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.
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/time.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/smp_lock.h>
26 #include <linux/backing-dev.h>
27 #include <linux/mpage.h>
28 #include <linux/swap.h>
29 #include <linux/writeback.h>
30 #include <linux/statfs.h>
31 #include <linux/compat.h>
32 #include <linux/version.h>
35 #include "transaction.h"
36 #include "btrfs_inode.h"
37 #include "ordered-data.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 noinline insert_inline_extent(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root, struct inode *inode,
86 u64 offset, size_t size,
87 struct page **pages, size_t page_offset,
91 struct btrfs_path *path;
92 struct extent_buffer *leaf;
95 struct btrfs_file_extent_item *ei;
103 path = btrfs_alloc_path();
107 btrfs_set_trans_block_group(trans, inode);
109 key.objectid = inode->i_ino;
111 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
113 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
119 struct btrfs_key found_key;
121 if (path->slots[0] == 0)
125 leaf = path->nodes[0];
126 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
128 if (found_key.objectid != inode->i_ino)
131 if (found_key.type != BTRFS_EXTENT_DATA_KEY)
133 ei = btrfs_item_ptr(leaf, path->slots[0],
134 struct btrfs_file_extent_item);
136 if (btrfs_file_extent_type(leaf, ei) !=
137 BTRFS_FILE_EXTENT_INLINE) {
140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
147 leaf = path->nodes[0];
148 ei = btrfs_item_ptr(leaf, path->slots[0],
149 struct btrfs_file_extent_item);
151 if (btrfs_file_extent_type(leaf, ei) !=
152 BTRFS_FILE_EXTENT_INLINE) {
154 btrfs_print_leaf(root, leaf);
155 printk("found wasn't inline offset %Lu inode %lu\n",
156 offset, inode->i_ino);
159 found_size = btrfs_file_extent_inline_len(leaf,
160 btrfs_item_nr(leaf, path->slots[0]));
161 found_end = key.offset + found_size;
163 if (found_end < offset + size) {
164 btrfs_release_path(root, path);
165 ret = btrfs_search_slot(trans, root, &key, path,
166 offset + size - found_end, 1);
169 ret = btrfs_extend_item(trans, root, path,
170 offset + size - found_end);
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
179 if (found_end < offset) {
180 ptr = btrfs_file_extent_inline_start(ei) + found_size;
181 memset_extent_buffer(leaf, 0, ptr, offset - found_end);
185 btrfs_release_path(root, path);
186 datasize = offset + size - key.offset;
187 datasize = btrfs_file_extent_calc_inline_size(datasize);
188 ret = btrfs_insert_empty_item(trans, root, path, &key,
192 printk("got bad ret %d\n", ret);
195 leaf = path->nodes[0];
196 ei = btrfs_item_ptr(leaf, path->slots[0],
197 struct btrfs_file_extent_item);
198 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
199 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
201 ptr = btrfs_file_extent_inline_start(ei) + offset - key.offset;
207 kaddr = kmap_atomic(page, KM_USER0);
208 cur_size = min_t(size_t, PAGE_CACHE_SIZE - page_offset, size);
209 write_extent_buffer(leaf, kaddr + page_offset, ptr, cur_size);
210 kunmap_atomic(kaddr, KM_USER0);
214 if (i >= num_pages) {
215 printk("i %d num_pages %d\n", i, num_pages);
219 btrfs_mark_buffer_dirty(leaf);
221 btrfs_free_path(path);
225 static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
226 struct btrfs_root *root,
235 struct inode *inode = fdentry(file)->d_inode;
236 struct extent_map *em;
237 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
241 u64 end_of_last_block;
242 u64 end_pos = pos + write_bytes;
244 loff_t isize = i_size_read(inode);
245 em = alloc_extent_map(GFP_NOFS);
249 em->bdev = inode->i_sb->s_bdev;
251 start_pos = pos & ~((u64)root->sectorsize - 1);
252 num_bytes = (write_bytes + pos - start_pos +
253 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
255 end_of_last_block = start_pos + num_bytes - 1;
257 lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
258 mutex_lock(&root->fs_info->fs_mutex);
259 trans = btrfs_start_transaction(root, 1);
264 btrfs_set_trans_block_group(trans, inode);
265 inode->i_blocks += num_bytes >> 9;
268 if ((end_of_last_block & 4095) == 0) {
269 printk("strange end of last %Lu %zu %Lu\n", start_pos, write_bytes, end_of_last_block);
271 set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
273 /* FIXME...EIEIO, ENOSPC and more */
275 /* insert any holes we need to create */
