blob: 4b833972273a75218eb775cf8caf64dc1be80ed7 [file] [log] [blame]
Chris Mason6cbd5572007-06-12 09:07:21 -04001/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
Chris Mason39279cc2007-06-12 06:35:45 -040019#include <linux/fs.h>
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>
Chris Mason39279cc2007-06-12 06:35:45 -040025#include <linux/backing-dev.h>
26#include <linux/mpage.h>
27#include <linux/swap.h>
28#include <linux/writeback.h>
29#include <linux/statfs.h>
30#include <linux/compat.h>
31#include "ctree.h"
32#include "disk-io.h"
33#include "transaction.h"
34#include "btrfs_inode.h"
35#include "ioctl.h"
36#include "print-tree.h"
Chris Masone02119d2008-09-05 16:13:11 -040037#include "tree-log.h"
38#include "locking.h"
Jeff Mahoney12fa8ec2008-05-02 15:03:58 -040039#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040040
41
Chris Masond352ac62008-09-29 15:18:18 -040042/* simple helper to fault in pages and copy. This should go away
43 * and be replaced with calls into generic code.
44 */
Chris Masond3977122009-01-05 21:25:51 -050045static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
Chris Masona1b32a52008-09-05 16:09:51 -040046 int write_bytes,
47 struct page **prepared_pages,
Chris Masond3977122009-01-05 21:25:51 -050048 const char __user *buf)
Chris Mason39279cc2007-06-12 06:35:45 -040049{
50 long page_fault = 0;
51 int i;
52 int offset = pos & (PAGE_CACHE_SIZE - 1);
53
54 for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
55 size_t count = min_t(size_t,
56 PAGE_CACHE_SIZE - offset, write_bytes);
57 struct page *page = prepared_pages[i];
58 fault_in_pages_readable(buf, count);
59
60 /* Copy data from userspace to the current page */
61 kmap(page);
62 page_fault = __copy_from_user(page_address(page) + offset,
63 buf, count);
64 /* Flush processor's dcache for this page */
65 flush_dcache_page(page);
66 kunmap(page);
67 buf += count;
68 write_bytes -= count;
69
70 if (page_fault)
71 break;
72 }
73 return page_fault ? -EFAULT : 0;
74}
75
Chris Masond352ac62008-09-29 15:18:18 -040076/*
77 * unlocks pages after btrfs_file_write is done with them
78 */
Chris Masond3977122009-01-05 21:25:51 -050079static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
Chris Mason39279cc2007-06-12 06:35:45 -040080{
81 size_t i;
82 for (i = 0; i < num_pages; i++) {
83 if (!pages[i])
84 break;
Chris Masond352ac62008-09-29 15:18:18 -040085 /* page checked is some magic around finding pages that
86 * have been modified without going through btrfs_set_page_dirty
87 * clear it here
88 */
Chris Mason4a096752008-07-21 10:29:44 -040089 ClearPageChecked(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -040090 unlock_page(pages[i]);
91 mark_page_accessed(pages[i]);
92 page_cache_release(pages[i]);
93 }
94}
95
Chris Masond352ac62008-09-29 15:18:18 -040096/*
97 * after copy_from_user, pages need to be dirtied and we need to make
98 * sure holes are created between the current EOF and the start of
99 * any next extents (if required).
100 *
101 * this also makes the decision about creating an inline extent vs
102 * doing real data extents, marking pages dirty and delalloc as required.
103 */
Chris Masond3977122009-01-05 21:25:51 -0500104static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
Chris Mason39279cc2007-06-12 06:35:45 -0400105 struct btrfs_root *root,
106 struct file *file,
107 struct page **pages,
108 size_t num_pages,
109 loff_t pos,
110 size_t write_bytes)
111{
Chris Mason39279cc2007-06-12 06:35:45 -0400112 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400113 int i;
Chris Mason6da6aba2007-12-18 16:15:09 -0500114 struct inode *inode = fdentry(file)->d_inode;
Chris Masond1310b22008-01-24 16:13:08 -0500115 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
Chris Masondb945352007-10-15 16:15:53 -0400116 u64 hint_byte;
117 u64 num_bytes;
Chris Masona52d9a82007-08-27 16:49:44 -0400118 u64 start_pos;
119 u64 end_of_last_block;
120 u64 end_pos = pos + write_bytes;
121 loff_t isize = i_size_read(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400122
Chris Mason5f39d392007-10-15 16:14:19 -0400123 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -0400124 num_bytes = (write_bytes + pos - start_pos +
125 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
Chris Mason39279cc2007-06-12 06:35:45 -0400126
Chris Masondb945352007-10-15 16:15:53 -0400127 end_of_last_block = start_pos + num_bytes - 1;
128
Chris Masond1310b22008-01-24 16:13:08 -0500129 lock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
Chris Mason37d1aee2008-07-31 10:48:37 -0400130 trans = btrfs_join_transaction(root, 1);
Chris Masona52d9a82007-08-27 16:49:44 -0400131 if (!trans) {
132 err = -ENOMEM;
133 goto out_unlock;
134 }
135 btrfs_set_trans_block_group(trans, inode);
Chris Masondb945352007-10-15 16:15:53 -0400136 hint_byte = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400137
Chris Masond1310b22008-01-24 16:13:08 -0500138 set_extent_uptodate(io_tree, start_pos, end_of_last_block, GFP_NOFS);
Chris Masona52d9a82007-08-27 16:49:44 -0400139
Chris Masonc8b97812008-10-29 14:49:59 -0400140 /* check for reserved extents on each page, we don't want
141 * to reset the delalloc bit on things that already have
142 * extents reserved.
Chris Masona52d9a82007-08-27 16:49:44 -0400143 */
Chris Masonc8b97812008-10-29 14:49:59 -0400144 btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
145 for (i = 0; i < num_pages; i++) {
146 struct page *p = pages[i];
147 SetPageUptodate(p);
148 ClearPageChecked(p);
149 set_page_dirty(p);
Chris Masona52d9a82007-08-27 16:49:44 -0400150 }
151 if (end_pos > isize) {
152 i_size_write(inode, end_pos);
Chris Masonf597bb12009-06-27 21:06:22 -0400153 /* we've only changed i_size in ram, and we haven't updated
154 * the disk i_size. There is no need to log the inode
155 * at this time.
156 */
Chris Mason39279cc2007-06-12 06:35:45 -0400157 }
Chris Mason017e5362008-07-28 15:32:51 -0400158 err = btrfs_end_transaction(trans, root);
Chris Masona52d9a82007-08-27 16:49:44 -0400159out_unlock:
Chris Masond1310b22008-01-24 16:13:08 -0500160 unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
Chris Mason39279cc2007-06-12 06:35:45 -0400161 return err;
162}
163
Chris Masond352ac62008-09-29 15:18:18 -0400164/*
165 * this drops all the extents in the cache that intersect the range
166 * [start, end]. Existing extents are split as required.
