275 * is deleted from the tree.
277 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root, struct inode *inode,
279 u64 start, u64 end, u64 *hint_block)
282 struct btrfs_key key;
283 struct btrfs_leaf *leaf;
285 struct btrfs_file_extent_item *extent;
288 struct btrfs_file_extent_item old;
289 struct btrfs_path *path;
290 u64 search_start = start;
297 btrfs_drop_extent_cache(inode, start, end - 1);
299 path = btrfs_alloc_path();
304 btrfs_release_path(root, path);
305 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
310 if (path->slots[0] == 0) {
322 leaf = btrfs_buffer_leaf(path->nodes[0]);
323 slot = path->slots[0];
325 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
326 if (key.offset >= end || key.objectid != inode->i_ino) {
329 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
333 search_start = key.offset;
336 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
337 extent = btrfs_item_ptr(leaf, slot,
338 struct btrfs_file_extent_item);
339 found_type = btrfs_file_extent_type(extent);
340 if (found_type == BTRFS_FILE_EXTENT_REG) {
341 extent_end = key.offset +
342 (btrfs_file_extent_num_blocks(extent) <<
345 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
347 extent_end = key.offset +
348 btrfs_file_extent_inline_len(leaf->items +
352 extent_end = search_start;
355 /* we found nothing we can drop */
356 if ((!found_extent && !found_inline) ||
357 search_start >= extent_end) {
360 nritems = btrfs_header_nritems(
361 btrfs_buffer_header(path->nodes[0]));
362 if (slot >= nritems - 1) {
363 nextret = btrfs_next_leaf(root, path);
373 /* FIXME, there's only one inline extent allowed right now */
375 u64 mask = root->blocksize - 1;
376 search_start = (extent_end + mask) & ~mask;
378 search_start = extent_end;
380 if (end < extent_end && end >= key.offset) {
383 btrfs_file_extent_disk_blocknr(extent);
384 u64 disk_num_blocks =
385 btrfs_file_extent_disk_num_blocks(extent);
386 memcpy(&old, extent, sizeof(old));
387 if (disk_blocknr != 0) {
388 ret = btrfs_inc_extent_ref(trans, root,
389 disk_blocknr, disk_num_blocks);
393 WARN_ON(found_inline);
396 /* truncate existing extent */
397 if (start > key.offset) {
401 WARN_ON(start & (root->blocksize - 1));
403 new_num = (start - key.offset) >>
405 old_num = btrfs_file_extent_num_blocks(extent);
407 btrfs_file_extent_disk_blocknr(extent);
408 if (btrfs_file_extent_disk_blocknr(extent)) {
410 (old_num - new_num) << 3;
412 btrfs_set_file_extent_num_blocks(extent,
414 btrfs_mark_buffer_dirty(path->nodes[0]);
419 /* delete the entire extent */
421 u64 disk_blocknr = 0;
422 u64 disk_num_blocks = 0;
423 u64 extent_num_blocks = 0;
426 btrfs_file_extent_disk_blocknr(extent);
428 btrfs_file_extent_disk_num_blocks(extent);
430 btrfs_file_extent_num_blocks(extent);
432 btrfs_file_extent_disk_blocknr(extent);
434 ret = btrfs_del_item(trans, root, path);
435 /* TODO update progress marker and return */
437 btrfs_release_path(root, path);
439 if (found_extent && disk_blocknr != 0) {
440 inode->i_blocks -= extent_num_blocks << 3;
441 ret = btrfs_free_extent(trans, root,
447 if (!bookend && search_start >= end) {
454 /* create bookend, splitting the extent in two */
455 if (bookend && found_extent) {
456 struct btrfs_key ins;
457 ins.objectid = inode->i_ino;
460 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
461 btrfs_release_path(root, path);
462 ret = btrfs_insert_empty_item(trans, root, path, &ins,
466 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
467 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu\n", ret , ins.objectid, ins.flags, ins.offset, start, end, key.offset, extent_end);
470 extent = btrfs_item_ptr(
471 btrfs_buffer_leaf(path->nodes[0]),
473 struct btrfs_file_extent_item);
474 btrfs_set_file_extent_disk_blocknr(extent,
475 btrfs_file_extent_disk_blocknr(&old));
476 btrfs_set_file_extent_disk_num_blocks(extent,
477 btrfs_file_extent_disk_num_blocks(&old));
479 btrfs_set_file_extent_offset(extent,
480 btrfs_file_extent_offset(&old) +
481 ((end - key.offset) >> inode->i_blkbits));
482 WARN_ON(btrfs_file_extent_num_blocks(&old) <
483 (extent_end - end) >> inode->i_blkbits);
484 btrfs_set_file_extent_num_blocks(extent,
485 (extent_end - end) >> inode->i_blkbits);
487 btrfs_set_file_extent_type(extent,
488 BTRFS_FILE_EXTENT_REG);
489 btrfs_set_file_extent_generation(extent,
490 btrfs_file_extent_generation(&old));
491 btrfs_mark_buffer_dirty(path->nodes[0]);
492 if (btrfs_file_extent_disk_blocknr(&old) != 0) {
494 btrfs_file_extent_num_blocks(extent) << 3;
501 btrfs_free_path(path);
506 * this gets pages into the page cache and locks them down
508 static int prepare_pages(struct btrfs_root *root,
513 unsigned long first_index,
514 unsigned long last_index,
518 unsigned long index = pos >> PAGE_CACHE_SHIFT;