276 if (inode->i_size < start_pos) {
277 u64 last_pos_in_file;
279 u64 mask = root->sectorsize - 1;
280 last_pos_in_file = (isize + mask) & ~mask;
281 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
283 if (last_pos_in_file < start_pos) {
284 err = btrfs_drop_extents(trans, root, inode,
286 last_pos_in_file + hole_size,
292 err = btrfs_insert_file_extent(trans, root,
302 * either allocate an extent for the new bytes or setup the key
303 * to show we are doing inline data in the extent
305 inline_size = end_pos;
306 if (isize >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
307 inline_size > 8192 ||
308 inline_size >= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
310 u64 existing_delalloc = 0;
312 for (i = 0; i < num_pages; i++) {
313 struct page *p = pages[i];
317 last_end = (u64)(pages[num_pages -1]->index) <<
319 last_end += PAGE_CACHE_SIZE - 1;
320 if (start_pos < isize) {
321 u64 delalloc_start = start_pos;
322 existing_delalloc = count_range_bits(em_tree,
324 end_of_last_block, (u64)-1,
327 set_extent_delalloc(em_tree, start_pos, end_of_last_block,
329 spin_lock(&root->fs_info->delalloc_lock);
330 root->fs_info->delalloc_bytes += (end_of_last_block + 1 -
331 start_pos) - existing_delalloc;
332 spin_unlock(&root->fs_info->delalloc_lock);
333 btrfs_add_ordered_inode(inode);
336 /* step one, delete the existing extents in this range */
337 aligned_end = (pos + write_bytes + root->sectorsize - 1) &
338 ~((u64)root->sectorsize - 1);
339 err = btrfs_drop_extents(trans, root, inode, start_pos,
340 aligned_end, aligned_end, &hint_byte);
343 if (isize > inline_size)
344 inline_size = min_t(u64, isize, aligned_end);
345 inline_size -= start_pos;
346 err = insert_inline_extent(trans, root, inode, start_pos,
347 inline_size, pages, 0, num_pages);
350 if (end_pos > isize) {
351 i_size_write(inode, end_pos);
352 btrfs_update_inode(trans, root, inode);
355 err = btrfs_end_transaction(trans, root);
357 mutex_unlock(&root->fs_info->fs_mutex);
358 unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
363 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
365 struct extent_map *em;
366 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
369 em = lookup_extent_mapping(em_tree, start, end);
372 remove_extent_mapping(em_tree, em);
375 /* once for the tree*/
382 * this is very complex, but the basic idea is to drop all extents
383 * in the range start - end. hint_block is filled in with a block number
384 * that would be a good hint to the block allocator for this file.
386 * If an extent intersects the range but is not entirely inside the range
387 * it is either truncated or split. Anything entirely inside the range
388 * is deleted from the tree.
390 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
391 struct btrfs_root *root, struct inode *inode,
392 u64 start, u64 end, u64 inline_limit, u64 *hint_byte)
395 u64 search_start = start;
396 struct extent_buffer *leaf;
397 struct btrfs_file_extent_item *extent;
398 struct btrfs_path *path;
399 struct btrfs_key key;
400 struct btrfs_file_extent_item old;
410 btrfs_drop_extent_cache(inode, start, end - 1);
412 path = btrfs_alloc_path();
417 btrfs_release_path(root, path);
418 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
423 if (path->slots[0] == 0) {
435 leaf = path->nodes[0];
436 slot = path->slots[0];
438 btrfs_item_key_to_cpu(leaf, &key, slot);
439 if (key.offset >= end || key.objectid != inode->i_ino) {
442 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
446 search_start = key.offset;
449 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
450 extent = btrfs_item_ptr(leaf, slot,
451 struct btrfs_file_extent_item);
452 found_type = btrfs_file_extent_type(leaf, extent);
453 if (found_type == BTRFS_FILE_EXTENT_REG) {
455 btrfs_file_extent_disk_bytenr(leaf,
458 *hint_byte = extent_end;
460 extent_end = key.offset +
461 btrfs_file_extent_num_bytes(leaf, extent);
463 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
464 struct btrfs_item *item;
465 item = btrfs_item_nr(leaf, slot);
467 extent_end = key.offset +
468 btrfs_file_extent_inline_len(leaf, item);
471 extent_end = search_start;
474 /* we found nothing we can drop */
475 if ((!found_extent && !found_inline) ||
476 search_start >= extent_end) {
479 nritems = btrfs_header_nritems(leaf);
480 if (slot >= nritems - 1) {
481 nextret = btrfs_next_leaf(root, path);
