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;
190 if (last_pos_in_file < start_pos) {
191 err = btrfs_drop_extents(trans, root, inode,
193 last_pos_in_file + hole_size,
198 hole_size >>= inode->i_blkbits;
199 err = btrfs_insert_file_extent(trans, root,
209 * either allocate an extent for the new bytes or setup the key
210 * to show we are doing inline data in the extent
212 if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
213 pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
215 for (i = 0; i < num_pages; i++) {
216 struct page *p = pages[i];
220 last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
221 last_end += PAGE_CACHE_SIZE - 1;
222 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
225 struct page *p = pages[0];
226 /* step one, delete the existing extents in this range */
227 /* FIXME blocksize != pagesize */
228 err = btrfs_drop_extents(trans, root, inode, start_pos,
229 (pos + write_bytes + root->blocksize -1) &
230 ~((u64)root->blocksize - 1), &hint_block);
234 err = insert_inline_extent(trans, root, inode, start_pos,
235 end_pos - start_pos, p, 0);
237 em->start = start_pos;
239 em->block_start = EXTENT_MAP_INLINE;
240 em->block_end = EXTENT_MAP_INLINE;
241 add_extent_mapping(em_tree, em);
243 if (end_pos > isize) {
244 i_size_write(inode, end_pos);
245 btrfs_update_inode(trans, root, inode);
248 err = btrfs_end_transaction(trans, root);
250 mutex_unlock(&root->fs_info->fs_mutex);
251 unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
256 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
258 struct extent_map *em;
259 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
262 em = lookup_extent_mapping(em_tree, start, end);
265 remove_extent_mapping(em_tree, em);
268 /* once for the tree*/
275 * this is very complex, but the basic idea is to drop all extents
276 * in the range start - end. hint_block is filled in with a block number
277 * that would be a good hint to the block allocator for this file.
279 * If an extent intersects the range but is not entirely inside the range
280 * it is either truncated or split. Anything entirely inside the range
281 * is deleted from the tree.
283 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
284 struct btrfs_root *root, struct inode *inode,
285 u64 start, u64 end, u64 *hint_block)
288 struct btrfs_key key;
289 struct btrfs_leaf *leaf;
291 struct btrfs_file_extent_item *extent;
294 struct btrfs_file_extent_item old;
295 struct btrfs_path *path;
296 u64 search_start = start;
303 btrfs_drop_extent_cache(inode, start, end - 1);
305 path = btrfs_alloc_path();
310 btrfs_release_path(root, path);
311 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
316 if (path->slots[0] == 0) {
328 leaf = btrfs_buffer_leaf(path->nodes[0]);
329 slot = path->slots[0];
331 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
332 if (key.offset >= end || key.objectid != inode->i_ino) {
335 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
339 search_start = key.offset;
342 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
343 extent = btrfs_item_ptr(leaf, slot,
344 struct btrfs_file_extent_item);
345 found_type = btrfs_file_extent_type(extent);
346 if (found_type == BTRFS_FILE_EXTENT_REG) {
347 extent_end = key.offset +
348 (btrfs_file_extent_num_blocks(extent) <<
351 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
353 extent_end = key.offset +
354 btrfs_file_extent_inline_len(leaf->items +
358 extent_end = search_start;
361 /* we found nothing we can drop */
362 if ((!found_extent && !found_inline) ||
363 search_start >= extent_end) {
366 nritems = btrfs_header_nritems(
367 btrfs_buffer_header(path->nodes[0]));
368 if (slot >= nritems - 1) {
369 nextret = btrfs_next_leaf(root, path);
379 /* FIXME, there's only one inline extent allowed right now */
381 u64 mask = root->blocksize - 1;
382 search_start = (extent_end + mask) & ~mask;
