blob: 75899a01dded75042b4c99f15a44cc83882730a4 [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>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010027#include <linux/falloc.h>
Chris Mason39279cc2007-06-12 06:35:45 -040028#include <linux/swap.h>
29#include <linux/writeback.h>
30#include <linux/statfs.h>
31#include <linux/compat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/slab.h>
Chris Mason39279cc2007-06-12 06:35:45 -040033#include "ctree.h"
34#include "disk-io.h"
35#include "transaction.h"
36#include "btrfs_inode.h"
37#include "ioctl.h"
38#include "print-tree.h"
Chris Masone02119d2008-09-05 16:13:11 -040039#include "tree-log.h"
40#include "locking.h"
Jeff Mahoney12fa8ec2008-05-02 15:03:58 -040041#include "compat.h"
Chris Mason39279cc2007-06-12 06:35:45 -040042
43
Chris Masond352ac62008-09-29 15:18:18 -040044/* simple helper to fault in pages and copy. This should go away
45 * and be replaced with calls into generic code.
46 */
Chris Masond3977122009-01-05 21:25:51 -050047static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -050048 size_t write_bytes,
Chris Masona1b32a52008-09-05 16:09:51 -040049 struct page **prepared_pages,
Josef Bacik11c65dc2010-05-23 11:07:21 -040050 struct iov_iter *i)
Chris Mason39279cc2007-06-12 06:35:45 -040051{
Xin Zhong914ee292010-12-09 09:30:14 +000052 size_t copied = 0;
Josef Bacikd0215f32011-01-25 14:57:24 -050053 size_t total_copied = 0;
Josef Bacik11c65dc2010-05-23 11:07:21 -040054 int pg = 0;
Chris Mason39279cc2007-06-12 06:35:45 -040055 int offset = pos & (PAGE_CACHE_SIZE - 1);
56
Josef Bacik11c65dc2010-05-23 11:07:21 -040057 while (write_bytes > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -040058 size_t count = min_t(size_t,
59 PAGE_CACHE_SIZE - offset, write_bytes);
Josef Bacik11c65dc2010-05-23 11:07:21 -040060 struct page *page = prepared_pages[pg];
Xin Zhong914ee292010-12-09 09:30:14 +000061 /*
62 * Copy data from userspace to the current page
63 *
64 * Disable pagefault to avoid recursive lock since
65 * the pages are already locked
66 */
67 pagefault_disable();
68 copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
69 pagefault_enable();
Josef Bacik11c65dc2010-05-23 11:07:21 -040070
Chris Mason39279cc2007-06-12 06:35:45 -040071 /* Flush processor's dcache for this page */
72 flush_dcache_page(page);
Chris Mason31339ac2011-03-07 11:10:24 -050073
74 /*
75 * if we get a partial write, we can end up with
76 * partially up to date pages. These add
77 * a lot of complexity, so make sure they don't
78 * happen by forcing this copy to be retried.
79 *
80 * The rest of the btrfs_file_write code will fall
81 * back to page at a time copies after we return 0.
82 */
83 if (!PageUptodate(page) && copied < count)
84 copied = 0;
85
Josef Bacik11c65dc2010-05-23 11:07:21 -040086 iov_iter_advance(i, copied);
87 write_bytes -= copied;
Xin Zhong914ee292010-12-09 09:30:14 +000088 total_copied += copied;
Chris Mason39279cc2007-06-12 06:35:45 -040089
Xin Zhong914ee292010-12-09 09:30:14 +000090 /* Return to btrfs_file_aio_write to fault page */
Josef Bacik9f570b82011-01-25 12:42:37 -050091 if (unlikely(copied == 0))
Xin Zhong914ee292010-12-09 09:30:14 +000092 break;
Josef Bacik11c65dc2010-05-23 11:07:21 -040093
94 if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
95 offset += copied;
96 } else {
97 pg++;
98 offset = 0;
99 }
Chris Mason39279cc2007-06-12 06:35:45 -0400100 }
Xin Zhong914ee292010-12-09 09:30:14 +0000101 return total_copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400102}
103
Chris Masond352ac62008-09-29 15:18:18 -0400104/*
105 * unlocks pages after btrfs_file_write is done with them
106 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400107void btrfs_drop_pages(struct page **pages, size_t num_pages)
Chris Mason39279cc2007-06-12 06:35:45 -0400108{
109 size_t i;
110 for (i = 0; i < num_pages; i++) {
Chris Masond352ac62008-09-29 15:18:18 -0400111 /* page checked is some magic around finding pages that
112 * have been modified without going through btrfs_set_page_dirty
113 * clear it here
114 */
Chris Mason4a096752008-07-21 10:29:44 -0400115 ClearPageChecked(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400116 unlock_page(pages[i]);
117 mark_page_accessed(pages[i]);
118 page_cache_release(pages[i]);
119 }
120}
121
Chris Masond352ac62008-09-29 15:18:18 -0400122/*
123 * after copy_from_user, pages need to be dirtied and we need to make
124 * sure holes are created between the current EOF and the start of
125 * any next extents (if required).
126 *
127 * this also makes the decision about creating an inline extent vs
128 * doing real data extents, marking pages dirty and delalloc as required.
129 */
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400130int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
131 struct page **pages, size_t num_pages,
132 loff_t pos, size_t write_bytes,
133 struct extent_state **cached)
Chris Mason39279cc2007-06-12 06:35:45 -0400134{
Chris Mason39279cc2007-06-12 06:35:45 -0400135 int err = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400136 int i;
Chris Masondb945352007-10-15 16:15:53 -0400137 u64 num_bytes;
Chris Masona52d9a82007-08-27 16:49:44 -0400138 u64 start_pos;
139 u64 end_of_last_block;
140 u64 end_pos = pos + write_bytes;
141 loff_t isize = i_size_read(inode);
Chris Mason39279cc2007-06-12 06:35:45 -0400142
Chris Mason5f39d392007-10-15 16:14:19 -0400143 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masondb945352007-10-15 16:15:53 -0400144 num_bytes = (write_bytes + pos - start_pos +
145 root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
Chris Mason39279cc2007-06-12 06:35:45 -0400146
Chris Masondb945352007-10-15 16:15:53 -0400147 end_of_last_block = start_pos + num_bytes - 1;
Josef Bacik2ac55d42010-02-03 19:33:23 +0000148 err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400149 cached);
Josef Bacikd0215f32011-01-25 14:57:24 -0500150 if (err)
151 return err;
Josef Bacik9ed74f22009-09-11 16:12:44 -0400152
Chris Masonc8b97812008-10-29 14:49:59 -0400153 for (i = 0; i < num_pages; i++) {
154 struct page *p = pages[i];
155 SetPageUptodate(p);
156 ClearPageChecked(p);
157 set_page_dirty(p);
Chris Masona52d9a82007-08-27 16:49:44 -0400158 }
Josef Bacik9f570b82011-01-25 12:42:37 -0500159
160 /*
161 * we've only changed i_size in ram, and we haven't updated
162 * the disk i_size. There is no need to log the inode
163 * at this time.