492 u64 mask = root->sectorsize - 1;
493 search_start = (extent_end + mask) & ~mask;
495 search_start = extent_end;
496 if (end <= extent_end && start >= key.offset && found_inline) {
497 *hint_byte = EXTENT_MAP_INLINE;
500 if (end < extent_end && end >= key.offset) {
503 btrfs_file_extent_disk_bytenr(leaf, extent);
505 btrfs_file_extent_disk_num_bytes(leaf,
507 read_extent_buffer(leaf, &old,
508 (unsigned long)extent,
510 if (disk_bytenr != 0) {
511 ret = btrfs_inc_extent_ref(trans, root,
512 disk_bytenr, disk_num_bytes,
513 root->root_key.objectid,
520 if (found_inline && start <= key.offset &&
521 inline_limit < extent_end)
524 /* truncate existing extent */
525 if (start > key.offset) {
529 WARN_ON(start & (root->sectorsize - 1));
531 new_num = start - key.offset;
532 old_num = btrfs_file_extent_num_bytes(leaf,
535 btrfs_file_extent_disk_bytenr(leaf,
537 if (btrfs_file_extent_disk_bytenr(leaf,
540 (old_num - new_num) >> 9;
542 btrfs_set_file_extent_num_bytes(leaf, extent,
544 btrfs_mark_buffer_dirty(leaf);
545 } else if (key.offset < inline_limit &&
546 (end > extent_end) &&
547 (inline_limit < extent_end)) {
549 new_size = btrfs_file_extent_calc_inline_size(
550 inline_limit - key.offset);
551 btrfs_truncate_item(trans, root, path,
555 /* delete the entire extent */
558 u64 disk_num_bytes = 0;
559 u64 extent_num_bytes = 0;
563 root_gen = btrfs_header_generation(leaf);
564 root_owner = btrfs_header_owner(leaf);
567 btrfs_file_extent_disk_bytenr(leaf,
570 btrfs_file_extent_disk_num_bytes(leaf,
573 btrfs_file_extent_num_bytes(leaf, extent);
575 btrfs_file_extent_disk_bytenr(leaf,
578 ret = btrfs_del_item(trans, root, path);
579 /* TODO update progress marker and return */
581 btrfs_release_path(root, path);
583 if (found_extent && disk_bytenr != 0) {
584 inode->i_blocks -= extent_num_bytes >> 9;
585 ret = btrfs_free_extent(trans, root,
589 root_gen, inode->i_ino,
594 if (!bookend && search_start >= end) {
601 if (bookend && found_inline && start <= key.offset &&
602 inline_limit < extent_end && key.offset <= inline_limit) {
604 new_size = btrfs_file_extent_calc_inline_size(
605 extent_end - inline_limit);
606 btrfs_truncate_item(trans, root, path, new_size, 0);
608 /* create bookend, splitting the extent in two */
609 if (bookend && found_extent) {
610 struct btrfs_key ins;
611 ins.objectid = inode->i_ino;
613 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
614 btrfs_release_path(root, path);
615 ret = btrfs_insert_empty_item(trans, root, path, &ins,
618 leaf = path->nodes[0];
620 btrfs_print_leaf(root, leaf);
621 printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu keep was %d\n", ret , ins.objectid, ins.type, ins.offset, start, end, key.offset, extent_end, keep);
624 extent = btrfs_item_ptr(leaf, path->slots[0],
625 struct btrfs_file_extent_item);
626 write_extent_buffer(leaf, &old,
627 (unsigned long)extent, sizeof(old));
629 btrfs_set_file_extent_offset(leaf, extent,
630 le64_to_cpu(old.offset) + end - key.offset);
631 WARN_ON(le64_to_cpu(old.num_bytes) <
633 btrfs_set_file_extent_num_bytes(leaf, extent,
635 btrfs_set_file_extent_type(leaf, extent,
636 BTRFS_FILE_EXTENT_REG);
638 btrfs_mark_buffer_dirty(path->nodes[0]);
639 if (le64_to_cpu(old.disk_bytenr) != 0) {
641 btrfs_file_extent_num_bytes(leaf,
649 btrfs_free_path(path);
654 * this gets pages into the page cache and locks them down
656 static int prepare_pages(struct btrfs_root *root, struct file *file,
657 struct page **pages, size_t num_pages,
658 loff_t pos, unsigned long first_index,
659 unsigned long last_index, size_t write_bytes)
662 unsigned long index = pos >> PAGE_CACHE_SHIFT;
663 struct inode *inode = fdentry(file)->d_inode;
667 start_pos = pos & ~((u64)root->sectorsize - 1);
669 memset(pages, 0, num_pages * sizeof(struct page *));
671 for (i = 0; i < num_pages; i++) {
672 pages[i] = grab_cache_page(inode->i_mapping, index + i);
677 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
678 ClearPageDirty(pages[i]);
680 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
682 wait_on_page_writeback(pages[i]);
683 set_page_extent_mapped(pages[i]);
684 WARN_ON(!PageLocked(pages[i]));
689 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
690 size_t count, loff_t *ppos)
694 ssize_t num_written = 0;
697 struct inode *inode = fdentry(file)->d_inode;
698 struct btrfs_root *root = BTRFS_I(inode)->root;
699 struct page **pages = NULL;