384 search_start = extent_end;
386 if (end < extent_end && end >= key.offset) {
389 btrfs_file_extent_disk_blocknr(extent);
390 u64 disk_num_blocks =
391 btrfs_file_extent_disk_num_blocks(extent);
392 memcpy(&old, extent, sizeof(old));
393 if (disk_blocknr != 0) {
394 ret = btrfs_inc_extent_ref(trans, root,
395 disk_blocknr, disk_num_blocks);
399 WARN_ON(found_inline);
402 /* truncate existing extent */
403 if (start > key.offset) {
407 WARN_ON(start & (root->blocksize - 1));
409 new_num = (start - key.offset) >>
411 old_num = btrfs_file_extent_num_blocks(extent);
413 btrfs_file_extent_disk_blocknr(extent);
414 if (btrfs_file_extent_disk_blocknr(extent)) {
416 (old_num - new_num) << 3;
418 btrfs_set_file_extent_num_blocks(extent,
420 btrfs_mark_buffer_dirty(path->nodes[0]);
425 /* delete the entire extent */
427 u64 disk_blocknr = 0;
428 u64 disk_num_blocks = 0;
429 u64 extent_num_blocks = 0;
432 btrfs_file_extent_disk_blocknr(extent);
434 btrfs_file_extent_disk_num_blocks(extent);
436 btrfs_file_extent_num_blocks(extent);
438 btrfs_file_extent_disk_blocknr(extent);
440 ret = btrfs_del_item(trans, root, path);
441 /* TODO update progress marker and return */
443 btrfs_release_path(root, path);
445 if (found_extent && disk_blocknr != 0) {
446 inode->i_blocks -= extent_num_blocks << 3;
447 ret = btrfs_free_extent(trans, root,
453 if (!bookend && search_start >= end) {
460 /* create bookend, splitting the extent in two */
461 if (bookend && found_extent) {
462 struct btrfs_key ins;
463 ins.objectid = inode->i_ino;
466 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
467 btrfs_release_path(root, path);
468 ret = btrfs_insert_empty_item(trans, root, path, &ins,
472 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
473 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.flags, ins.offset, start, end, key.offset, extent_end, keep);
476 extent = btrfs_item_ptr(
477 btrfs_buffer_leaf(path->nodes[0]),
479 struct btrfs_file_extent_item);
480 btrfs_set_file_extent_disk_blocknr(extent,
481 btrfs_file_extent_disk_blocknr(&old));
482 btrfs_set_file_extent_disk_num_blocks(extent,
483 btrfs_file_extent_disk_num_blocks(&old));
485 btrfs_set_file_extent_offset(extent,
486 btrfs_file_extent_offset(&old) +
487 ((end - key.offset) >> inode->i_blkbits));
488 WARN_ON(btrfs_file_extent_num_blocks(&old) <
489 (extent_end - end) >> inode->i_blkbits);
490 btrfs_set_file_extent_num_blocks(extent,
491 (extent_end - end) >> inode->i_blkbits);
493 btrfs_set_file_extent_type(extent,
494 BTRFS_FILE_EXTENT_REG);
495 btrfs_set_file_extent_generation(extent,
496 btrfs_file_extent_generation(&old));
497 btrfs_mark_buffer_dirty(path->nodes[0]);
498 if (btrfs_file_extent_disk_blocknr(&old) != 0) {
500 btrfs_file_extent_num_blocks(extent) << 3;
507 btrfs_free_path(path);
512 * this gets pages into the page cache and locks them down
514 static int prepare_pages(struct btrfs_root *root,
519 unsigned long first_index,
520 unsigned long last_index,
524 unsigned long index = pos >> PAGE_CACHE_SHIFT;
525 struct inode *inode = file->f_path.dentry->d_inode;
530 start_pos = pos & ~((u64)root->blocksize - 1);
531 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
534 memset(pages, 0, num_pages * sizeof(struct page *));
536 for (i = 0; i < num_pages; i++) {
537 pages[i] = grab_cache_page(inode->i_mapping, index + i);
542 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
543 wait_on_page_writeback(pages[i]);
544 if (!PagePrivate(pages[i])) {
545 SetPagePrivate(pages[i]);
546 set_page_private(pages[i], 1);
547 WARN_ON(!pages[i]->mapping->a_ops->invalidatepage);
548 page_cache_get(pages[i]);
550 WARN_ON(!PageLocked(pages[i]));
555 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
556 size_t count, loff_t *ppos)
559 size_t num_written = 0;
562 struct inode *inode = file->f_path.dentry->d_inode;
563 struct btrfs_root *root = BTRFS_I(inode)->root;
