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 bh = page_buffers(pages[i]);
108 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
109 struct btrfs_key key;
110 struct btrfs_path *path;
114 mutex_lock(&root->fs_info->fs_mutex);
115 trans = btrfs_start_transaction(root, 1);
116 btrfs_set_trans_block_group(trans, inode);
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);
139 kaddr = kmap_atomic(bh->b_page, KM_USER0);
140 btrfs_memcpy(root, path->nodes[0]->b_data,
141 ptr, kaddr + bh_offset(bh),
142 offset + write_bytes);
143 kunmap_atomic(kaddr, KM_USER0);
145 mark_buffer_dirty(path->nodes[0]);
146 btrfs_free_path(path);
147 ret = btrfs_end_transaction(trans, root);
149 mutex_unlock(&root->fs_info->fs_mutex);
152 ret = btrfs_commit_write(file, pages[i], offset,
153 offset + this_write);
159 WARN_ON(this_write > write_bytes);
160 write_bytes -= this_write;
167 * this is very complex, but the basic idea is to drop all extents
168 * in the range start - end. hint_block is filled in with a block number
169 * that would be a good hint to the block allocator for this file.
171 * If an extent intersects the range but is not entirely inside the range
172 * it is either truncated or split. Anything entirely inside the range
173 * is deleted from the tree.
175 int btrfs_drop_extents(struct btrfs_trans_handle *trans,
176 struct btrfs_root *root, struct inode *inode,
177 u64 start, u64 end, u64 *hint_block)
180 struct btrfs_key key;
181 struct btrfs_leaf *leaf;
183 struct btrfs_file_extent_item *extent;
186 struct btrfs_file_extent_item old;
187 struct btrfs_path *path;
188 u64 search_start = start;
194 path = btrfs_alloc_path();
198 btrfs_release_path(root, path);
199 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
204 if (path->slots[0] == 0) {
216 leaf = btrfs_buffer_leaf(path->nodes[0]);
217 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) {
223 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
226 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
227 extent = btrfs_item_ptr(leaf, slot,
228 struct btrfs_file_extent_item);
229 found_type = btrfs_file_extent_type(extent);
230 if (found_type == BTRFS_FILE_EXTENT_REG) {
231 extent_end = key.offset +
232 (btrfs_file_extent_num_blocks(extent) <<
235 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
237 extent_end = key.offset +
238 btrfs_file_extent_inline_len(leaf->items +
242 extent_end = search_start;
245 /* we found nothing we can drop */
246 if ((!found_extent && !found_inline) ||
247 search_start >= extent_end) {
250 nritems = btrfs_header_nritems(
251 btrfs_buffer_header(path->nodes[0]));
252 if (slot >= nritems - 1) {
253 nextret = btrfs_next_leaf(root, path);
262 /* FIXME, there's only one inline extent allowed right now */
264 u64 mask = root->blocksize - 1;
265 search_start = (extent_end + mask) & ~mask;
267 search_start = extent_end;
269 if (end < extent_end && end >= key.offset) {
272 btrfs_file_extent_disk_blocknr(extent);
273 u64 disk_num_blocks =
274 btrfs_file_extent_disk_num_blocks(extent);
275 memcpy(&old, extent, sizeof(old));
276 if (disk_blocknr != 0) {
277 ret = btrfs_inc_extent_ref(trans, root,
278 disk_blocknr, disk_num_blocks);
282 WARN_ON(found_inline);
285 /* truncate existing extent */
286 if (start > key.offset) {
290 WARN_ON(start & (root->blocksize - 1));
292 new_num = (start - key.offset) >>
294 old_num = btrfs_file_extent_num_blocks(extent);
296 btrfs_file_extent_disk_blocknr(extent);
297 if (btrfs_file_extent_disk_blocknr(extent)) {
299 (old_num - new_num) << 3;
301 btrfs_set_file_extent_num_blocks(extent,
303 mark_buffer_dirty(path->nodes[0]);
308 /* delete the entire extent */
310 u64 disk_blocknr = 0;
311 u64 disk_num_blocks = 0;
312 u64 extent_num_blocks = 0;
315 btrfs_file_extent_disk_blocknr(extent);
317 btrfs_file_extent_disk_num_blocks(extent);
319 btrfs_file_extent_num_blocks(extent);
321 btrfs_file_extent_disk_blocknr(extent);
323 ret = btrfs_del_item(trans, root, path);
325 btrfs_release_path(root, path);
327 if (found_extent && disk_blocknr != 0) {
328 inode->i_blocks -= extent_num_blocks << 3;