164 */
165 if (end_pos > isize)
Chris Masona52d9a82007-08-27 16:49:44 -0400166 i_size_write(inode, end_pos);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400167 return 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400168}
169
Chris Masond352ac62008-09-29 15:18:18 -0400170/*
171 * this drops all the extents in the cache that intersect the range
172 * [start, end]. Existing extents are split as required.
173 */
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400174int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
175 int skip_pinned)
Chris Masona52d9a82007-08-27 16:49:44 -0400176{
177 struct extent_map *em;
Chris Mason3b951512008-04-17 11:29:12 -0400178 struct extent_map *split = NULL;
179 struct extent_map *split2 = NULL;
Chris Masona52d9a82007-08-27 16:49:44 -0400180 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Yan39b56372008-02-15 10:40:50 -0500181 u64 len = end - start + 1;
Chris Mason3b951512008-04-17 11:29:12 -0400182 int ret;
183 int testend = 1;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400184 unsigned long flags;
Chris Masonc8b97812008-10-29 14:49:59 -0400185 int compressed = 0;
Chris Masona52d9a82007-08-27 16:49:44 -0400186
Chris Masone6dcd2d2008-07-17 12:53:50 -0400187 WARN_ON(end < start);
Chris Mason3b951512008-04-17 11:29:12 -0400188 if (end == (u64)-1) {
Yan39b56372008-02-15 10:40:50 -0500189 len = (u64)-1;
Chris Mason3b951512008-04-17 11:29:12 -0400190 testend = 0;
191 }
Chris Masond3977122009-01-05 21:25:51 -0500192 while (1) {
Chris Mason3b951512008-04-17 11:29:12 -0400193 if (!split)
194 split = alloc_extent_map(GFP_NOFS);
195 if (!split2)
196 split2 = alloc_extent_map(GFP_NOFS);
Tsutomu Itohc26a9202011-02-14 00:45:29 +0000197 BUG_ON(!split || !split2);
Chris Mason3b951512008-04-17 11:29:12 -0400198
Chris Mason890871b2009-09-02 16:24:52 -0400199 write_lock(&em_tree->lock);
Yan39b56372008-02-15 10:40:50 -0500200 em = lookup_extent_mapping(em_tree, start, len);
Chris Masond1310b22008-01-24 16:13:08 -0500201 if (!em) {
Chris Mason890871b2009-09-02 16:24:52 -0400202 write_unlock(&em_tree->lock);
Chris Masona52d9a82007-08-27 16:49:44 -0400203 break;
Chris Masond1310b22008-01-24 16:13:08 -0500204 }
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400205 flags = em->flags;
206 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
Yan, Zheng55ef6892009-11-12 09:36:44 +0000207 if (testend && em->start + em->len >= start + len) {
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400208 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400209 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400210 break;
211 }
Yan, Zheng55ef6892009-11-12 09:36:44 +0000212 start = em->start + em->len;
213 if (testend)
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400214 len = start + len - (em->start + em->len);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400215 free_extent_map(em);
Chris Masona1ed8352009-09-11 12:27:37 -0400216 write_unlock(&em_tree->lock);
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400217 continue;
218 }
Chris Masonc8b97812008-10-29 14:49:59 -0400219 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
Chris Mason3ce7e672008-07-31 15:42:54 -0400220 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
Chris Masona52d9a82007-08-27 16:49:44 -0400221 remove_extent_mapping(em_tree, em);
Chris Mason3b951512008-04-17 11:29:12 -0400222
223 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
224 em->start < start) {
225 split->start = em->start;
226 split->len = start - em->start;
Yan Zhengff5b7ee2008-11-10 07:34:43 -0500227 split->orig_start = em->orig_start;
Chris Mason3b951512008-04-17 11:29:12 -0400228 split->block_start = em->block_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400229
230 if (compressed)
231 split->block_len = em->block_len;
232 else
233 split->block_len = split->len;
234
Chris Mason3b951512008-04-17 11:29:12 -0400235 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400236 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800237 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400238 ret = add_extent_mapping(em_tree, split);
239 BUG_ON(ret);
240 free_extent_map(split);
241 split = split2;
242 split2 = NULL;
243 }
244 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
245 testend && em->start + em->len > start + len) {
246 u64 diff = start + len - em->start;
247
248 split->start = start + len;
249 split->len = em->start + em->len - (start + len);
250 split->bdev = em->bdev;
Zheng Yan5b21f2e2008-09-26 10:05:38 -0400251 split->flags = flags;
Li Zefan261507a02010-12-17 14:21:50 +0800252 split->compress_type = em->compress_type;
Chris Mason3b951512008-04-17 11:29:12 -0400253
Chris Masonc8b97812008-10-29 14:49:59 -0400254 if (compressed) {
255 split->block_len = em->block_len;
256 split->block_start = em->block_start;
Chris Mason445a6942008-11-10 11:53:33 -0500257 split->orig_start = em->orig_start;
Chris Masonc8b97812008-10-29 14:49:59 -0400258 } else {
259 split->block_len = split->len;
260 split->block_start = em->block_start + diff;
Chris Mason445a6942008-11-10 11:53:33 -0500261 split->orig_start = split->start;
Chris Masonc8b97812008-10-29 14:49:59 -0400262 }
Chris Mason3b951512008-04-17 11:29:12 -0400263
264 ret = add_extent_mapping(em_tree, split);
265 BUG_ON(ret);
266 free_extent_map(split);
267 split = NULL;
268 }
Chris Mason890871b2009-09-02 16:24:52 -0400269 write_unlock(&em_tree->lock);
Chris Masond1310b22008-01-24 16:13:08 -0500270
Chris Masona52d9a82007-08-27 16:49:44 -0400271 /* once for us */
272 free_extent_map(em);
273 /* once for the tree*/
274 free_extent_map(em);
275 }
Chris Mason3b951512008-04-17 11:29:12 -0400276 if (split)
277 free_extent_map(split);
278 if (split2)
279 free_extent_map(split2);
Chris Masona52d9a82007-08-27 16:49:44 -0400280 return 0;
281}
282
Chris Mason39279cc2007-06-12 06:35:45 -0400283/*
284 * this is very complex, but the basic idea is to drop all extents
285 * in the range start - end. hint_block is filled in with a block number
286 * that would be a good hint to the block allocator for this file.