564 struct page **pages = NULL;
566 struct page *pinned[2];
567 unsigned long first_index;
568 unsigned long last_index;
570 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
571 PAGE_CACHE_SIZE / (sizeof(struct page *)));
574 if (file->f_flags & O_DIRECT)
577 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
578 current->backing_dev_info = inode->i_mapping->backing_dev_info;
579 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
584 err = remove_suid(file->f_path.dentry);
587 file_update_time(file);
589 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
591 mutex_lock(&inode->i_mutex);
592 first_index = pos >> PAGE_CACHE_SHIFT;
593 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
596 * there are lots of better ways to do this, but this code
597 * makes sure the first and last page in the file range are
598 * up to date and ready for cow
600 if ((pos & (PAGE_CACHE_SIZE - 1))) {
601 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
602 if (!PageUptodate(pinned[0])) {
603 ret = btrfs_readpage(NULL, pinned[0]);
605 wait_on_page_locked(pinned[0]);
607 unlock_page(pinned[0]);
610 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
611 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
612 if (!PageUptodate(pinned[1])) {
613 ret = btrfs_readpage(NULL, pinned[1]);
615 wait_on_page_locked(pinned[1]);
617 unlock_page(pinned[1]);
622 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
623 size_t write_bytes = min(count, nrptrs *
624 (size_t)PAGE_CACHE_SIZE -
626 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
629 WARN_ON(num_pages > nrptrs);
630 memset(pages, 0, sizeof(pages));
631 ret = prepare_pages(root, file, pages, num_pages,
632 pos, first_index, last_index,
637 ret = btrfs_copy_from_user(pos, num_pages,
638 write_bytes, pages, buf);
640 btrfs_drop_pages(pages, num_pages);
644 ret = dirty_and_release_pages(NULL, root, file, pages,
645 num_pages, pos, write_bytes);
646 btrfs_drop_pages(pages, num_pages);
651 count -= write_bytes;
653 num_written += write_bytes;
655 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
656 btrfs_btree_balance_dirty(root);
659 mutex_unlock(&inode->i_mutex);
663 page_cache_release(pinned[0]);
665 page_cache_release(pinned[1]);
667 current->backing_dev_info = NULL;
668 return num_written ? num_written : err;
671 static int btrfs_sync_file(struct file *file,
672 struct dentry *dentry, int datasync)
674 struct inode *inode = dentry->d_inode;
675 struct btrfs_root *root = BTRFS_I(inode)->root;
677 struct btrfs_trans_handle *trans;
680 * check the transaction that last modified this inode
681 * and see if its already been committed
683 mutex_lock(&root->fs_info->fs_mutex);
684 if (!BTRFS_I(inode)->last_trans)
686 mutex_lock(&root->fs_info->trans_mutex);
687 if (BTRFS_I(inode)->last_trans <=
688 root->fs_info->last_trans_committed) {
689 BTRFS_I(inode)->last_trans = 0;
690 mutex_unlock(&root->fs_info->trans_mutex);
693 mutex_unlock(&root->fs_info->trans_mutex);
696 * ok we haven't committed the transaction yet, lets do a commit
698 trans = btrfs_start_transaction(root, 1);
703 ret = btrfs_commit_transaction(trans, root);
705 mutex_unlock(&root->fs_info->fs_mutex);
706 return ret > 0 ? EIO : ret;
709 static struct vm_operations_struct btrfs_file_vm_ops = {
710 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
711 .nopage = filemap_nopage,
712 .populate = filemap_populate,
714 .fault = filemap_fault,
716 .page_mkwrite = btrfs_page_mkwrite,
719 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
721 vma->vm_ops = &btrfs_file_vm_ops;
726 struct file_operations btrfs_file_operations = {
727 .llseek = generic_file_llseek,
728 .read = do_sync_read,
729 .aio_read = generic_file_aio_read,
730 .write = btrfs_file_write,
731 .mmap = btrfs_file_mmap,
732 .open = generic_file_open,
733 .ioctl = btrfs_ioctl,
734 .fsync = btrfs_sync_file,
736 .compat_ioctl = btrfs_compat_ioctl,