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/module.h>
20 #include <linux/buffer_head.h>
22 #include <linux/pagemap.h>
23 #include <linux/highmem.h>
24 #include <linux/time.h>
25 #include <linux/init.h>
26 #include <linux/string.h>
27 #include <linux/smp_lock.h>
28 #include <linux/backing-dev.h>
29 #include <linux/mpage.h>
30 #include <linux/swap.h>
31 #include <linux/writeback.h>
32 #include <linux/statfs.h>
33 #include <linux/compat.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 dirty_and_release_pages(struct btrfs_trans_handle *trans,
85 struct btrfs_root *root,
97 struct inode *inode = file->f_path.dentry->d_inode;
98 struct buffer_head *bh;
99 struct btrfs_file_extent_item *ei;
101 for (i = 0; i < num_pages; i++) {
102 offset = pos & (PAGE_CACHE_SIZE -1);
103 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
104 /* FIXME, one block at a time */
106 mutex_lock(&root->fs_info->fs_mutex);
107 trans = btrfs_start_transaction(root, 1);
108 btrfs_set_trans_block_group(trans, inode);
110 bh = page_buffers(pages[i]);
112 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
113 struct btrfs_key key;
114 struct btrfs_path *path;
118 /* create an inline extent, and copy the data in */
119 path = btrfs_alloc_path();
121 key.objectid = inode->i_ino;
122 key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
124 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
125 BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
127 btrfs_file_extent_calc_inline_size(write_bytes);
129 ret = btrfs_insert_empty_item(trans, root, path, &key,
132 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
133 path->slots[0], struct btrfs_file_extent_item);
134 btrfs_set_file_extent_generation(ei, trans->transid);
135 btrfs_set_file_extent_type(ei,
136 BTRFS_FILE_EXTENT_INLINE);
137 ptr = btrfs_file_extent_inline_start(ei);
138 btrfs_memcpy(root, path->nodes[0]->b_data,
139 ptr, bh->b_data, offset + write_bytes);
140 mark_buffer_dirty(path->nodes[0]);
141 btrfs_free_path(path);
142 } else if (buffer_mapped(bh)) {
143 /* csum the file data */
144 btrfs_csum_file_block(trans, root, inode->i_ino,
145 pages[i]->index << PAGE_CACHE_SHIFT,
146 kmap(pages[i]), PAGE_CACHE_SIZE);
149 SetPageChecked(pages[i]);
150 ret = btrfs_end_transaction(trans, root);
152 mutex_unlock(&root->fs_info->fs_mutex);
154 ret = btrfs_commit_write(file, pages[i], offset,
155 offset + this_write);
161 WARN_ON(this_write > write_bytes);
162 write_bytes -= this_write;
169 * this is very complex, but the basic idea is to drop all extents
170 * in the range start - end. hint_block is filled in with a block number
171 * that would be a good hint to the block allocator for this file.
173 * If an extent intersects the range but is not entirely inside the range
174 * it is either truncated or split. Anything entirely inside the range
175 * is deleted from the tree.