329 ret = btrfs_free_extent(trans, root,
335 if (!bookend && search_start >= end) {
342 /* create bookend, splitting the extent in two */
343 if (bookend && found_extent) {
344 struct btrfs_key ins;
345 ins.objectid = inode->i_ino;
348 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
349 btrfs_release_path(root, path);
350 ret = btrfs_insert_empty_item(trans, root, path, &ins,
354 btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
355 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);
358 extent = btrfs_item_ptr(
359 btrfs_buffer_leaf(path->nodes[0]),
361 struct btrfs_file_extent_item);
362 btrfs_set_file_extent_disk_blocknr(extent,
363 btrfs_file_extent_disk_blocknr(&old));
364 btrfs_set_file_extent_disk_num_blocks(extent,
365 btrfs_file_extent_disk_num_blocks(&old));
367 btrfs_set_file_extent_offset(extent,
368 btrfs_file_extent_offset(&old) +
369 ((end - key.offset) >> inode->i_blkbits));
370 WARN_ON(btrfs_file_extent_num_blocks(&old) <
371 (extent_end - end) >> inode->i_blkbits);
372 btrfs_set_file_extent_num_blocks(extent,
373 (extent_end - end) >> inode->i_blkbits);
375 btrfs_set_file_extent_type(extent,
376 BTRFS_FILE_EXTENT_REG);
377 btrfs_set_file_extent_generation(extent,
378 btrfs_file_extent_generation(&old));
379 btrfs_mark_buffer_dirty(path->nodes[0]);
380 if (btrfs_file_extent_disk_blocknr(&old) != 0) {
382 btrfs_file_extent_num_blocks(extent) << 3;
389 btrfs_free_path(path);
394 * this gets pages into the page cache and locks them down
396 static int prepare_pages(struct btrfs_root *root,
401 unsigned long first_index,
402 unsigned long last_index,
406 unsigned long index = pos >> PAGE_CACHE_SHIFT;
407 struct inode *inode = file->f_path.dentry->d_inode;
411 struct buffer_head *bh;
412 struct buffer_head *head;
413 loff_t isize = i_size_read(inode);
414 struct btrfs_trans_handle *trans;
417 u64 alloc_extent_start;
419 struct btrfs_key ins;
421 start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
422 num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
425 memset(pages, 0, num_pages * sizeof(struct page *));
427 for (i = 0; i < num_pages; i++) {
428 pages[i] = grab_cache_page(inode->i_mapping, index + i);
435 mutex_lock(&root->fs_info->fs_mutex);
436 trans = btrfs_start_transaction(root, 1);
439 mutex_unlock(&root->fs_info->fs_mutex);
442 btrfs_set_trans_block_group(trans, inode);
443 /* FIXME blocksize != 4096 */
444 inode->i_blocks += num_blocks << 3;
447 /* FIXME...EIEIO, ENOSPC and more */
449 /* step one, delete the existing extents in this range */
450 /* FIXME blocksize != pagesize */
451 if (start_pos < inode->i_size) {
452 err = btrfs_drop_extents(trans, root, inode,
453 start_pos, (pos + write_bytes + root->blocksize -1) &
454 ~((u64)root->blocksize - 1), &hint_block);
458 /* insert any holes we need to create */
459 if (inode->i_size < start_pos) {
460 u64 last_pos_in_file;
462 u64 mask = root->blocksize - 1;
463 last_pos_in_file = (isize + mask) & ~mask;
464 hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
465 hole_size >>= inode->i_blkbits;
466 if (last_pos_in_file < start_pos) {
467 err = btrfs_insert_file_extent(trans, root,
476 * either allocate an extent for the new bytes or setup the key
477 * to show we are doing inline data in the extent
479 if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
480 pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
481 err = btrfs_alloc_extent(trans, root, inode->i_ino,
482 num_blocks, hint_block, (u64)-1,
485 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
486 start_pos, ins.objectid, ins.offset,
494 alloc_extent_start = ins.objectid;
495 err = btrfs_end_transaction(trans, root);
496 mutex_unlock(&root->fs_info->fs_mutex);
498 for (i = 0; i < num_pages; i++) {
499 cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
500 wait_on_page_writeback(pages[i]);
501 offset = pos & (PAGE_CACHE_SIZE -1);
502 this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
503 if (!page_has_buffers(pages[i])) {
504 create_empty_buffers(pages[i],
505 root->fs_info->sb->s_blocksize,