287 *
288 * If an extent intersects the range but is not entirely inside the range
289 * it is either truncated or split. Anything entirely inside the range
290 * is deleted from the tree.
291 */
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000292int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
293 u64 start, u64 end, u64 *hint_byte, int drop_cache)
Chris Mason39279cc2007-06-12 06:35:45 -0400294{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000295 struct btrfs_root *root = BTRFS_I(inode)->root;
Chris Mason00f5c792007-11-30 10:09:33 -0500296 struct extent_buffer *leaf;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000297 struct btrfs_file_extent_item *fi;
Chris Mason00f5c792007-11-30 10:09:33 -0500298 struct btrfs_path *path;
299 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000300 struct btrfs_key new_key;
301 u64 search_start = start;
302 u64 disk_bytenr = 0;
303 u64 num_bytes = 0;
304 u64 extent_offset = 0;
305 u64 extent_end = 0;
306 int del_nr = 0;
307 int del_slot = 0;
308 int extent_type;
Chris Masonccd467d2007-06-28 15:57:36 -0400309 int recow;
Chris Mason00f5c792007-11-30 10:09:33 -0500310 int ret;
Chris Mason39279cc2007-06-12 06:35:45 -0400311
Chris Masona1ed8352009-09-11 12:27:37 -0400312 if (drop_cache)
313 btrfs_drop_extent_cache(inode, start, end - 1, 0);
Chris Masona52d9a82007-08-27 16:49:44 -0400314
Chris Mason39279cc2007-06-12 06:35:45 -0400315 path = btrfs_alloc_path();
316 if (!path)
317 return -ENOMEM;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000318
Chris Masond3977122009-01-05 21:25:51 -0500319 while (1) {
Chris Masonccd467d2007-06-28 15:57:36 -0400320 recow = 0;
Chris Mason39279cc2007-06-12 06:35:45 -0400321 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
322 search_start, -1);
323 if (ret < 0)
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000324 break;
325 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
326 leaf = path->nodes[0];
327 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
328 if (key.objectid == inode->i_ino &&
329 key.type == BTRFS_EXTENT_DATA_KEY)
330 path->slots[0]--;
Chris Mason39279cc2007-06-12 06:35:45 -0400331 }
Chris Mason8c2383c2007-06-18 09:57:58 -0400332 ret = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000333next_slot:
334 leaf = path->nodes[0];
335 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
336 BUG_ON(del_nr > 0);
337 ret = btrfs_next_leaf(root, path);
338 if (ret < 0)
339 break;
340 if (ret > 0) {
341 ret = 0;
342 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400343 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000344 leaf = path->nodes[0];
345 recow = 1;
346 }
347
348 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
349 if (key.objectid > inode->i_ino ||
350 key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
351 break;
352
353 fi = btrfs_item_ptr(leaf, path->slots[0],
354 struct btrfs_file_extent_item);
355 extent_type = btrfs_file_extent_type(leaf, fi);
356
357 if (extent_type == BTRFS_FILE_EXTENT_REG ||
358 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
359 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
360 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
361 extent_offset = btrfs_file_extent_offset(leaf, fi);
362 extent_end = key.offset +
363 btrfs_file_extent_num_bytes(leaf, fi);
364 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
365 extent_end = key.offset +
366 btrfs_file_extent_inline_len(leaf, fi);
Chris Mason8c2383c2007-06-18 09:57:58 -0400367 } else {
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000368 WARN_ON(1);
Chris Mason8c2383c2007-06-18 09:57:58 -0400369 extent_end = search_start;
Chris Mason39279cc2007-06-12 06:35:45 -0400370 }
371
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000372 if (extent_end <= search_start) {
373 path->slots[0]++;
Chris Mason8c2383c2007-06-18 09:57:58 -0400374 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400375 }
376
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000377 search_start = max(key.offset, start);
378 if (recow) {
Zheng Yan31840ae2008-09-23 13:14:14 -0400379 btrfs_release_path(root, path);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000380 continue;
Chris Mason39279cc2007-06-12 06:35:45 -0400381 }
Chris Mason771ed682008-11-06 22:02:51 -0500382
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000383 /*
384 * | - range to drop - |
385 * | -------- extent -------- |
386 */
387 if (start > key.offset && end < extent_end) {
388 BUG_ON(del_nr > 0);
389 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
390
391 memcpy(&new_key, &key, sizeof(new_key));
392 new_key.offset = start;
393 ret = btrfs_duplicate_item(trans, root, path,
394 &new_key);
395 if (ret == -EAGAIN) {
396 btrfs_release_path(root, path);
397 continue;
398 }
399 if (ret < 0)
400 break;
Chris Mason8c2383c2007-06-18 09:57:58 -0400401
Chris Mason5f39d392007-10-15 16:14:19 -0400402 leaf = path->nodes[0];
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000403 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
404 struct btrfs_file_extent_item);
405 btrfs_set_file_extent_num_bytes(leaf, fi,
406 start - key.offset);
Chris Mason39279cc2007-06-12 06:35:45 -0400407
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000408 fi = btrfs_item_ptr(leaf, path->slots[0],
409 struct btrfs_file_extent_item);
Chris Masonc8b97812008-10-29 14:49:59 -0400410
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000411 extent_offset += start - key.offset;
412 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
413 btrfs_set_file_extent_num_bytes(leaf, fi,
414 extent_end - start);
415 btrfs_mark_buffer_dirty(leaf);
Chris Masondb945352007-10-15 16:15:53 -0400416
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000417 if (disk_bytenr > 0) {
418 ret = btrfs_inc_extent_ref(trans, root,
419 disk_bytenr, num_bytes, 0,
420 root->root_key.objectid,
421 new_key.objectid,
422 start - extent_offset);
Zheng Yan31840ae2008-09-23 13:14:14 -0400423 BUG_ON(ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000424 *hint_byte = disk_bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -0400425 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000426 key.offset = start;
427 }
428 /*
429 * | ---- range to drop ----- |
430 * | -------- extent -------- |
431 */
432 if (start <= key.offset && end < extent_end) {
433 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
434
435 memcpy(&new_key, &key, sizeof(new_key));
436 new_key.offset = end;
437 btrfs_set_item_key_safe(trans, root, path, &new_key);