177 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
178 struct btrfs_root *root, struct inode *inode,
179 u64 start, u64 end, u64 *hint_block)
182 struct btrfs_key key;
183 struct btrfs_leaf *leaf;
185 struct btrfs_file_extent_item *extent;
188 struct btrfs_file_extent_item old;
189 struct btrfs_path *path;
190 u64 search_start = start;
196 path = btrfs_alloc_path();
200 btrfs_release_path(root, path);
201 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
206 if (path->slots[0] == 0) {
217 leaf = btrfs_buffer_leaf(path->nodes[0]);
218 slot = path->slots[0];
219 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
220 if (key.offset >= end || key.objectid != inode->i_ino) {
224 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
228 extent = btrfs_item_ptr(leaf, slot,
229 struct btrfs_file_extent_item);
230 found_type = btrfs_file_extent_type(extent);
231 if (found_type == BTRFS_FILE_EXTENT_REG) {
232 extent_end = key.offset +
233 (btrfs_file_extent_num_blocks(extent) <<
236 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
238 extent_end = key.offset +
239 btrfs_file_extent_inline_len(leaf->items + slot);
242 /* we found nothing we can drop */
243 if (!found_extent && !found_inline) {
248 /* we found nothing inside the range */
249 if (search_start >= extent_end) {
254 /* FIXME, there's only one inline extent allowed right now */
256 u64 mask = root->blocksize - 1;
257 search_start = (extent_end + mask) & ~mask;
259 search_start = extent_end;
261 if (end < extent_end && end >= key.offset) {
264 btrfs_file_extent_disk_blocknr(extent);
265 u64 disk_num_blocks =
266 btrfs_file_extent_disk_num_blocks(extent);
267 memcpy(&old, extent, sizeof(old));
268 if (disk_blocknr != 0) {
269 ret = btrfs_inc_extent_ref(trans, root,
270 disk_blocknr, disk_num_blocks);
274 WARN_ON(found_inline);
278 /* truncate existing extent */
279 if (start > key.offset) {
283 WARN_ON(start & (root->blocksize - 1));
285 new_num = (start - key.offset) >>
287 old_num = btrfs_file_extent_num_blocks(extent);
289 btrfs_file_extent_disk_blocknr(extent);
290 if (btrfs_file_extent_disk_blocknr(extent)) {
292 (old_num - new_num) << 3;
294 btrfs_set_file_extent_num_blocks(extent,
296 mark_buffer_dirty(path->nodes[0]);
301 /* delete the entire extent */
303 u64 disk_blocknr = 0;
304 u64 disk_num_blocks = 0;
305 u64 extent_num_blocks = 0;
308 btrfs_file_extent_disk_blocknr(extent);
310 btrfs_file_extent_disk_num_blocks(extent);
312 btrfs_file_extent_num_blocks(extent);
314 btrfs_file_extent_disk_blocknr(extent);
316 ret = btrfs_del_item(trans, root, path);
318 btrfs_release_path(root, path);
320 if (found_extent && disk_blocknr != 0) {
321 inode->i_blocks -= extent_num_blocks << 3;
322 ret = btrfs_free_extent(trans, root,
328 if (!bookend && search_start >= end) {
335 /* create bookend, splitting the extent in two */
336 if (bookend && found_extent) {
337 struct btrfs_key ins;
338 ins.objectid = inode->i_ino;
341 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
343 btrfs_release_path(root, path);
344 ret = btrfs_insert_empty_item(trans, root, path, &ins,
347 extent = btrfs_item_ptr(
348 btrfs_buffer_leaf(path->nodes[0]),
350 struct btrfs_file_extent_item);
351 btrfs_set_file_extent_disk_blocknr(extent,
352 btrfs_file_extent_disk_blocknr(&old));
353 btrfs_set_file_extent_disk_num_blocks(extent,
354 btrfs_file_extent_disk_num_blocks(&old));
356 btrfs_set_file_extent_offset(extent,
357 btrfs_file_extent_offset(&old) +
358 ((end - key.offset) >> inode->i_blkbits));
359 WARN_ON(btrfs_file_extent_num_blocks(&old) <
360 (extent_end - end) >> inode->i_blkbits);
361 btrfs_set_file_extent_num_blocks(extent,
362 (extent_end - end) >> inode->i_blkbits);
364 btrfs_set_file_extent_type(extent,
365 BTRFS_FILE_EXTENT_REG);
366 btrfs_set_file_extent_generation(extent,
367 btrfs_file_extent_generation(&old));
368 btrfs_mark_buffer_dirty(path->nodes[0]);
369 if (btrfs_file_extent_disk_blocknr(&old) != 0) {
371 btrfs_file_extent_num_blocks(extent) << 3;
378 btrfs_free_path(path);
383 * this gets pages into the page cache and locks them down
385 static int prepare_pages(struct btrfs_root *root,
390 unsigned long first_index,
391 unsigned long last_index,
393 u64 alloc_extent_start)
396 unsigned long index = pos >> PAGE_CACHE_SHIFT;