167 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400168int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
169 int skip_pinned)
Chris Masona52d9a82007-08-27 16:49:44 -0400170{
171 struct extent_map *em;
Chris Mason3b951512008-04-17 11:29:12 -0400172 struct extent_map *split = NULL;
173 struct extent_map *split2 = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -0400174 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan39b56372008-02-15 10:40:50 -0500175 u64 len = end - start + 1;
Chris Mason3b951512008-04-17 11:29:12 -0400176 int ret;
177 int testend = 1;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400178 unsigned long flags;
Chris Masonc8b97812008-10-29 14:49:59 -0400179 int compressed = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400180
Chris Masone6dcd2d2008-07-17 12:53:50 -0400181 WARN_ON(end < start);
Chris Mason3b951512008-04-17 11:29:12 -0400182 if (end == (u64)-1) {
Yan39b56372008-02-15 10:40:50 -0500183 len = (u64)-1;
Chris Mason3b951512008-04-17 11:29:12 -0400184 testend = 0;
185 }
Chris Masond3977122009-01-05 21:25:51 -0500186 while (1) {
Chris Mason3b951512008-04-17 11:29:12 -0400187 if (!split)
188 split = alloc_extent_map(GFP_NOFS);
189 if (!split2)
190 split2 = alloc_extent_map(GFP_NOFS);
191
Chris Masond1310b22008-01-24 16:13:08 -0500192 spin_lock(&em_tree->lock);
Yan39b56372008-02-15 10:40:50 -0500193 em = lookup_extent_mapping(em_tree, start, len);
Chris Masond1310b22008-01-24 16:13:08 -0500194 if (!em) {
195 spin_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -0400196 break;
Chris Masond1310b22008-01-24 16:13:08 -0500197 }
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400198 flags = em->flags;
199 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
200 spin_unlock(&em_tree->lock);
201 if (em->start <= start &&
202 (!testend || em->start + em->len >= start + len)) {
203 free_extent_map(em);
204 break;
205 }
206 if (start < em->start) {
207 len = em->start - start;
208 } else {
209 len = start + len - (em->start + em->len);
210 start = em->start + em->len;
211 }
212 free_extent_map(em);
213 continue;
214 }
Chris Masonc8b97812008-10-29 14:49:59 -0400215 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Chris Mason3ce7e672008-07-31 15:42:54 -0400216 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -0400217 remove_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -0400218
219 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
220 em->start < start) {
221 split->start = em->start;
222 split->len = start - em->start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500223 split->orig_start = em->orig_start;
Chris Mason3b951512008-04-17 11:29:12 -0400224 split->block_start = em->block_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400225
226 if (compressed)
227 split->block_len = em->block_len;
228 else
229 split->block_len = split->len;
230
Chris Mason3b951512008-04-17 11:29:12 -0400231 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400232 split->flags = flags;
Chris Mason3b951512008-04-17 11:29:12 -0400233 ret = add_extent_mapping(em_tree, split);
234 BUG_ON(ret);
235 free_extent_map(split);
236 split = split2;
237 split2 = NULL;
238 }
239 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
240 testend && em->start + em->len > start + len) {
241 u64 diff = start + len - em->start;
242
243 split->start = start + len;
244 split->len = em->start + em->len - (start + len);
245 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400246 split->flags = flags;
Chris Mason3b951512008-04-17 11:29:12 -0400247
Chris Masonc8b97812008-10-29 14:49:59 -0400248 if (compressed) {
249 split->block_len = em->block_len;
250 split->block_start = em->block_start;
Chris Mason445a6942008-11-10 11:53:33 -0500251 split->orig_start = em->orig_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400252 } else {
253 split->block_len = split->len;
254 split->block_start = em->block_start + diff;
Chris Mason445a6942008-11-10 11:53:33 -0500255 split->orig_start = split->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400256 }
Chris Mason3b951512008-04-17 11:29:12 -0400257
258 ret = add_extent_mapping(em_tree, split);
259 BUG_ON(ret);
260 free_extent_map(split);
261 split = NULL;
262 }
Chris Masond1310b22008-01-24 16:13:08 -0500263 spin_unlock(&em_tree->lock);
264
Chris Masona52d9a82007-08-27 16:49:44 -0400265 /* once for us */
266 free_extent_map(em);
267 /* once for the tree*/
268 free_extent_map(em);
269 }
Chris Mason3b951512008-04-17 11:29:12 -0400270 if (split)
271 free_extent_map(split);
272 if (split2)
273 free_extent_map(split2);
Chris Masona52d9a82007-08-27 16:49:44 -0400274 return 0;
275}
276
Chris Mason39279cc2007-06-12 06:35:45 -0400277/*
278 * this is very complex, but the basic idea is to drop all extents
279 * in the range start - end. hint_block is filled in with a block number
280 * that would be a good hint to the block allocator for this file.
281 *
282 * If an extent intersects the range but is not entirely inside the range
283 * it is either truncated or split. Anything entirely inside the range
284 * is deleted from the tree.
Chris Masond352ac62008-09-29 15:18:18 -0400285 *
286 * inline_limit is used to tell this code which offsets in the file to keep
287 * if they contain inline extents.
Chris Mason39279cc2007-06-12 06:35:45 -0400288 */
Chris Masond3977122009-01-05 21:25:51 -0500289noinline int btrfs_drop_extents(struct btrfs_trans_handle *trans,
Chris Mason39279cc2007-06-12 06:35:45 -0400290 struct btrfs_root *root, struct inode *inode,
Chris Masone980b502009-04-24 14:39:24 -0400291 u64 start, u64 end, u64 locked_end,
292 u64 inline_limit, u64 *hint_byte)
Chris Mason39279cc2007-06-12 06:35:45 -0400293{
Chris Mason39279cc2007-06-12 06:35:45 -0400294 u64 extent_end = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400295 u64 search_start = start;
Chris Masonc8b97812008-10-29 14:49:59 -0400296 u64 ram_bytes = 0;
Chris Mason771ed682008-11-06 22:02:51 -0500297 u64 disk_bytenr = 0;
Chris Masone980b502009-04-24 14:39:24 -0400298 u64 orig_locked_end = locked_end;
Chris Mason70b99e62008-10-31 12:46:39 -0400299 u8 compression;
300 u8 encryption;
Chris Masonc8b97812008-10-29 14:49:59 -0400301 u16 other_encoding = 0;
Chris Mason00f5c792007-11-30 10:09:33 -0500302 struct extent_buffer *leaf;
303 struct btrfs_file_extent_item *extent;
304 struct btrfs_path *path;
305 struct btrfs_key key;
306 struct btrfs_file_extent_item old;
307 int keep;
308 int slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400309 int bookend;
Yan Zhengd899e052008-10-30 14:25:28 -0400310 int found_type = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400311 int found_extent;
312 int found_inline;
Chris Masonccd467d2007-06-28 15:57:36 -0400313 int recow;
Chris Mason00f5c792007-11-30 10:09:33 -0500314 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -0400315
Chris Masonc8b97812008-10-29 14:49:59 -0400316 inline_limit = 0;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400317 btrfs_drop_extent_cache(inode, start, end - 1, 0);
Chris Masona52d9a82007-08-27 16:49:44 -0400318