438
439 extent_offset += end - key.offset;
440 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
441 btrfs_set_file_extent_num_bytes(leaf, fi,
442 extent_end - end);
443 btrfs_mark_buffer_dirty(leaf);
444 if (disk_bytenr > 0) {
445 inode_sub_bytes(inode, end - key.offset);
446 *hint_byte = disk_bytenr;
447 }
448 break;
Zheng Yan31840ae2008-09-23 13:14:14 -0400449 }
450
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000451 search_start = extent_end;
452 /*
453 * | ---- range to drop ----- |
454 * | -------- extent -------- |
455 */
456 if (start > key.offset && end >= extent_end) {
457 BUG_ON(del_nr > 0);
458 BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
459
460 btrfs_set_file_extent_num_bytes(leaf, fi,
461 start - key.offset);
462 btrfs_mark_buffer_dirty(leaf);
463 if (disk_bytenr > 0) {
464 inode_sub_bytes(inode, extent_end - start);
465 *hint_byte = disk_bytenr;
466 }
467 if (end == extent_end)
468 break;
469
470 path->slots[0]++;
471 goto next_slot;
Chris Mason39279cc2007-06-12 06:35:45 -0400472 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000473
474 /*
475 * | ---- range to drop ----- |
476 * | ------ extent ------ |
477 */
478 if (start <= key.offset && end >= extent_end) {
479 if (del_nr == 0) {
480 del_slot = path->slots[0];
481 del_nr = 1;
482 } else {
483 BUG_ON(del_slot + del_nr != path->slots[0]);
484 del_nr++;
485 }
486
487 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
488 inode_sub_bytes(inode,
489 extent_end - key.offset);
490 extent_end = ALIGN(extent_end,
491 root->sectorsize);
492 } else if (disk_bytenr > 0) {
493 ret = btrfs_free_extent(trans, root,
494 disk_bytenr, num_bytes, 0,
495 root->root_key.objectid,
496 key.objectid, key.offset -
497 extent_offset);
498 BUG_ON(ret);
499 inode_sub_bytes(inode,
500 extent_end - key.offset);
501 *hint_byte = disk_bytenr;
502 }
503
504 if (end == extent_end)
505 break;
506
507 if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
508 path->slots[0]++;
509 goto next_slot;
510 }
511
512 ret = btrfs_del_items(trans, root, path, del_slot,
513 del_nr);
514 BUG_ON(ret);
515
516 del_nr = 0;
517 del_slot = 0;
518
519 btrfs_release_path(root, path);
520 continue;
521 }
522
523 BUG_ON(1);
Chris Mason39279cc2007-06-12 06:35:45 -0400524 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000525
526 if (del_nr > 0) {
527 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
528 BUG_ON(ret);
529 }
530
Chris Mason39279cc2007-06-12 06:35:45 -0400531 btrfs_free_path(path);
532 return ret;
533}
534
Yan Zhengd899e052008-10-30 14:25:28 -0400535static int extent_mergeable(struct extent_buffer *leaf, int slot,
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000536 u64 objectid, u64 bytenr, u64 orig_offset,
537 u64 *start, u64 *end)
Yan Zhengd899e052008-10-30 14:25:28 -0400538{
539 struct btrfs_file_extent_item *fi;
540 struct btrfs_key key;
541 u64 extent_end;
542
543 if (slot < 0 || slot >= btrfs_header_nritems(leaf))
544 return 0;
545
546 btrfs_item_key_to_cpu(leaf, &key, slot);
547 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
548 return 0;
549
550 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
551 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
552 btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000553 btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Yan Zhengd899e052008-10-30 14:25:28 -0400554 btrfs_file_extent_compression(leaf, fi) ||
555 btrfs_file_extent_encryption(leaf, fi) ||
556 btrfs_file_extent_other_encoding(leaf, fi))
557 return 0;
558
559 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
560 if ((*start && *start != key.offset) || (*end && *end != extent_end))
561 return 0;
562
563 *start = key.offset;
564 *end = extent_end;
565 return 1;
566}
567
568/*
569 * Mark extent in the range start - end as written.
570 *
571 * This changes extent type from 'pre-allocated' to 'regular'. If only
572 * part of extent is marked as written, the extent will be split into
573 * two or three.
574 */
575int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
Yan Zhengd899e052008-10-30 14:25:28 -0400576 struct inode *inode, u64 start, u64 end)
577{
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000578 struct btrfs_root *root = BTRFS_I(inode)->root;
Yan Zhengd899e052008-10-30 14:25:28 -0400579 struct extent_buffer *leaf;
580 struct btrfs_path *path;
581 struct btrfs_file_extent_item *fi;
582 struct btrfs_key key;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000583 struct btrfs_key new_key;
Yan Zhengd899e052008-10-30 14:25:28 -0400584 u64 bytenr;
585 u64 num_bytes;
586 u64 extent_end;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400587 u64 orig_offset;
Yan Zhengd899e052008-10-30 14:25:28 -0400588 u64 other_start;
589 u64 other_end;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000590 u64 split;
591 int del_nr = 0;
592 int del_slot = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000593 int recow;
Yan Zhengd899e052008-10-30 14:25:28 -0400594 int ret;
595
596 btrfs_drop_extent_cache(inode, start, end - 1, 0);
597
598 path = btrfs_alloc_path();
599 BUG_ON(!path);
600again:
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000601 recow = 0;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000602 split = start;
Yan Zhengd899e052008-10-30 14:25:28 -0400603 key.objectid = inode->i_ino;
604 key.type = BTRFS_EXTENT_DATA_KEY;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000605 key.offset = split;
Yan Zhengd899e052008-10-30 14:25:28 -0400606
607 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
Josef Bacik41415732011-03-16 13:59:32 -0400608 if (ret < 0)
609 goto out;
Yan Zhengd899e052008-10-30 14:25:28 -0400610 if (ret > 0 && path->slots[0] > 0)
611 path->slots[0]--;
612
613 leaf = path->nodes[0];
614 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
615 BUG_ON(key.objectid != inode->i_ino ||
616 key.type != BTRFS_EXTENT_DATA_KEY);
617 fi = btrfs_item_ptr(leaf, path->slots[0],
618 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000619 BUG_ON(btrfs_file_extent_type(leaf, fi) !=
620 BTRFS_FILE_EXTENT_PREALLOC);
Yan Zhengd899e052008-10-30 14:25:28 -0400621 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
622 BUG_ON(key.offset > start || extent_end < end);
623
624 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
625 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400626 orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000627 memcpy(&new_key, &key, sizeof(new_key));
628
629 if (start == key.offset && end < extent_end) {