701 struct page *pinned[2];
702 unsigned long first_index;
703 unsigned long last_index;
705 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
706 PAGE_CACHE_SIZE / (sizeof(struct page *)));
709 if (file->f_flags & O_DIRECT)
715 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
716 current->backing_dev_info = inode->i_mapping->backing_dev_info;
717 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
722 err = remove_suid(fdentry(file));
725 file_update_time(file);
727 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
729 mutex_lock(&inode->i_mutex);
730 first_index = pos >> PAGE_CACHE_SHIFT;
731 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
734 * there are lots of better ways to do this, but this code
735 * makes sure the first and last page in the file range are
736 * up to date and ready for cow
738 if ((pos & (PAGE_CACHE_SIZE - 1))) {
739 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
740 if (!PageUptodate(pinned[0])) {
741 ret = btrfs_readpage(NULL, pinned[0]);
743 wait_on_page_locked(pinned[0]);
745 unlock_page(pinned[0]);
748 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
749 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
750 if (!PageUptodate(pinned[1])) {
751 ret = btrfs_readpage(NULL, pinned[1]);
753 wait_on_page_locked(pinned[1]);
755 unlock_page(pinned[1]);
760 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
761 size_t write_bytes = min(count, nrptrs *
762 (size_t)PAGE_CACHE_SIZE -
764 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
767 WARN_ON(num_pages > nrptrs);
768 memset(pages, 0, sizeof(pages));
770 mutex_lock(&root->fs_info->fs_mutex);
771 ret = btrfs_check_free_space(root, write_bytes, 0);
772 mutex_unlock(&root->fs_info->fs_mutex);
776 ret = prepare_pages(root, file, pages, num_pages,
777 pos, first_index, last_index,
782 ret = btrfs_copy_from_user(pos, num_pages,
783 write_bytes, pages, buf);
785 btrfs_drop_pages(pages, num_pages);
789 ret = dirty_and_release_pages(NULL, root, file, pages,
790 num_pages, pos, write_bytes);
791 btrfs_drop_pages(pages, num_pages);
796 count -= write_bytes;
798 num_written += write_bytes;
800 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
801 if (num_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
802 btrfs_btree_balance_dirty(root, 1);
806 mutex_unlock(&inode->i_mutex);
811 page_cache_release(pinned[0]);
813 page_cache_release(pinned[1]);
816 if (num_written > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
817 err = sync_page_range(inode, inode->i_mapping,
818 start_pos, num_written);
822 current->backing_dev_info = NULL;
823 return num_written ? num_written : err;
826 static int btrfs_sync_file(struct file *file,
827 struct dentry *dentry, int datasync)
829 struct inode *inode = dentry->d_inode;
830 struct btrfs_root *root = BTRFS_I(inode)->root;
832 struct btrfs_trans_handle *trans;
835 * check the transaction that last modified this inode
836 * and see if its already been committed
838 mutex_lock(&root->fs_info->fs_mutex);
839 if (!BTRFS_I(inode)->last_trans)
841 mutex_lock(&root->fs_info->trans_mutex);
842 if (BTRFS_I(inode)->last_trans <=
843 root->fs_info->last_trans_committed) {
844 BTRFS_I(inode)->last_trans = 0;
845 mutex_unlock(&root->fs_info->trans_mutex);
848 mutex_unlock(&root->fs_info->trans_mutex);
851 * ok we haven't committed the transaction yet, lets do a commit
853 trans = btrfs_start_transaction(root, 1);
858 ret = btrfs_commit_transaction(trans, root);
860 mutex_unlock(&root->fs_info->fs_mutex);
861 return ret > 0 ? EIO : ret;
864 static struct vm_operations_struct btrfs_file_vm_ops = {
865 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
866 .nopage = filemap_nopage,
867 .populate = filemap_populate,
869 .fault = filemap_fault,
871 .page_mkwrite = btrfs_page_mkwrite,
874 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
876 vma->vm_ops = &btrfs_file_vm_ops;
881 struct file_operations btrfs_file_operations = {
882 .llseek = generic_file_llseek,
883 .read = do_sync_read,
884 .aio_read = generic_file_aio_read,
885 .splice_read = generic_file_splice_read,
886 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
887 .sendfile = generic_file_sendfile,
889 .write = btrfs_file_write,
890 .mmap = btrfs_file_mmap,
891 .open = generic_file_open,
892 .fsync = btrfs_sync_file,
893 .unlocked_ioctl = btrfs_ioctl,
895 .compat_ioctl = btrfs_ioctl,