Chris Mason39279cc2007-06-12 06:35:45 -0400319 path = btrfs_alloc_path();
320 if (!path)
321 return -ENOMEM;
Chris Masond3977122009-01-05 21:25:51 -0500322 while (1) {
Chris Masonccd467d2007-06-28 15:57:36 -0400323 recow = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400324 btrfs_release_path(root, path);
325 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
326 search_start, -1);
327 if (ret < 0)
328 goto out;
329 if (ret > 0) {
330 if (path->slots[0] == 0) {
331 ret = 0;
332 goto out;
333 }
334 path->slots[0]--;
335 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400336next_slot:
Chris Mason39279cc2007-06-12 06:35:45 -0400337 keep = 0;
338 bookend = 0;
339 found_extent = 0;
340 found_inline = 0;
Chris Mason70b99e62008-10-31 12:46:39 -0400341 compression = 0;
342 encryption = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400343 extent = NULL;
Chris Mason5f39d392007-10-15 16:14:19 -0400344 leaf = path->nodes[0];
Chris Mason39279cc2007-06-12 06:35:45 -0400345 slot = path->slots[0];
Chris Mason8c2383c2007-06-18 09:57:58 -0400346 ret = 0;
Chris Mason5f39d392007-10-15 16:14:19 -0400347 btrfs_item_key_to_cpu(leaf, &key, slot);
Yan72610092008-02-05 15:40:36 -0500348 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY &&
349 key.offset >= end) {
Chris Mason39279cc2007-06-12 06:35:45 -0400350 goto out;
351 }
Yan72610092008-02-05 15:40:36 -0500352 if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY ||
353 key.objectid != inode->i_ino) {
Chris Mason39279cc2007-06-12 06:35:45 -0400354 goto out;
355 }
Chris Masonccd467d2007-06-28 15:57:36 -0400356 if (recow) {
Yan Zheng8247b412008-11-11 09:33:29 -0500357 search_start = max(key.offset, start);
Chris Masonccd467d2007-06-28 15:57:36 -0400358 continue;
359 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400360 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
361 extent = btrfs_item_ptr(leaf, slot,
362 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -0400363 found_type = btrfs_file_extent_type(leaf, extent);
Chris Masonc8b97812008-10-29 14:49:59 -0400364 compression = btrfs_file_extent_compression(leaf,
365 extent);
366 encryption = btrfs_file_extent_encryption(leaf,
367 extent);
368 other_encoding = btrfs_file_extent_other_encoding(leaf,
369 extent);
Yan Zhengd899e052008-10-30 14:25:28 -0400370 if (found_type == BTRFS_FILE_EXTENT_REG ||
371 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Chris Mason257d0ce2007-11-07 21:08:16 -0500372 extent_end =
373 btrfs_file_extent_disk_bytenr(leaf,
374 extent);
375 if (extent_end)
376 *hint_byte = extent_end;
377
Chris Mason8c2383c2007-06-18 09:57:58 -0400378 extent_end = key.offset +
Chris Masondb945352007-10-15 16:15:53 -0400379 btrfs_file_extent_num_bytes(leaf, extent);
Chris Masonc8b97812008-10-29 14:49:59 -0400380 ram_bytes = btrfs_file_extent_ram_bytes(leaf,
381 extent);
Chris Mason8c2383c2007-06-18 09:57:58 -0400382 found_extent = 1;
383 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
384 found_inline = 1;
385 extent_end = key.offset +
Chris Masonc8b97812008-10-29 14:49:59 -0400386 btrfs_file_extent_inline_len(leaf, extent);
Chris Mason8c2383c2007-06-18 09:57:58 -0400387 }
388 } else {
389 extent_end = search_start;
Chris Mason39279cc2007-06-12 06:35:45 -0400390 }
391
392 /* we found nothing we can drop */
Chris Mason8c2383c2007-06-18 09:57:58 -0400393 if ((!found_extent && !found_inline) ||
394 search_start >= extent_end) {
395 int nextret;
396 u32 nritems;
Chris Mason5f39d392007-10-15 16:14:19 -0400397 nritems = btrfs_header_nritems(leaf);
Chris Mason8c2383c2007-06-18 09:57:58 -0400398 if (slot >= nritems - 1) {
399 nextret = btrfs_next_leaf(root, path);
400 if (nextret)
401 goto out;
Chris Masonccd467d2007-06-28 15:57:36 -0400402 recow = 1;
Chris Mason8c2383c2007-06-18 09:57:58 -0400403 } else {
404 path->slots[0]++;
405 }
406 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400407 }
408
Chris Masonc8b97812008-10-29 14:49:59 -0400409 if (end <= extent_end && start >= key.offset && found_inline)
Chris Mason179e29e2007-11-01 11:28:41 -0400410 *hint_byte = EXTENT_MAP_INLINE;
Zheng Yan31840ae2008-09-23 13:14:14 -0400411
412 if (found_extent) {
413 read_extent_buffer(leaf, &old, (unsigned long)extent,
414 sizeof(old));
Zheng Yan31840ae2008-09-23 13:14:14 -0400415 }
416
Chris Mason39279cc2007-06-12 06:35:45 -0400417 if (end < extent_end && end >= key.offset) {
Chris Mason179e29e2007-11-01 11:28:41 -0400418 bookend = 1;
Yan0181e582008-01-30 14:39:54 -0500419 if (found_inline && start <= key.offset)
Chris Mason179e29e2007-11-01 11:28:41 -0400420 keep = 1;
Chris Mason39279cc2007-06-12 06:35:45 -0400421 }
Yan Zheng66435582008-10-30 14:19:50 -0400422
Chris Mason771ed682008-11-06 22:02:51 -0500423 if (bookend && found_extent) {
424 if (locked_end < extent_end) {
425 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
426 locked_end, extent_end - 1,
427 GFP_NOFS);
428 if (!ret) {
429 btrfs_release_path(root, path);
430 lock_extent(&BTRFS_I(inode)->io_tree,
431 locked_end, extent_end - 1,
432 GFP_NOFS);
433 locked_end = extent_end;
434 continue;
435 }
Yan Zheng66435582008-10-30 14:19:50 -0400436 locked_end = extent_end;
Yan Zheng66435582008-10-30 14:19:50 -0400437 }
Chris Mason771ed682008-11-06 22:02:51 -0500438 disk_bytenr = le64_to_cpu(old.disk_bytenr);
439 if (disk_bytenr != 0) {
440 ret = btrfs_inc_extent_ref(trans, root,
441 disk_bytenr,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400442 le64_to_cpu(old.disk_num_bytes), 0,
443 root->root_key.objectid,
444 key.objectid, key.offset -
445 le64_to_cpu(old.offset));
Chris Mason771ed682008-11-06 22:02:51 -0500446 BUG_ON(ret);
447 }
Yan Zheng66435582008-10-30 14:19:50 -0400448 }
449
450 if (found_inline) {
451 u64 mask = root->sectorsize - 1;
452 search_start = (extent_end + mask) & ~mask;
453 } else
454 search_start = extent_end;
455
Chris Mason39279cc2007-06-12 06:35:45 -0400456 /* truncate existing extent */
457 if (start > key.offset) {
458 u64 new_num;
459 u64 old_num;
460 keep = 1;
Chris Mason5f39d392007-10-15 16:14:19 -0400461 WARN_ON(start & (root->sectorsize - 1));
Chris Mason39279cc2007-06-12 06:35:45 -0400462 if (found_extent) {
Chris Masondb945352007-10-15 16:15:53 -0400463 new_num = start - key.offset;
464 old_num = btrfs_file_extent_num_bytes(leaf,
465 extent);
466 *hint_byte =
467 btrfs_file_extent_disk_bytenr(leaf,
468 extent);
469 if (btrfs_file_extent_disk_bytenr(leaf,
470 extent)) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400471 inode_sub_bytes(inode, old_num -
472 new_num);
Chris Mason39279cc2007-06-12 06:35:45 -0400473 }
Chris Mason771ed682008-11-06 22:02:51 -0500474 btrfs_set_file_extent_num_bytes(leaf,
475 extent, new_num);
Chris Mason5f39d392007-10-15 16:14:19 -0400476 btrfs_mark_buffer_dirty(leaf);
Chris Mason00f5c792007-11-30 10:09:33 -0500477 } else if (key.offset < inline_limit &&
478 (end > extent_end) &&
479 (inline_limit < extent_end)) {
Chris Mason3326d1b2007-10-15 16:18:25 -0400480 u32 new_size;
481 new_size = btrfs_file_extent_calc_inline_size(