630 other_start = 0;
631 other_end = start;
632 if (extent_mergeable(leaf, path->slots[0] - 1,
633 inode->i_ino, bytenr, orig_offset,
634 &other_start, &other_end)) {
635 new_key.offset = end;
636 btrfs_set_item_key_safe(trans, root, path, &new_key);
637 fi = btrfs_item_ptr(leaf, path->slots[0],
638 struct btrfs_file_extent_item);
639 btrfs_set_file_extent_num_bytes(leaf, fi,
640 extent_end - end);
641 btrfs_set_file_extent_offset(leaf, fi,
642 end - orig_offset);
643 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
644 struct btrfs_file_extent_item);
645 btrfs_set_file_extent_num_bytes(leaf, fi,
646 end - other_start);
647 btrfs_mark_buffer_dirty(leaf);
648 goto out;
649 }
650 }
651
652 if (start > key.offset && end == extent_end) {
653 other_start = end;
654 other_end = 0;
655 if (extent_mergeable(leaf, path->slots[0] + 1,
656 inode->i_ino, bytenr, orig_offset,
657 &other_start, &other_end)) {
658 fi = btrfs_item_ptr(leaf, path->slots[0],
659 struct btrfs_file_extent_item);
660 btrfs_set_file_extent_num_bytes(leaf, fi,
661 start - key.offset);
662 path->slots[0]++;
663 new_key.offset = start;
664 btrfs_set_item_key_safe(trans, root, path, &new_key);
665
666 fi = btrfs_item_ptr(leaf, path->slots[0],
667 struct btrfs_file_extent_item);
668 btrfs_set_file_extent_num_bytes(leaf, fi,
669 other_end - start);
670 btrfs_set_file_extent_offset(leaf, fi,
671 start - orig_offset);
672 btrfs_mark_buffer_dirty(leaf);
673 goto out;
674 }
675 }
Yan Zhengd899e052008-10-30 14:25:28 -0400676
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000677 while (start > key.offset || end < extent_end) {
678 if (key.offset == start)
679 split = end;
Yan Zhengd899e052008-10-30 14:25:28 -0400680
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000681 new_key.offset = split;
682 ret = btrfs_duplicate_item(trans, root, path, &new_key);
683 if (ret == -EAGAIN) {
684 btrfs_release_path(root, path);
685 goto again;
Yan Zhengd899e052008-10-30 14:25:28 -0400686 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000687 BUG_ON(ret < 0);
Yan Zhengd899e052008-10-30 14:25:28 -0400688
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000689 leaf = path->nodes[0];
690 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Yan Zhengd899e052008-10-30 14:25:28 -0400691 struct btrfs_file_extent_item);
Yan Zhengd899e052008-10-30 14:25:28 -0400692 btrfs_set_file_extent_num_bytes(leaf, fi,
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000693 split - key.offset);
694
695 fi = btrfs_item_ptr(leaf, path->slots[0],
696 struct btrfs_file_extent_item);
697
698 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
699 btrfs_set_file_extent_num_bytes(leaf, fi,
700 extent_end - split);
Yan Zhengd899e052008-10-30 14:25:28 -0400701 btrfs_mark_buffer_dirty(leaf);
702
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000703 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
704 root->root_key.objectid,
705 inode->i_ino, orig_offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400706 BUG_ON(ret);
Yan Zhengd899e052008-10-30 14:25:28 -0400707
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000708 if (split == start) {
709 key.offset = start;
710 } else {
711 BUG_ON(start != key.offset);
Yan Zhengd899e052008-10-30 14:25:28 -0400712 path->slots[0]--;
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000713 extent_end = end;
Yan Zhengd899e052008-10-30 14:25:28 -0400714 }
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000715 recow = 1;
Yan Zhengd899e052008-10-30 14:25:28 -0400716 }
717
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000718 other_start = end;
719 other_end = 0;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000720 if (extent_mergeable(leaf, path->slots[0] + 1,
721 inode->i_ino, bytenr, orig_offset,
722 &other_start, &other_end)) {
723 if (recow) {
724 btrfs_release_path(root, path);
725 goto again;
726 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000727 extent_end = other_end;
728 del_slot = path->slots[0] + 1;
729 del_nr++;
730 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
731 0, root->root_key.objectid,
732 inode->i_ino, orig_offset);
733 BUG_ON(ret);
734 }
735 other_start = 0;
736 other_end = start;
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000737 if (extent_mergeable(leaf, path->slots[0] - 1,
738 inode->i_ino, bytenr, orig_offset,
739 &other_start, &other_end)) {
740 if (recow) {
741 btrfs_release_path(root, path);
742 goto again;
743 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000744 key.offset = other_start;
745 del_slot = path->slots[0];
746 del_nr++;
747 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
748 0, root->root_key.objectid,
749 inode->i_ino, orig_offset);
750 BUG_ON(ret);
751 }
752 if (del_nr == 0) {
Shaohua Li3f6fae92010-02-11 07:43:00 +0000753 fi = btrfs_item_ptr(leaf, path->slots[0],
754 struct btrfs_file_extent_item);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000755 btrfs_set_file_extent_type(leaf, fi,
756 BTRFS_FILE_EXTENT_REG);
757 btrfs_mark_buffer_dirty(leaf);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000758 } else {
Shaohua Li3f6fae92010-02-11 07:43:00 +0000759 fi = btrfs_item_ptr(leaf, del_slot - 1,
760 struct btrfs_file_extent_item);
Yan, Zheng6c7d54a2010-01-15 08:43:09 +0000761 btrfs_set_file_extent_type(leaf, fi,
762 BTRFS_FILE_EXTENT_REG);
763 btrfs_set_file_extent_num_bytes(leaf, fi,
764 extent_end - key.offset);
765 btrfs_mark_buffer_dirty(leaf);
766
767 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
768 BUG_ON(ret);
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000769 }
Yan, Zheng920bbbf2009-11-12 09:34:08 +0000770out:
Yan Zhengd899e052008-10-30 14:25:28 -0400771 btrfs_free_path(path);
772 return 0;
773}
774
Chris Mason39279cc2007-06-12 06:35:45 -0400775/*
Chris Masonb1bf8622011-02-28 09:52:08 -0500776 * on error we return an unlocked page and the error value
777 * on success we return a locked page and 0
778 */
779static int prepare_uptodate_page(struct page *page, u64 pos)
780{
781 int ret = 0;
782
783 if ((pos & (PAGE_CACHE_SIZE - 1)) && !PageUptodate(page)) {
784 ret = btrfs_readpage(NULL, page);
785 if (ret)
786 return ret;
787 lock_page(page);
788 if (!PageUptodate(page)) {
789 unlock_page(page);
790 return -EIO;
791 }
792 }
793 return 0;
794}
795
796/*
Chris Masond352ac62008-09-29 15:18:18 -0400797 * this gets pages into the page cache and locks them down, it also properly
798 * waits for data=ordered extents to finish before allowing the pages to be
799 * modified.