519 struct inode *inode = file->f_path.dentry->d_inode;
524 start_pos = pos & ~((u64)root->blocksize - 1);
525 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
528 memset(pages, 0, num_pages * sizeof(struct page *));
530 for (i = 0; i < num_pages; i++) {
531 pages[i] = grab_cache_page(inode->i_mapping, index + i);
536 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
537 wait_on_page_writeback(pages[i]);
538 if (!PagePrivate(pages[i])) {
539 SetPagePrivate(pages[i]);
540 set_page_private(pages[i], 1);
541 WARN_ON(!pages[i]->mapping->a_ops->invalidatepage);
542 page_cache_get(pages[i]);
544 WARN_ON(!PageLocked(pages[i]));
549 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
550 size_t count, loff_t *ppos)
553 size_t num_written = 0;
556 struct inode *inode = file->f_path.dentry->d_inode;
557 struct btrfs_root *root = BTRFS_I(inode)->root;
558 struct page **pages = NULL;
560 struct page *pinned[2];
561 unsigned long first_index;
562 unsigned long last_index;
564 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
565 PAGE_CACHE_SIZE / (sizeof(struct page *)));
568 if (file->f_flags & O_DIRECT)
571 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
572 current->backing_dev_info = inode->i_mapping->backing_dev_info;
573 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
578 err = remove_suid(file->f_path.dentry);
581 file_update_time(file);
583 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
585 mutex_lock(&inode->i_mutex);
586 first_index = pos >> PAGE_CACHE_SHIFT;
587 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
590 * there are lots of better ways to do this, but this code
591 * makes sure the first and last page in the file range are
592 * up to date and ready for cow
594 if ((pos & (PAGE_CACHE_SIZE - 1))) {
595 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
596 if (!PageUptodate(pinned[0])) {
597 ret = btrfs_readpage(NULL, pinned[0]);
599 wait_on_page_locked(pinned[0]);
601 unlock_page(pinned[0]);
604 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
605 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
606 if (!PageUptodate(pinned[1])) {
607 ret = btrfs_readpage(NULL, pinned[1]);
609 wait_on_page_locked(pinned[1]);
611 unlock_page(pinned[1]);
616 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
617 size_t write_bytes = min(count, nrptrs *
618 (size_t)PAGE_CACHE_SIZE -
620 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
623 WARN_ON(num_pages > nrptrs);
624 memset(pages, 0, sizeof(pages));
625 ret = prepare_pages(root, file, pages, num_pages,
626 pos, first_index, last_index,
631 ret = btrfs_copy_from_user(pos, num_pages,
632 write_bytes, pages, buf);
634 btrfs_drop_pages(pages, num_pages);
638 ret = dirty_and_release_pages(NULL, root, file, pages,
639 num_pages, pos, write_bytes);
640 btrfs_drop_pages(pages, num_pages);
645 count -= write_bytes;
647 num_written += write_bytes;
649 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
650 btrfs_btree_balance_dirty(root);
653 mutex_unlock(&inode->i_mutex);
657 page_cache_release(pinned[0]);
659 page_cache_release(pinned[1]);
661 current->backing_dev_info = NULL;
662 return num_written ? num_written : err;
665 static int btrfs_sync_file(struct file *file,
666 struct dentry *dentry, int datasync)
668 struct inode *inode = dentry->d_inode;
669 struct btrfs_root *root = BTRFS_I(inode)->root;
671 struct btrfs_trans_handle *trans;
674 * check the transaction that last modified this inode
675 * and see if its already been committed
677 mutex_lock(&root->fs_info->fs_mutex);
678 if (!BTRFS_I(inode)->last_trans)
680 mutex_lock(&root->fs_info->trans_mutex);
681 if (BTRFS_I(inode)->last_trans <=
682 root->fs_info->last_trans_committed) {
683 BTRFS_I(inode)->last_trans = 0;
684 mutex_unlock(&root->fs_info->trans_mutex);
687 mutex_unlock(&root->fs_info->trans_mutex);
690 * ok we haven't committed the transaction yet, lets do a commit
692 trans = btrfs_start_transaction(root, 1);
697 ret = btrfs_commit_transaction(trans, root);
699 mutex_unlock(&root->fs_info->fs_mutex);
700 return ret > 0 ? EIO : ret;
703 static struct vm_operations_struct btrfs_file_vm_ops = {
704 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
705 .nopage = filemap_nopage,
706 .populate = filemap_populate,
708 .fault = filemap_fault,
710 .page_mkwrite = btrfs_page_mkwrite,
713 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
715 vma->vm_ops = &btrfs_file_vm_ops;
720 struct file_operations btrfs_file_operations = {
721 .llseek = generic_file_llseek,
722 .read = do_sync_read,
723 .aio_read = generic_file_aio_read,
724 .write = btrfs_file_write,
725 .mmap = btrfs_file_mmap,
726 .open = generic_file_open,
727 .ioctl = btrfs_ioctl,
728 .fsync = btrfs_sync_file,
730 .compat_ioctl = btrfs_compat_ioctl,