Chris Mason00f5c792007-11-30 10:09:33 -0500482 inline_limit - key.offset);
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400483 inode_sub_bytes(inode, extent_end -
484 inline_limit);
Chris Mason70b99e62008-10-31 12:46:39 -0400485 btrfs_set_file_extent_ram_bytes(leaf, extent,
486 new_size);
487 if (!compression && !encryption) {
488 btrfs_truncate_item(trans, root, path,
489 new_size, 1);
490 }
Chris Mason39279cc2007-06-12 06:35:45 -0400491 }
492 }
493 /* delete the entire extent */
494 if (!keep) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400495 if (found_inline)
496 inode_sub_bytes(inode, extent_end -
497 key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400498 ret = btrfs_del_item(trans, root, path);
Chris Mason54aa1f42007-06-22 14:16:25 -0400499 /* TODO update progress marker and return */
Chris Mason39279cc2007-06-12 06:35:45 -0400500 BUG_ON(ret);
Chris Mason39279cc2007-06-12 06:35:45 -0400501 extent = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -0400502 btrfs_release_path(root, path);
503 /* the extent will be freed later */
Chris Mason39279cc2007-06-12 06:35:45 -0400504 }
Yan0181e582008-01-30 14:39:54 -0500505 if (bookend && found_inline && start <= key.offset) {
Chris Mason179e29e2007-11-01 11:28:41 -0400506 u32 new_size;
507 new_size = btrfs_file_extent_calc_inline_size(
Yan0181e582008-01-30 14:39:54 -0500508 extent_end - end);
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400509 inode_sub_bytes(inode, end - key.offset);
Chris Mason70b99e62008-10-31 12:46:39 -0400510 btrfs_set_file_extent_ram_bytes(leaf, extent,
511 new_size);
512 if (!compression && !encryption)
513 ret = btrfs_truncate_item(trans, root, path,
514 new_size, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -0400515 BUG_ON(ret);
Chris Mason179e29e2007-11-01 11:28:41 -0400516 }
Chris Mason39279cc2007-06-12 06:35:45 -0400517 /* create bookend, splitting the extent in two */
518 if (bookend && found_extent) {
519 struct btrfs_key ins;
520 ins.objectid = inode->i_ino;
521 ins.offset = end;
Chris Mason39279cc2007-06-12 06:35:45 -0400522 btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
Chris Mason771ed682008-11-06 22:02:51 -0500523
Chris Mason39279cc2007-06-12 06:35:45 -0400524 btrfs_release_path(root, path);
Chris Masonb9473432009-03-13 11:00:37 -0400525 path->leave_spinning = 1;
Chris Mason39279cc2007-06-12 06:35:45 -0400526 ret = btrfs_insert_empty_item(trans, root, path, &ins,
527 sizeof(*extent));
Zheng Yan31840ae2008-09-23 13:14:14 -0400528 BUG_ON(ret);
Chris Mason8c2383c2007-06-18 09:57:58 -0400529
Chris Mason5f39d392007-10-15 16:14:19 -0400530 leaf = path->nodes[0];
Chris Mason5f39d392007-10-15 16:14:19 -0400531 extent = btrfs_item_ptr(leaf, path->slots[0],
532 struct btrfs_file_extent_item);
533 write_extent_buffer(leaf, &old,
534 (unsigned long)extent, sizeof(old));
Chris Mason39279cc2007-06-12 06:35:45 -0400535
Chris Masonc8b97812008-10-29 14:49:59 -0400536 btrfs_set_file_extent_compression(leaf, extent,
537 compression);
538 btrfs_set_file_extent_encryption(leaf, extent,
539 encryption);
540 btrfs_set_file_extent_other_encoding(leaf, extent,
541 other_encoding);
Chris Mason5f39d392007-10-15 16:14:19 -0400542 btrfs_set_file_extent_offset(leaf, extent,
Chris Masondb945352007-10-15 16:15:53 -0400543 le64_to_cpu(old.offset) + end - key.offset);
544 WARN_ON(le64_to_cpu(old.num_bytes) <
545 (extent_end - end));
546 btrfs_set_file_extent_num_bytes(leaf, extent,
547 extent_end - end);
Chris Masonc8b97812008-10-29 14:49:59 -0400548
549 /*
550 * set the ram bytes to the size of the full extent
551 * before splitting. This is a worst case flag,
552 * but its the best we can do because we don't know
553 * how splitting affects compression
554 */
555 btrfs_set_file_extent_ram_bytes(leaf, extent,
556 ram_bytes);
Yan Zhengd899e052008-10-30 14:25:28 -0400557 btrfs_set_file_extent_type(leaf, extent, found_type);
Chris Masondb945352007-10-15 16:15:53 -0400558
Chris Masonb9473432009-03-13 11:00:37 -0400559 btrfs_unlock_up_safe(path, 1);
Chris Mason39279cc2007-06-12 06:35:45 -0400560 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Masonb9473432009-03-13 11:00:37 -0400561 btrfs_set_lock_blocking(path->nodes[0]);
Zheng Yan31840ae2008-09-23 13:14:14 -0400562
Chris Masonb9473432009-03-13 11:00:37 -0400563 path->leave_spinning = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -0400564 btrfs_release_path(root, path);
Chris Masond3977122009-01-05 21:25:51 -0500565 if (disk_bytenr != 0)
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400566 inode_add_bytes(inode, extent_end - end);
Zheng Yan31840ae2008-09-23 13:14:14 -0400567 }
568
569 if (found_extent && !keep) {
Christoph Hellwig6e430f92008-12-02 06:36:09 -0500570 u64 old_disk_bytenr = le64_to_cpu(old.disk_bytenr);
Zheng Yan31840ae2008-09-23 13:14:14 -0400571
Christoph Hellwig6e430f92008-12-02 06:36:09 -0500572 if (old_disk_bytenr != 0) {
Yan Zhenga76a3cd2008-10-09 11:46:29 -0400573 inode_sub_bytes(inode,
574 le64_to_cpu(old.num_bytes));
Zheng Yan31840ae2008-09-23 13:14:14 -0400575 ret = btrfs_free_extent(trans, root,
Christoph Hellwig6e430f92008-12-02 06:36:09 -0500576 old_disk_bytenr,
Zheng Yan31840ae2008-09-23 13:14:14 -0400577 le64_to_cpu(old.disk_num_bytes),
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400578 0, root->root_key.objectid,
579 key.objectid, key.offset -
580 le64_to_cpu(old.offset));
Zheng Yan31840ae2008-09-23 13:14:14 -0400581 BUG_ON(ret);
Christoph Hellwig6e430f92008-12-02 06:36:09 -0500582 *hint_byte = old_disk_bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -0400583 }
584 }
585
586 if (search_start >= end) {
Chris Mason39279cc2007-06-12 06:35:45 -0400587 ret = 0;
588 goto out;
589 }
590 }
591out:
592 btrfs_free_path(path);
Chris Masone980b502009-04-24 14:39:24 -0400593 if (locked_end > orig_locked_end) {
594 unlock_extent(&BTRFS_I(inode)->io_tree, orig_locked_end,
595 locked_end - 1, GFP_NOFS);
Yan Zheng66435582008-10-30 14:19:50 -0400596 }
Chris Mason39279cc2007-06-12 06:35:45 -0400597 return ret;
598}
599
Yan Zhengd899e052008-10-30 14:25:28 -0400600static int extent_mergeable(struct extent_buffer *leaf, int slot,
601 u64 objectid, u64 bytenr, u64 *start, u64 *end)
602{
603 struct btrfs_file_extent_item *fi;
604 struct btrfs_key key;
605 u64 extent_end;
606
607 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
608 return 0;
609
610 btrfs_item_key_to_cpu(leaf, &key, slot);
611 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
612 return 0;
613
614 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
615 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
616 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
617 btrfs_file_extent_compression(leaf, fi) ||
618 btrfs_file_extent_encryption(leaf, fi) ||
619 btrfs_file_extent_other_encoding(leaf, fi))
620 return 0;
621
622 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
623 if ((*start && *start != key.offset) || (*end && *end != extent_end))
624 return 0;
625
626 *start = key.offset;
627 *end = extent_end;
628 return 1;
629}
630
631/*
632 * Mark extent in the range start - end as written.