Chris Mason39279cc2007-06-12 06:35:45 -0400800 */
Chris Masond3977122009-01-05 21:25:51 -0500801static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
Chris Mason98ed5172008-01-03 10:01:48 -0500802 struct page **pages, size_t num_pages,
803 loff_t pos, unsigned long first_index,
804 unsigned long last_index, size_t write_bytes)
Chris Mason39279cc2007-06-12 06:35:45 -0400805{
Josef Bacik2ac55d42010-02-03 19:33:23 +0000806 struct extent_state *cached_state = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -0400807 int i;
808 unsigned long index = pos >> PAGE_CACHE_SHIFT;
Chris Mason6da6aba2007-12-18 16:15:09 -0500809 struct inode *inode = fdentry(file)->d_inode;
Chris Mason39279cc2007-06-12 06:35:45 -0400810 int err = 0;
Chris Masonb1bf8622011-02-28 09:52:08 -0500811 int faili = 0;
Chris Mason8c2383c2007-06-18 09:57:58 -0400812 u64 start_pos;
Chris Masone6dcd2d2008-07-17 12:53:50 -0400813 u64 last_pos;
Chris Mason8c2383c2007-06-18 09:57:58 -0400814
Chris Mason5f39d392007-10-15 16:14:19 -0400815 start_pos = pos & ~((u64)root->sectorsize - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400816 last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -0400817
Yan Zheng9036c102008-10-30 14:19:41 -0400818 if (start_pos > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -0500819 err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
Yan Zheng9036c102008-10-30 14:19:41 -0400820 if (err)
821 return err;
822 }
823
Chris Masone6dcd2d2008-07-17 12:53:50 -0400824again:
Chris Mason39279cc2007-06-12 06:35:45 -0400825 for (i = 0; i < num_pages; i++) {
826 pages[i] = grab_cache_page(inode->i_mapping, index + i);
827 if (!pages[i]) {
Chris Masonb1bf8622011-02-28 09:52:08 -0500828 faili = i - 1;
829 err = -ENOMEM;
830 goto fail;
831 }
832
833 if (i == 0)
834 err = prepare_uptodate_page(pages[i], pos);
835 if (i == num_pages - 1)
836 err = prepare_uptodate_page(pages[i],
837 pos + write_bytes);
838 if (err) {
839 page_cache_release(pages[i]);
840 faili = i - 1;
841 goto fail;
Chris Mason39279cc2007-06-12 06:35:45 -0400842 }
Chris Masonccd467d2007-06-28 15:57:36 -0400843 wait_on_page_writeback(pages[i]);
Chris Mason39279cc2007-06-12 06:35:45 -0400844 }
Chris Masonb1bf8622011-02-28 09:52:08 -0500845 err = 0;
Chris Mason07627042008-02-19 11:29:24 -0500846 if (start_pos < inode->i_size) {
Chris Masone6dcd2d2008-07-17 12:53:50 -0400847 struct btrfs_ordered_extent *ordered;
Josef Bacik2ac55d42010-02-03 19:33:23 +0000848 lock_extent_bits(&BTRFS_I(inode)->io_tree,
849 start_pos, last_pos - 1, 0, &cached_state,
850 GFP_NOFS);
Chris Masond3977122009-01-05 21:25:51 -0500851 ordered = btrfs_lookup_first_ordered_extent(inode,
852 last_pos - 1);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400853 if (ordered &&
854 ordered->file_offset + ordered->len > start_pos &&
855 ordered->file_offset < last_pos) {
856 btrfs_put_ordered_extent(ordered);
Josef Bacik2ac55d42010-02-03 19:33:23 +0000857 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
858 start_pos, last_pos - 1,
859 &cached_state, GFP_NOFS);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400860 for (i = 0; i < num_pages; i++) {
861 unlock_page(pages[i]);
862 page_cache_release(pages[i]);
863 }
864 btrfs_wait_ordered_range(inode, start_pos,
865 last_pos - start_pos);
866 goto again;
867 }
868 if (ordered)
869 btrfs_put_ordered_extent(ordered);
870
Josef Bacik2ac55d42010-02-03 19:33:23 +0000871 clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
Josef Bacik32c00af2009-10-08 13:34:05 -0400872 last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
Josef Bacik2ac55d42010-02-03 19:33:23 +0000873 EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
Chris Mason07627042008-02-19 11:29:24 -0500874 GFP_NOFS);
Josef Bacik2ac55d42010-02-03 19:33:23 +0000875 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
876 start_pos, last_pos - 1, &cached_state,
877 GFP_NOFS);
Chris Mason07627042008-02-19 11:29:24 -0500878 }
Chris Masone6dcd2d2008-07-17 12:53:50 -0400879 for (i = 0; i < num_pages; i++) {
Chris Masonf87f0572008-08-01 11:27:23 -0400880 clear_page_dirty_for_io(pages[i]);
Chris Masone6dcd2d2008-07-17 12:53:50 -0400881 set_page_extent_mapped(pages[i]);
882 WARN_ON(!PageLocked(pages[i]));
883 }
Chris Mason39279cc2007-06-12 06:35:45 -0400884 return 0;
Chris Masonb1bf8622011-02-28 09:52:08 -0500885fail:
886 while (faili >= 0) {
887 unlock_page(pages[faili]);
888 page_cache_release(pages[faili]);
889 faili--;
890 }
891 return err;
892
Chris Mason39279cc2007-06-12 06:35:45 -0400893}