397 struct inode *inode = file->f_path.dentry->d_inode;
401 struct buffer_head *bh;
402 struct buffer_head *head;
403 loff_t isize = i_size_read(inode);
405 memset(pages, 0, num_pages * sizeof(struct page *));
407 for (i = 0; i < num_pages; i++) {
408 pages[i] = grab_cache_page(inode->i_mapping, index + i);
413 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
414 wait_on_page_writeback(pages[i]);
415 offset = pos & (PAGE_CACHE_SIZE -1);
416 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
417 if (!page_has_buffers(pages[i])) {
418 create_empty_buffers(pages[i],
419 root->fs_info->sb->s_blocksize,
422 head = page_buffers(pages[i]);
425 err = btrfs_map_bh_to_logical(root, bh,
429 goto failed_truncate;
430 bh = bh->b_this_page;
431 if (alloc_extent_start)
432 alloc_extent_start++;
433 } while (bh != head);
435 WARN_ON(this_write > write_bytes);
436 write_bytes -= this_write;
441 btrfs_drop_pages(pages, num_pages);
445 btrfs_drop_pages(pages, num_pages);
447 vmtruncate(inode, isize);
451 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
452 size_t count, loff_t *ppos)
455 size_t num_written = 0;
458 struct inode *inode = file->f_path.dentry->d_inode;
459 struct btrfs_root *root = BTRFS_I(inode)->root;
460 struct page *pages[8];
461 struct page *pinned[2];
462 unsigned long first_index;
463 unsigned long last_index;
466 u64 alloc_extent_start;
468 struct btrfs_trans_handle *trans;
469 struct btrfs_key ins;
472 if (file->f_flags & O_DIRECT)
475 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
476 current->backing_dev_info = inode->i_mapping->backing_dev_info;
477 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
482 err = remove_suid(file->f_path.dentry);
485 file_update_time(file);
487 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
488 num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
491 mutex_lock(&inode->i_mutex);
492 first_index = pos >> PAGE_CACHE_SHIFT;
493 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
496 * there are lots of better ways to do this, but this code
497 * makes sure the first and last page in the file range are
498 * up to date and ready for cow
500 if ((pos & (PAGE_CACHE_SIZE - 1))) {
501 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
502 if (!PageUptodate(pinned[0])) {
503 ret = mpage_readpage(pinned[0], btrfs_get_block);
505 wait_on_page_locked(pinned[0]);
507 unlock_page(pinned[0]);
510 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
511 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
512 if (!PageUptodate(pinned[1])) {
513 ret = mpage_readpage(pinned[1], btrfs_get_block);
515 wait_on_page_locked(pinned[1]);
517 unlock_page(pinned[1]);
521 mutex_lock(&root->fs_info->fs_mutex);
522 trans = btrfs_start_transaction(root, 1);
525 mutex_unlock(&root->fs_info->fs_mutex);
528 btrfs_set_trans_block_group(trans, inode);
529 /* FIXME blocksize != 4096 */
530 inode->i_blocks += num_blocks << 3;
533 /* FIXME...EIEIO, ENOSPC and more */
535 /* step one, delete the existing extents in this range */
536 if (start_pos < inode->i_size) {
537 /* FIXME blocksize != pagesize */
538 ret = btrfs_drop_extents(trans, root, inode,
540 (pos + count + root->blocksize -1) &
541 ~((u64)root->blocksize - 1),
546 /* insert any holes we need to create */
547 if (inode->i_size < start_pos) {
548 u64 last_pos_in_file;
550 u64 mask = root->blocksize - 1;
551 last_pos_in_file = (inode->i_size + mask) & ~mask;
552 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
553 hole_size >>= inode->i_blkbits;
554 if (last_pos_in_file < start_pos) {
555 ret = btrfs_insert_file_extent(trans, root,
564 * either allocate an extent for the new bytes or setup the key
565 * to show we are doing inline data in the extent
567 if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
568 pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
569 ret = btrfs_alloc_extent(trans, root, inode->i_ino,
570 num_blocks, hint_block, (u64)-1,
573 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
574 start_pos, ins.objectid, ins.offset,
582 alloc_extent_start = ins.objectid;
583 ret = btrfs_end_transaction(trans, root);
584 mutex_unlock(&root->fs_info->fs_mutex);
587 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
588 size_t write_bytes = min(count,