633 *
634 * This changes extent type from 'pre-allocated' to 'regular'. If only
635 * part of extent is marked as written, the extent will be split into
636 * two or three.
637 */
638int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
639 struct btrfs_root *root,
640 struct inode *inode, u64 start, u64 end)
641{
642 struct extent_buffer *leaf;
643 struct btrfs_path *path;
644 struct btrfs_file_extent_item *fi;
645 struct btrfs_key key;
646 u64 bytenr;
647 u64 num_bytes;
648 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400649 u64 orig_offset;
Yan Zhengd899e052008-10-30 14:25:28 -0400650 u64 other_start;
651 u64 other_end;
652 u64 split = start;
653 u64 locked_end = end;
654 int extent_type;
655 int split_end = 1;
656 int ret;
657
658 btrfs_drop_extent_cache(inode, start, end - 1, 0);
659
660 path = btrfs_alloc_path();
661 BUG_ON(!path);
662again:
663 key.objectid = inode->i_ino;
664 key.type = BTRFS_EXTENT_DATA_KEY;
665 if (split == start)
666 key.offset = split;
667 else
668 key.offset = split - 1;
669
670 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
671 if (ret > 0 && path->slots[0] > 0)
672 path->slots[0]--;
673
674 leaf = path->nodes[0];
675 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
676 BUG_ON(key.objectid != inode->i_ino ||
677 key.type != BTRFS_EXTENT_DATA_KEY);
678 fi = btrfs_item_ptr(leaf, path->slots[0],
679 struct btrfs_file_extent_item);
680 extent_type = btrfs_file_extent_type(leaf, fi);
681 BUG_ON(extent_type != BTRFS_FILE_EXTENT_PREALLOC);
682 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
683 BUG_ON(key.offset > start || extent_end < end);
684
685 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
686 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400687 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
Yan Zhengd899e052008-10-30 14:25:28 -0400688
689 if (key.offset == start)
690 split = end;
691
692 if (key.offset == start && extent_end == end) {
693 int del_nr = 0;
694 int del_slot = 0;
Yan Zhengd899e052008-10-30 14:25:28 -0400695 other_start = end;
696 other_end = 0;
697 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
698 bytenr, &other_start, &other_end)) {
699 extent_end = other_end;
700 del_slot = path->slots[0] + 1;
701 del_nr++;
702 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400703 0, root->root_key.objectid,
704 inode->i_ino, orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400705 BUG_ON(ret);
706 }
707 other_start = 0;
708 other_end = start;
709 if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
710 bytenr, &other_start, &other_end)) {
711 key.offset = other_start;
712 del_slot = path->slots[0];
713 del_nr++;
714 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400715 0, root->root_key.objectid,
716 inode->i_ino, orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400717 BUG_ON(ret);
718 }
719 split_end = 0;
720 if (del_nr == 0) {
721 btrfs_set_file_extent_type(leaf, fi,
722 BTRFS_FILE_EXTENT_REG);
723 goto done;
724 }
725
726 fi = btrfs_item_ptr(leaf, del_slot - 1,
727 struct btrfs_file_extent_item);
728 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
729 btrfs_set_file_extent_num_bytes(leaf, fi,
730 extent_end - key.offset);
731 btrfs_mark_buffer_dirty(leaf);
732
733 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
734 BUG_ON(ret);
Chris Mason546888d2009-04-21 11:53:38 -0400735 goto release;
Yan Zhengd899e052008-10-30 14:25:28 -0400736 } else if (split == start) {
737 if (locked_end < extent_end) {
738 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
739 locked_end, extent_end - 1, GFP_NOFS);
740 if (!ret) {
741 btrfs_release_path(root, path);
742 lock_extent(&BTRFS_I(inode)->io_tree,
743 locked_end, extent_end - 1, GFP_NOFS);
744 locked_end = extent_end;
745 goto again;
746 }
747 locked_end = extent_end;
748 }
749 btrfs_set_file_extent_num_bytes(leaf, fi, split - key.offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400750 } else {
751 BUG_ON(key.offset != start);
Yan Zhengd899e052008-10-30 14:25:28 -0400752 key.offset = split;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400753 btrfs_set_file_extent_offset(leaf, fi, key.offset -
754 orig_offset);
755 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - split);
Yan Zhengd899e052008-10-30 14:25:28 -0400756 btrfs_set_item_key_safe(trans, root, path, &key);
757 extent_end = split;
758 }
759
760 if (extent_end == end) {
761 split_end = 0;
762 extent_type = BTRFS_FILE_EXTENT_REG;
763 }
764 if (extent_end == end && split == start) {
765 other_start = end;
766 other_end = 0;
767 if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
768 bytenr, &other_start, &other_end)) {
769 path->slots[0]++;
770 fi = btrfs_item_ptr(leaf, path->slots[0],
771 struct btrfs_file_extent_item);
772 key.offset = split;
773 btrfs_set_item_key_safe(trans, root, path, &key);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400774 btrfs_set_file_extent_offset(leaf, fi, key.offset -
775 orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400776 btrfs_set_file_extent_num_bytes(leaf, fi,
777 other_end - split);
778 goto done;
779 }
780 }
781 if (extent_end == end && split == end) {
782 other_start = 0;
783 other_end = start;
784 if (extent_mergeable(leaf, path->slots[0] - 1 , inode->i_ino,
785 bytenr, &other_start, &other_end)) {
786 path->slots[0]--;
787 fi = btrfs_item_ptr(leaf, path->slots[0],
788 struct btrfs_file_extent_item);
789 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end -
790 other_start);
791 goto done;
792 }
793 }
794
795 btrfs_mark_buffer_dirty(leaf);
Yan Zhengc36047d2008-11-12 14:19:50 -0500796
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400797 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
798 root->root_key.objectid,
799 inode->i_ino, orig_offset);
Yan Zhengc36047d2008-11-12 14:19:50 -0500800 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -0400801 btrfs_release_path(root, path);
802
803 key.offset = start;
804 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*fi));
805 BUG_ON(ret);
806
807 leaf = path->nodes[0];
808 fi = btrfs_item_ptr(leaf, path->slots[0],
809 struct btrfs_file_extent_item);
810 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
811 btrfs_set_file_extent_type(leaf, fi, extent_type);
812 btrfs_set_file_extent_disk_bytenr(leaf, fi, bytenr);
813 btrfs_set_file_extent_disk_num_bytes(leaf, fi, num_bytes);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400814 btrfs_set_file_extent_offset(leaf, fi, key.offset - orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400815 btrfs_set_file_extent_num_bytes(leaf, fi, extent_end - key.offset);
816 btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
817 btrfs_set_file_extent_compression(leaf, fi, 0);
818 btrfs_set_file_extent_encryption(leaf, fi, 0);
819 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
Yan Zhengd899e052008-10-30 14:25:28 -0400820done:
821 btrfs_mark_buffer_dirty(leaf);
Chris Mason546888d2009-04-21 11:53:38 -0400822
823release:
Yan Zhengd899e052008-10-30 14:25:28 -0400824 btrfs_release_path(root, path);
825 if (split_end && split == start) {
826 split = end;
827 goto again;
828 }
829 if (locked_end > end) {
830 unlock_extent(&BTRFS_I(inode)->io_tree, end, locked_end - 1,
831 GFP_NOFS);
832 }
833 btrfs_free_path(path);
834 return 0;
835}
836
Chris Mason39279cc2007-06-12 06:35:45 -0400837/*
Chris Masond352ac62008-09-29 15:18:18 -0400838 * this gets pages into the page cache and locks them down, it also properly
839 * waits for data=ordered extents to finish before allowing the pages to be
840 * modified.