508 head = page_buffers(pages[i]);
511 err = btrfs_map_bh_to_logical(root, bh,
515 goto failed_truncate;
516 bh = bh->b_this_page;
517 if (alloc_extent_start)
518 alloc_extent_start++;
519 } while (bh != head);
521 WARN_ON(this_write > write_bytes);
522 write_bytes -= this_write;
527 btrfs_drop_pages(pages, num_pages);
531 btrfs_drop_pages(pages, num_pages);
533 vmtruncate(inode, isize);
537 mutex_unlock(&root->fs_info->fs_mutex);
542 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
543 size_t count, loff_t *ppos)
546 size_t num_written = 0;
549 struct inode *inode = file->f_path.dentry->d_inode;
550 struct btrfs_root *root = BTRFS_I(inode)->root;
551 struct page **pages = NULL;
553 struct page *pinned[2];
554 unsigned long first_index;
555 unsigned long last_index;
557 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
558 PAGE_CACHE_SIZE / (sizeof(struct page *)));
561 if (file->f_flags & O_DIRECT)
564 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
565 current->backing_dev_info = inode->i_mapping->backing_dev_info;
566 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
571 err = remove_suid(file->f_path.dentry);
574 file_update_time(file);
576 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
578 mutex_lock(&inode->i_mutex);
579 first_index = pos >> PAGE_CACHE_SHIFT;
580 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
583 * there are lots of better ways to do this, but this code
584 * makes sure the first and last page in the file range are
585 * up to date and ready for cow
587 if ((pos & (PAGE_CACHE_SIZE - 1))) {
588 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
589 if (!PageUptodate(pinned[0])) {
590 ret = btrfs_readpage(NULL, pinned[0]);
592 wait_on_page_locked(pinned[0]);
594 unlock_page(pinned[0]);
597 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
598 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
599 if (!PageUptodate(pinned[1])) {
600 ret = btrfs_readpage(NULL, pinned[1]);
602 wait_on_page_locked(pinned[1]);
604 unlock_page(pinned[1]);
609 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
610 size_t write_bytes = min(count, nrptrs *
611 (size_t)PAGE_CACHE_SIZE -
613 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
616 WARN_ON(num_pages > nrptrs);
617 memset(pages, 0, sizeof(pages));
618 ret = prepare_pages(root, file, pages, num_pages,
619 pos, first_index, last_index,
623 ret = btrfs_copy_from_user(pos, num_pages,
624 write_bytes, pages, buf);
627 ret = dirty_and_release_pages(NULL, root, file, pages,
628 num_pages, pos, write_bytes);
630 btrfs_drop_pages(pages, num_pages);
633 count -= write_bytes;
635 num_written += write_bytes;
637 balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
638 btrfs_btree_balance_dirty(root);
641 mutex_unlock(&inode->i_mutex);
645 page_cache_release(pinned[0]);
647 page_cache_release(pinned[1]);
649 current->backing_dev_info = NULL;
650 mark_inode_dirty(inode);
651 return num_written ? num_written : err;
654 static int btrfs_sync_file(struct file *file,
655 struct dentry *dentry, int datasync)
657 struct inode *inode = dentry->d_inode;
658 struct btrfs_root *root = BTRFS_I(inode)->root;
660 struct btrfs_trans_handle *trans;
663 * FIXME, use inode generation number to check if we can skip the
666 mutex_lock(&root->fs_info->fs_mutex);
667 trans = btrfs_start_transaction(root, 1);
672 ret = btrfs_commit_transaction(trans, root);
673 mutex_unlock(&root->fs_info->fs_mutex);
675 return ret > 0 ? EIO : ret;
678 static struct vm_operations_struct btrfs_file_vm_ops = {
679 .nopage = filemap_nopage,
680 .populate = filemap_populate,
681 .page_mkwrite = btrfs_page_mkwrite,
684 static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
686 vma->vm_ops = &btrfs_file_vm_ops;
691 struct file_operations btrfs_file_operations = {
692 .llseek = generic_file_llseek,
693 .read = do_sync_read,
694 .aio_read = generic_file_aio_read,
695 .write = btrfs_file_write,
696 .mmap = btrfs_file_mmap,
697 .open = generic_file_open,
698 .ioctl = btrfs_ioctl,
699 .fsync = btrfs_sync_file,
701 .compat_ioctl = btrfs_compat_ioctl,