894
Josef Bacikd0215f32011-01-25 14:57:24 -0500895static noinline ssize_t __btrfs_buffered_write(struct file *file,
896 struct iov_iter *i,
897 loff_t pos)
Josef Bacik4b46fce2010-05-23 11:00:55 -0400898{
Josef Bacik11c65dc2010-05-23 11:07:21 -0400899 struct inode *inode = fdentry(file)->d_inode;
900 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik11c65dc2010-05-23 11:07:21 -0400901 struct page **pages = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -0400902 unsigned long first_index;
903 unsigned long last_index;
Josef Bacikd0215f32011-01-25 14:57:24 -0500904 size_t num_written = 0;
905 int nrptrs;
Tsutomu Itohc9149232011-03-30 00:57:23 +0000906 int ret = 0;
Chris Masoncb843a62008-10-03 12:30:02 -0400907
Josef Bacikd0215f32011-01-25 14:57:24 -0500908 nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
Josef Bacik11c65dc2010-05-23 11:07:21 -0400909 PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
910 (sizeof(struct page *)));
Chris Mason8c2383c2007-06-18 09:57:58 -0400911 pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
Josef Bacikd0215f32011-01-25 14:57:24 -0500912 if (!pages)
913 return -ENOMEM;
Chris Masonab93dbe2009-10-01 12:29:10 -0400914
Chris Mason39279cc2007-06-12 06:35:45 -0400915 first_index = pos >> PAGE_CACHE_SHIFT;
Josef Bacikd0215f32011-01-25 14:57:24 -0500916 last_index = (pos + iov_iter_count(i)) >> PAGE_CACHE_SHIFT;
Chris Mason39279cc2007-06-12 06:35:45 -0400917
Josef Bacikd0215f32011-01-25 14:57:24 -0500918 while (iov_iter_count(i) > 0) {
Chris Mason39279cc2007-06-12 06:35:45 -0400919 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
Josef Bacikd0215f32011-01-25 14:57:24 -0500920 size_t write_bytes = min(iov_iter_count(i),
Josef Bacik11c65dc2010-05-23 11:07:21 -0400921 nrptrs * (size_t)PAGE_CACHE_SIZE -
Chris Mason8c2383c2007-06-18 09:57:58 -0400922 offset);
Yan, Zheng3a909832011-01-18 13:34:40 +0800923 size_t num_pages = (write_bytes + offset +
924 PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Josef Bacikd0215f32011-01-25 14:57:24 -0500925 size_t dirty_pages;
926 size_t copied;
Chris Mason39279cc2007-06-12 06:35:45 -0400927
Chris Mason8c2383c2007-06-18 09:57:58 -0400928 WARN_ON(num_pages > nrptrs);
Chris Mason1832a6d2007-12-21 16:27:21 -0500929
Xin Zhong914ee292010-12-09 09:30:14 +0000930 /*
931 * Fault pages before locking them in prepare_pages
932 * to avoid recursive lock
933 */
Josef Bacikd0215f32011-01-25 14:57:24 -0500934 if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
Xin Zhong914ee292010-12-09 09:30:14 +0000935 ret = -EFAULT;
Josef Bacikd0215f32011-01-25 14:57:24 -0500936 break;
Xin Zhong914ee292010-12-09 09:30:14 +0000937 }
938
939 ret = btrfs_delalloc_reserve_space(inode,
940 num_pages << PAGE_CACHE_SHIFT);
Chris Mason1832a6d2007-12-21 16:27:21 -0500941 if (ret)
Josef Bacikd0215f32011-01-25 14:57:24 -0500942 break;
Chris Mason1832a6d2007-12-21 16:27:21 -0500943
Josef Bacik4a640012011-01-25 15:10:08 -0500944 /*
945 * This is going to setup the pages array with the number of
946 * pages we want, so we don't really need to worry about the
947 * contents of pages from loop to loop
948 */
Chris Mason39279cc2007-06-12 06:35:45 -0400949 ret = prepare_pages(root, file, pages, num_pages,
950 pos, first_index, last_index,
Chris Mason8c2383c2007-06-18 09:57:58 -0400951 write_bytes);
Josef Bacik6a632092009-02-20 11:00:09 -0500952 if (ret) {
Xin Zhong914ee292010-12-09 09:30:14 +0000953 btrfs_delalloc_release_space(inode,
954 num_pages << PAGE_CACHE_SHIFT);
Josef Bacikd0215f32011-01-25 14:57:24 -0500955 break;
Josef Bacik6a632092009-02-20 11:00:09 -0500956 }
Chris Mason39279cc2007-06-12 06:35:45 -0400957
Xin Zhong914ee292010-12-09 09:30:14 +0000958 copied = btrfs_copy_from_user(pos, num_pages,
Josef Bacikd0215f32011-01-25 14:57:24 -0500959 write_bytes, pages, i);
Chris Masonb1bf8622011-02-28 09:52:08 -0500960
961 /*
962 * if we have trouble faulting in the pages, fall
963 * back to one page at a time
964 */
965 if (copied < write_bytes)
966 nrptrs = 1;
967
968 if (copied == 0)
969 dirty_pages = 0;
970 else
971 dirty_pages = (copied + offset +
972 PAGE_CACHE_SIZE - 1) >>
973 PAGE_CACHE_SHIFT;
Xin Zhong914ee292010-12-09 09:30:14 +0000974
Josef Bacikd0215f32011-01-25 14:57:24 -0500975 /*
976 * If we had a short copy we need to release the excess delaloc
977 * bytes we reserved. We need to increment outstanding_extents
978 * because btrfs_delalloc_release_space will decrement it, but
979 * we still have an outstanding extent for the chunk we actually
980 * managed to copy.