Chris Mason39279cc2007-06-12 06:35:45 -0400841 */
Chris Masond3977122009-01-05 21:25:51 -0500842static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
Chris Mason98ed5172008-01-03 10:01:48 -0500843 struct page **pages, size_t num_pages,
844 loff_t pos, unsigned long first_index,
845 unsigned long last_index, size_t write_bytes)
Chris Mason39279cc2007-06-12 06:35:45 -0400846{
847 int i;
848 unsigned long index = pos >> PAGE_CACHE_SHIFT;
Chris Mason6da6aba2007-12-18 16:15:09 -0500849 struct inode *inode = fdentry(file)->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -0400850 int err = 0;
Chris Mason8c2383c2007-06-18 09:57:58 -0400851 u64 start_pos;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400852 u64 last_pos;
Chris Mason8c2383c2007-06-18 09:57:58 -0400853
Chris Mason5f39d392007-10-15 16:14:19 -0400854 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400855 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -0400856
Yan Zheng9036c102008-10-30 14:19:41 -0400857 if (start_pos > inode->i_size) {
858 err = btrfs_cont_expand(inode, start_pos);
859 if (err)
860 return err;
861 }
862
Chris Mason39279cc2007-06-12 06:35:45 -0400863 memset(pages, 0, num_pages * sizeof(struct page *));
Chris Masone6dcd2d2008-07-17 12:53:50 -0400864again:
Chris Mason39279cc2007-06-12 06:35:45 -0400865 for (i = 0; i < num_pages; i++) {
866 pages[i] = grab_cache_page(inode->i_mapping, index + i);
867 if (!pages[i]) {
868 err = -ENOMEM;
Chris Masona52d9a82007-08-27 16:49:44 -0400869 BUG_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -0400870 }
Chris Masonccd467d2007-06-28 15:57:36 -0400871 wait_on_page_writeback(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400872 }
Chris Mason07627042008-02-19 11:29:24 -0500873 if (start_pos < inode->i_size) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400874 struct btrfs_ordered_extent *ordered;
Chris Masond99cb302008-02-19 12:55:05 -0500875 lock_extent(&BTRFS_I(inode)->io_tree,
876 start_pos, last_pos - 1, GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500877 ordered = btrfs_lookup_first_ordered_extent(inode,
878 last_pos - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400879 if (ordered &&
880 ordered->file_offset + ordered->len > start_pos &&
881 ordered->file_offset < last_pos) {
882 btrfs_put_ordered_extent(ordered);
883 unlock_extent(&BTRFS_I(inode)->io_tree,
884 start_pos, last_pos - 1, GFP_NOFS);
885 for (i = 0; i < num_pages; i++) {
886 unlock_page(pages[i]);
887 page_cache_release(pages[i]);
888 }
889 btrfs_wait_ordered_range(inode, start_pos,
890 last_pos - start_pos);
891 goto again;
892 }
893 if (ordered)
894 btrfs_put_ordered_extent(ordered);
895
Chris Mason07627042008-02-19 11:29:24 -0500896 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
897 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC,
898 GFP_NOFS);
Chris Masond99cb302008-02-19 12:55:05 -0500899 unlock_extent(&BTRFS_I(inode)->io_tree,
900 start_pos, last_pos - 1, GFP_NOFS);
Chris Mason07627042008-02-19 11:29:24 -0500901 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400902 for (i = 0; i < num_pages; i++) {
Chris Masonf87f0572008-08-01 11:27:23 -0400903 clear_page_dirty_for_io(pages[i]);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400904 set_page_extent_mapped(pages[i]);
905 WARN_ON(!PageLocked(pages[i]));
906 }
Chris Mason39279cc2007-06-12 06:35:45 -0400907 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400908}
909
910static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
911 size_t count, loff_t *ppos)
912{
913 loff_t pos;
Chris Mason2ff3e9b2007-10-29 14:36:41 -0400914 loff_t start_pos;
915 ssize_t num_written = 0;
916 ssize_t err = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400917 int ret = 0;
Chris Mason6da6aba2007-12-18 16:15:09 -0500918 struct inode *inode = fdentry(file)->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -0400919 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason8c2383c2007-06-18 09:57:58 -0400920 struct page **pages = NULL;
921 int nrptrs;
Chris Mason39279cc2007-06-12 06:35:45 -0400922 struct page *pinned[2];
923 unsigned long first_index;
924 unsigned long last_index;
Chris Masoncb843a62008-10-03 12:30:02 -0400925 int will_write;
926
927 will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
928 (file->f_flags & O_DIRECT));
Chris Mason8c2383c2007-06-18 09:57:58 -0400929
930 nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
931 PAGE_CACHE_SIZE / (sizeof(struct page *)));
Chris Mason39279cc2007-06-12 06:35:45 -0400932 pinned[0] = NULL;
933 pinned[1] = NULL;
Chris Mason2ff3e9b2007-10-29 14:36:41 -0400934
Chris Mason39279cc2007-06-12 06:35:45 -0400935 pos = *ppos;
Chris Mason2ff3e9b2007-10-29 14:36:41 -0400936 start_pos = pos;
937
Chris Mason39279cc2007-06-12 06:35:45 -0400938 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
939 current->backing_dev_info = inode->i_mapping->backing_dev_info;
940 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
941 if (err)
Chris Mason1832a6d2007-12-21 16:27:21 -0500942 goto out_nolock;
Chris Mason39279cc2007-06-12 06:35:45 -0400943 if (count == 0)
Chris Mason1832a6d2007-12-21 16:27:21 -0500944 goto out_nolock;
Chris Mason2b1f55b2008-09-24 11:48:04 -0400945
Sven Wegener0ee0fda2008-07-30 16:54:26 -0400946 err = file_remove_suid(file);
Chris Mason39279cc2007-06-12 06:35:45 -0400947 if (err)
Chris Mason1832a6d2007-12-21 16:27:21 -0500948 goto out_nolock;
Chris Mason39279cc2007-06-12 06:35:45 -0400949 file_update_time(file);