589 (size_t)PAGE_CACHE_SIZE - offset);
590 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
593 memset(pages, 0, sizeof(pages));
594 ret = prepare_pages(root, file, pages, num_pages,
595 pos, first_index, last_index,
596 write_bytes, alloc_extent_start);
599 /* FIXME blocks != pagesize */
600 if (alloc_extent_start)
601 alloc_extent_start += num_pages;
602 ret = btrfs_copy_from_user(pos, num_pages,
603 write_bytes, pages, buf);
606 ret = dirty_and_release_pages(NULL, root, file, pages,
607 num_pages, pos, write_bytes);
609 btrfs_drop_pages(pages, num_pages);
612 count -= write_bytes;
614 num_written += write_bytes;
616 balance_dirty_pages_ratelimited(inode->i_mapping);
617 btrfs_btree_balance_dirty(root);
621 mutex_unlock(&inode->i_mutex);
624 page_cache_release(pinned[0]);
626 page_cache_release(pinned[1]);
628 current->backing_dev_info = NULL;
629 mark_inode_dirty(inode);
630 return num_written ? num_written : err;
634 * FIXME, do this by stuffing the csum we want in the info hanging off
635 * page->private. For now, verify file csums on read
637 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
638 unsigned long offset, unsigned long size)
641 unsigned long left, count = desc->count;
642 struct inode *inode = page->mapping->host;
647 if (!PageChecked(page)) {
648 /* FIXME, do it per block */
649 struct btrfs_root *root = BTRFS_I(inode)->root;
651 struct buffer_head *bh;
653 if (page_has_buffers(page)) {
654 bh = page_buffers(page);
655 if (!buffer_mapped(bh)) {
656 SetPageChecked(page);
661 ret = btrfs_csum_verify_file_block(root,
662 page->mapping->host->i_ino,
663 page->index << PAGE_CACHE_SHIFT,
664 kmap(page), PAGE_CACHE_SIZE);
666 if (ret != -ENOENT) {
667 printk("failed to verify ino %lu page %lu ret %d\n",
668 page->mapping->host->i_ino,
670 memset(page_address(page), 1, PAGE_CACHE_SIZE);
671 flush_dcache_page(page);
674 SetPageChecked(page);
679 * Faults on the destination of a read are common, so do it before
682 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
683 kaddr = kmap_atomic(page, KM_USER0);
684 left = __copy_to_user_inatomic(desc->arg.buf,
685 kaddr + offset, size);
686 kunmap_atomic(kaddr, KM_USER0);
691 /* Do it the slow way */
693 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
698 desc->error = -EFAULT;
701 desc->count = count - size;
702 desc->written += size;
703 desc->arg.buf += size;
708 * btrfs_file_aio_read - filesystem read routine, with a mod to csum verify
709 * @iocb: kernel I/O control block
710 * @iov: io vector request
711 * @nr_segs: number of segments in the iovec
712 * @pos: current file position
714 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
715 unsigned long nr_segs, loff_t pos)
717 struct file *filp = iocb->ki_filp;
721 loff_t *ppos = &iocb->ki_pos;
724 for (seg = 0; seg < nr_segs; seg++) {
725 const struct iovec *iv = &iov[seg];
728 * If any segment has a negative length, or the cumulative
729 * length ever wraps negative then return -EINVAL.
731 count += iv->iov_len;
732 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
734 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
739 count -= iv->iov_len; /* This segment is no good */
744 for (seg = 0; seg < nr_segs; seg++) {
745 read_descriptor_t desc;
748 desc.arg.buf = iov[seg].iov_base;
749 desc.count = iov[seg].iov_len;
753 do_generic_file_read(filp, ppos, &desc,
755 retval += desc.written;
757 retval = retval ?: desc.error;
765 static int btrfs_sync_file(struct file *file,
766 struct dentry *dentry, int datasync)
768 struct inode *inode = dentry->d_inode;
769 struct btrfs_root *root = BTRFS_I(inode)->root;
771 struct btrfs_trans_handle *trans;
774 * FIXME, use inode generation number to check if we can skip the
777 mutex_lock(&root->fs_info->fs_mutex);
778 trans = btrfs_start_transaction(root, 1);
783 ret = btrfs_commit_transaction(trans, root);
784 mutex_unlock(&root->fs_info->fs_mutex);
786 return ret > 0 ? EIO : ret;
789 struct file_operations btrfs_file_operations = {
790 .llseek = generic_file_llseek,
791 .read = do_sync_read,
792 .aio_read = btrfs_file_aio_read,
793 .write = btrfs_file_write,
794 .mmap = generic_file_mmap,
795 .open = generic_file_open,
796 .ioctl = btrfs_ioctl,
797 .fsync = btrfs_sync_file,
799 .compat_ioctl = btrfs_compat_ioctl,