950
Chris Mason8c2383c2007-06-18 09:57:58 -0400951 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
Chris Mason39279cc2007-06-12 06:35:45 -0400952
953 mutex_lock(&inode->i_mutex);
Chris Masonc3027eb2008-12-08 16:40:21 -0500954 BTRFS_I(inode)->sequence++;
Chris Mason39279cc2007-06-12 06:35:45 -0400955 first_index = pos >> PAGE_CACHE_SHIFT;
956 last_index = (pos + count) >> PAGE_CACHE_SHIFT;
957
958 /*
959 * there are lots of better ways to do this, but this code
960 * makes sure the first and last page in the file range are
961 * up to date and ready for cow
962 */
963 if ((pos & (PAGE_CACHE_SIZE - 1))) {
964 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
965 if (!PageUptodate(pinned[0])) {
Chris Mason9ebefb182007-06-15 13:50:00 -0400966 ret = btrfs_readpage(NULL, pinned[0]);
Chris Mason39279cc2007-06-12 06:35:45 -0400967 BUG_ON(ret);
968 wait_on_page_locked(pinned[0]);
969 } else {
970 unlock_page(pinned[0]);
971 }
972 }
973 if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
974 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
975 if (!PageUptodate(pinned[1])) {
Chris Mason9ebefb182007-06-15 13:50:00 -0400976 ret = btrfs_readpage(NULL, pinned[1]);
Chris Mason39279cc2007-06-12 06:35:45 -0400977 BUG_ON(ret);
978 wait_on_page_locked(pinned[1]);
979 } else {
980 unlock_page(pinned[1]);
981 }
982 }
983
Chris Masond3977122009-01-05 21:25:51 -0500984 while (count > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -0400985 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
Chris Mason11bd1432007-06-22 14:16:24 -0400986 size_t write_bytes = min(count, nrptrs *
987 (size_t)PAGE_CACHE_SIZE -
Chris Mason8c2383c2007-06-18 09:57:58 -0400988 offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400989 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
990 PAGE_CACHE_SHIFT;
991
Chris Mason8c2383c2007-06-18 09:57:58 -0400992 WARN_ON(num_pages > nrptrs);
yanhai zhu9aead432009-01-05 15:49:11 -0500993 memset(pages, 0, sizeof(struct page *) * nrptrs);
Chris Mason1832a6d2007-12-21 16:27:21 -0500994
Josef Bacik6a632092009-02-20 11:00:09 -0500995 ret = btrfs_check_data_free_space(root, inode, write_bytes);
Chris Mason1832a6d2007-12-21 16:27:21 -0500996 if (ret)
997 goto out;
998
Chris Mason39279cc2007-06-12 06:35:45 -0400999 ret = prepare_pages(root, file, pages, num_pages,
1000 pos, first_index, last_index,
Chris Mason8c2383c2007-06-18 09:57:58 -04001001 write_bytes);
Josef Bacik6a632092009-02-20 11:00:09 -05001002 if (ret) {
1003 btrfs_free_reserved_data_space(root, inode,
1004 write_bytes);
Chris Mason54aa1f42007-06-22 14:16:25 -04001005 goto out;
Josef Bacik6a632092009-02-20 11:00:09 -05001006 }
Chris Mason39279cc2007-06-12 06:35:45 -04001007
Chris Mason39279cc2007-06-12 06:35:45 -04001008 ret = btrfs_copy_from_user(pos, num_pages,
1009 write_bytes, pages, buf);
Chris Mason54aa1f42007-06-22 14:16:25 -04001010 if (ret) {
Josef Bacik6a632092009-02-20 11:00:09 -05001011 btrfs_free_reserved_data_space(root, inode,
1012 write_bytes);
Chris Mason54aa1f42007-06-22 14:16:25 -04001013 btrfs_drop_pages(pages, num_pages);
1014 goto out;
1015 }
Chris Mason39279cc2007-06-12 06:35:45 -04001016
1017 ret = dirty_and_release_pages(NULL, root, file, pages,
1018 num_pages, pos, write_bytes);
Chris Mason39279cc2007-06-12 06:35:45 -04001019 btrfs_drop_pages(pages, num_pages);
Josef Bacik6a632092009-02-20 11:00:09 -05001020 if (ret) {
1021 btrfs_free_reserved_data_space(root, inode,
1022 write_bytes);
Chris Mason54aa1f42007-06-22 14:16:25 -04001023 goto out;
Josef Bacik6a632092009-02-20 11:00:09 -05001024 }
Chris Mason39279cc2007-06-12 06:35:45 -04001025
Chris Masoncb843a62008-10-03 12:30:02 -04001026 if (will_write) {
1027 btrfs_fdatawrite_range(inode->i_mapping, pos,
1028 pos + write_bytes - 1,
Chris Masond313d7a2009-04-20 15:50:09 -04001029 WB_SYNC_ALL);
Chris Masoncb843a62008-10-03 12:30:02 -04001030 } else {
1031 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1032 num_pages);
1033 if (num_pages <
1034 (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1035 btrfs_btree_balance_dirty(root, 1);
1036 btrfs_throttle(root);
1037 }
1038
Chris Mason39279cc2007-06-12 06:35:45 -04001039 buf += write_bytes;
1040 count -= write_bytes;
1041 pos += write_bytes;
1042 num_written += write_bytes;
1043
Chris Mason39279cc2007-06-12 06:35:45 -04001044 cond_resched();
1045 }
Chris Mason39279cc2007-06-12 06:35:45 -04001046out:
Chris Mason1832a6d2007-12-21 16:27:21 -05001047 mutex_unlock(&inode->i_mutex);
Josef Bacik6a632092009-02-20 11:00:09 -05001048 if (ret)
1049 err = ret;
Chris Mason5b92ee72008-01-03 13:46:11 -05001050
Chris Mason1832a6d2007-12-21 16:27:21 -05001051out_nolock:
Chris Mason8c2383c2007-06-18 09:57:58 -04001052 kfree(pages);
Chris Mason39279cc2007-06-12 06:35:45 -04001053 if (pinned[0])
1054 page_cache_release(pinned[0]);
1055 if (pinned[1])
1056 page_cache_release(pinned[1]);
1057 *ppos = pos;
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001058
Chris Mason5a3f23d2009-03-31 13:27:11 -04001059 /*
1060 * we want to make sure fsync finds this change
1061 * but we haven't joined a transaction running right now.
1062 *
1063 * Later on, someone is sure to update the inode and get the
1064 * real transid recorded.
1065 *
1066 * We set last_trans now to the fs_info generation + 1,
1067 * this will either be one more than the running transaction
1068 * or the generation used for the next transaction if there isn't
1069 * one running right now.
1070 */
1071 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1072
Chris Masoncb843a62008-10-03 12:30:02 -04001073 if (num_written > 0 && will_write) {
Chris Masone02119d2008-09-05 16:13:11 -04001074 struct btrfs_trans_handle *trans;
1075
Chris Masoncb843a62008-10-03 12:30:02 -04001076 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
1077 if (err)
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001078 num_written = err;
Chris Masone02119d2008-09-05 16:13:11 -04001079
Chris Masoncb843a62008-10-03 12:30:02 -04001080 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
1081 trans = btrfs_start_transaction(root, 1);
1082 ret = btrfs_log_dentry_safe(trans, root,
1083 file->f_dentry);
1084 if (ret == 0) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001085 ret = btrfs_sync_log(trans, root);
1086 if (ret == 0)
1087 btrfs_end_transaction(trans, root);
1088 else
1089 btrfs_commit_transaction(trans, root);
Chris Masoncb843a62008-10-03 12:30:02 -04001090 } else {
1091 btrfs_commit_transaction(trans, root);
1092 }
Chris Masone02119d2008-09-05 16:13:11 -04001093 }
Chris Masoncb843a62008-10-03 12:30:02 -04001094 if (file->f_flags & O_DIRECT) {
1095 invalidate_mapping_pages(inode->i_mapping,
1096 start_pos >> PAGE_CACHE_SHIFT,
1097 (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
1098 }
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001099 }
Chris Mason39279cc2007-06-12 06:35:45 -04001100 current->backing_dev_info = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001101 return num_written ? num_written : err;
1102}
1103
Chris Masond3977122009-01-05 21:25:51 -05001104int btrfs_release_file(struct inode *inode, struct file *filp)
Mingminge1b81e62008-05-27 10:55:43 -04001105{
Chris Mason5a3f23d2009-03-31 13:27:11 -04001106 /*
1107 * ordered_data_close is set by settattr when we are about to truncate
1108 * a file from a non-zero size to a zero size. This tries to
1109 * flush down new bytes that may have been written if the
1110 * application were using truncate to replace a file in place.
1111 */
1112 if (BTRFS_I(inode)->ordered_data_close) {
1113 BTRFS_I(inode)->ordered_data_close = 0;
1114 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1115 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1116 filemap_flush(inode->i_mapping);
1117 }
Sage Weil6bf13c02008-06-10 10:07:39 -04001118 if (filp->private_data)
1119 btrfs_ioctl_trans_end(filp);
Mingminge1b81e62008-05-27 10:55:43 -04001120 return 0;
1121}
1122
Chris Masond352ac62008-09-29 15:18:18 -04001123/*
1124 * fsync call for both files and directories. This logs the inode into
1125 * the tree log instead of forcing full commits whenever possible.
1126 *
1127 * It needs to call filemap_fdatawait so that all ordered extent updates are
1128 * in the metadata btree are up to date for copying to the log.
1129 *
1130 * It drops the inode mutex before doing the tree log commit. This is an
1131 * important optimization for directories because holding the mutex prevents
1132 * new operations on the dir while we write to disk.
1133 */
Chris Masone02119d2008-09-05 16:13:11 -04001134int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
Chris Mason39279cc2007-06-12 06:35:45 -04001135{
1136 struct inode *inode = dentry->d_inode;
1137 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001138 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001139 struct btrfs_trans_handle *trans;
1140
1141 /*
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001142 * check the transaction that last modified this inode
1143 * and see if its already been committed
Chris Mason39279cc2007-06-12 06:35:45 -04001144 */
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001145 if (!BTRFS_I(inode)->last_trans)
1146 goto out;
Chris Masona2135012008-06-25 16:01:30 -04001147
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001148 mutex_lock(&root->fs_info->trans_mutex);
1149 if (BTRFS_I(inode)->last_trans <=
1150 root->fs_info->last_trans_committed) {
1151 BTRFS_I(inode)->last_trans = 0;
1152 mutex_unlock(&root->fs_info->trans_mutex);
1153 goto out;
1154 }
1155 mutex_unlock(&root->fs_info->trans_mutex);
1156
Yan Zheng7237f182009-01-21 12:54:03 -05001157 root->log_batch++;
Chris Mason580afd72008-12-08 19:15:39 -05001158 filemap_fdatawrite(inode->i_mapping);
1159 btrfs_wait_ordered_range(inode, 0, (u64)-1);
Yan Zheng7237f182009-01-21 12:54:03 -05001160 root->log_batch++;
Chris Masone02119d2008-09-05 16:13:11 -04001161
Hisashi Hifumi524724e2009-06-10 11:13:17 -04001162 if (datasync && !(inode->i_state & I_DIRTY_PAGES))
1163 goto out;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001164 /*
Chris Masona52d9a82007-08-27 16:49:44 -04001165 * ok we haven't committed the transaction yet, lets do a commit
1166 */
Chris Mason2cfbd502009-02-20 10:55:10 -05001167 if (file && file->private_data)
Sage Weil6bf13c02008-06-10 10:07:39 -04001168 btrfs_ioctl_trans_end(file);
1169
Chris Mason39279cc2007-06-12 06:35:45 -04001170 trans = btrfs_start_transaction(root, 1);
1171 if (!trans) {
1172 ret = -ENOMEM;
1173 goto out;
1174 }
Chris Masone02119d2008-09-05 16:13:11 -04001175
Chris Mason2cfbd502009-02-20 10:55:10 -05001176 ret = btrfs_log_dentry_safe(trans, root, dentry);
Chris Masond3977122009-01-05 21:25:51 -05001177 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04001178 goto out;
Chris Mason49eb7e42008-09-11 15:53:12 -04001179
1180 /* we've logged all the items and now have a consistent
1181 * version of the file in the log. It is possible that
1182 * someone will come in and modify the file, but that's
1183 * fine because the log is consistent on disk, and we
1184 * have references to all of the file's extents
1185 *
1186 * It is possible that someone will come in and log the
1187 * file again, but that will end up using the synchronization
1188 * inside btrfs_sync_log to keep things safe.
1189 */
Chris Mason2cfbd502009-02-20 10:55:10 -05001190 mutex_unlock(&dentry->d_inode->i_mutex);
Chris Mason49eb7e42008-09-11 15:53:12 -04001191
Chris Masone02119d2008-09-05 16:13:11 -04001192 if (ret > 0) {
1193 ret = btrfs_commit_transaction(trans, root);
1194 } else {
Chris Mason12fcfd22009-03-24 10:24:20 -04001195 ret = btrfs_sync_log(trans, root);
1196 if (ret == 0)
1197 ret = btrfs_end_transaction(trans, root);
1198 else
1199 ret = btrfs_commit_transaction(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001200 }
Chris Mason2cfbd502009-02-20 10:55:10 -05001201 mutex_lock(&dentry->d_inode->i_mutex);
Chris Mason39279cc2007-06-12 06:35:45 -04001202out:
1203 return ret > 0 ? EIO : ret;
1204}
1205
Chris Mason9ebefb182007-06-15 13:50:00 -04001206static struct vm_operations_struct btrfs_file_vm_ops = {
Chris Mason92fee662007-07-25 12:31:35 -04001207 .fault = filemap_fault,
Chris Mason9ebefb182007-06-15 13:50:00 -04001208 .page_mkwrite = btrfs_page_mkwrite,
1209};
1210
1211static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1212{
1213 vma->vm_ops = &btrfs_file_vm_ops;
1214 file_accessed(filp);
1215 return 0;
1216}
1217
Chris Mason39279cc2007-06-12 06:35:45 -04001218struct file_operations btrfs_file_operations = {
1219 .llseek = generic_file_llseek,
1220 .read = do_sync_read,
Chris Mason9ebefb182007-06-15 13:50:00 -04001221 .aio_read = generic_file_aio_read,
Chris Masone9906a92007-12-14 12:56:58 -05001222 .splice_read = generic_file_splice_read,
Chris Mason39279cc2007-06-12 06:35:45 -04001223 .write = btrfs_file_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001224 .mmap = btrfs_file_mmap,
Chris Mason39279cc2007-06-12 06:35:45 -04001225 .open = generic_file_open,
Mingminge1b81e62008-05-27 10:55:43 -04001226 .release = btrfs_release_file,
Chris Mason39279cc2007-06-12 06:35:45 -04001227 .fsync = btrfs_sync_file,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001228 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001229#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001230 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001231#endif
1232};