981 */
Xin Zhong914ee292010-12-09 09:30:14 +0000982 if (num_pages > dirty_pages) {
983 if (copied > 0)
984 atomic_inc(
985 &BTRFS_I(inode)->outstanding_extents);
986 btrfs_delalloc_release_space(inode,
987 (num_pages - dirty_pages) <<
988 PAGE_CACHE_SHIFT);
989 }
990
991 if (copied > 0) {
Josef Bacikbe1a12a2011-04-06 13:05:22 -0400992 ret = btrfs_dirty_pages(root, inode, pages,
993 dirty_pages, pos, copied,
994 NULL);
Josef Bacikd0215f32011-01-25 14:57:24 -0500995 if (ret) {
996 btrfs_delalloc_release_space(inode,
997 dirty_pages << PAGE_CACHE_SHIFT);
998 btrfs_drop_pages(pages, num_pages);
999 break;
1000 }
Chris Mason54aa1f42007-06-22 14:16:25 -04001001 }
Chris Mason39279cc2007-06-12 06:35:45 -04001002
Chris Mason39279cc2007-06-12 06:35:45 -04001003 btrfs_drop_pages(pages, num_pages);
Xin Zhong914ee292010-12-09 09:30:14 +00001004
Josef Bacikd0215f32011-01-25 14:57:24 -05001005 cond_resched();
1006
1007 balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1008 dirty_pages);
1009 if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1010 btrfs_btree_balance_dirty(root, 1);
1011 btrfs_throttle(root);
Chris Mason39279cc2007-06-12 06:35:45 -04001012
Xin Zhong914ee292010-12-09 09:30:14 +00001013 pos += copied;
1014 num_written += copied;
Chris Mason39279cc2007-06-12 06:35:45 -04001015 }
Chris Mason5b92ee72008-01-03 13:46:11 -05001016
Chris Mason8c2383c2007-06-18 09:57:58 -04001017 kfree(pages);
Josef Bacikd0215f32011-01-25 14:57:24 -05001018
1019 return num_written ? num_written : ret;
1020}
1021
1022static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1023 const struct iovec *iov,
1024 unsigned long nr_segs, loff_t pos,
1025 loff_t *ppos, size_t count, size_t ocount)
1026{
1027 struct file *file = iocb->ki_filp;
1028 struct inode *inode = fdentry(file)->d_inode;
1029 struct iov_iter i;
1030 ssize_t written;
1031 ssize_t written_buffered;
1032 loff_t endbyte;
1033 int err;
1034
1035 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1036 count, ocount);
1037
1038 /*
1039 * the generic O_DIRECT will update in-memory i_size after the
1040 * DIOs are done. But our endio handlers that update the on
1041 * disk i_size never update past the in memory i_size. So we
1042 * need one more update here to catch any additions to the
1043 * file
1044 */
1045 if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1046 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1047 mark_inode_dirty(inode);
1048 }
1049
1050 if (written < 0 || written == count)
1051 return written;
1052
1053 pos += written;
1054 count -= written;
1055 iov_iter_init(&i, iov, nr_segs, count, written);
1056 written_buffered = __btrfs_buffered_write(file, &i, pos);
1057 if (written_buffered < 0) {
1058 err = written_buffered;
1059 goto out;
1060 }
1061 endbyte = pos + written_buffered - 1;
1062 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1063 if (err)
1064 goto out;
1065 written += written_buffered;
1066 *ppos = pos + written_buffered;
1067 invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1068 endbyte >> PAGE_CACHE_SHIFT);
1069out:
1070 return written ? written : err;
1071}
1072
1073static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1074 const struct iovec *iov,
1075 unsigned long nr_segs, loff_t pos)
1076{
1077 struct file *file = iocb->ki_filp;
1078 struct inode *inode = fdentry(file)->d_inode;
1079 struct btrfs_root *root = BTRFS_I(inode)->root;
1080 loff_t *ppos = &iocb->ki_pos;
1081 ssize_t num_written = 0;
1082 ssize_t err = 0;
1083 size_t count, ocount;
1084
1085 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1086
1087 mutex_lock(&inode->i_mutex);
1088
1089 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1090 if (err) {
1091 mutex_unlock(&inode->i_mutex);
1092 goto out;
1093 }
1094 count = ocount;
1095
1096 current->backing_dev_info = inode->i_mapping->backing_dev_info;
1097 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1098 if (err) {
1099 mutex_unlock(&inode->i_mutex);
1100 goto out;
1101 }
1102
1103 if (count == 0) {
1104 mutex_unlock(&inode->i_mutex);
1105 goto out;
1106 }
1107
1108 err = file_remove_suid(file);
1109 if (err) {
1110 mutex_unlock(&inode->i_mutex);
1111 goto out;
1112 }
1113
1114 /*
1115 * If BTRFS flips readonly due to some impossible error
1116 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1117 * although we have opened a file as writable, we have
1118 * to stop this write operation to ensure FS consistency.
1119 */
1120 if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1121 mutex_unlock(&inode->i_mutex);
1122 err = -EROFS;
1123 goto out;
1124 }
1125
1126 file_update_time(file);
1127 BTRFS_I(inode)->sequence++;
1128
1129 if (unlikely(file->f_flags & O_DIRECT)) {
1130 num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1131 pos, ppos, count, ocount);
1132 } else {
1133 struct iov_iter i;
1134
1135 iov_iter_init(&i, iov, nr_segs, count, num_written);
1136
1137 num_written = __btrfs_buffered_write(file, &i, pos);
1138 if (num_written > 0)
1139 *ppos = pos + num_written;
1140 }
1141
1142 mutex_unlock(&inode->i_mutex);
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001143
Chris Mason5a3f23d2009-03-31 13:27:11 -04001144 /*
1145 * we want to make sure fsync finds this change
1146 * but we haven't joined a transaction running right now.
1147 *
1148 * Later on, someone is sure to update the inode and get the
1149 * real transid recorded.
1150 *
1151 * We set last_trans now to the fs_info generation + 1,
1152 * this will either be one more than the running transaction
1153 * or the generation used for the next transaction if there isn't
1154 * one running right now.
1155 */
1156 BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
Josef Bacikd0215f32011-01-25 14:57:24 -05001157 if (num_written > 0 || num_written == -EIOCBQUEUED) {
1158 err = generic_write_sync(file, pos, num_written);
1159 if (err < 0 && num_written > 0)
Chris Mason2ff3e9b2007-10-29 14:36:41 -04001160 num_written = err;
1161 }
Josef Bacikd0215f32011-01-25 14:57:24 -05001162out:
Chris Mason39279cc2007-06-12 06:35:45 -04001163 current->backing_dev_info = NULL;
Chris Mason39279cc2007-06-12 06:35:45 -04001164 return num_written ? num_written : err;
1165}
1166
Chris Masond3977122009-01-05 21:25:51 -05001167int btrfs_release_file(struct inode *inode, struct file *filp)
Mingminge1b81e62008-05-27 10:55:43 -04001168{
Chris Mason5a3f23d2009-03-31 13:27:11 -04001169 /*
1170 * ordered_data_close is set by settattr when we are about to truncate
1171 * a file from a non-zero size to a zero size. This tries to
1172 * flush down new bytes that may have been written if the
1173 * application were using truncate to replace a file in place.
1174 */
1175 if (BTRFS_I(inode)->ordered_data_close) {
1176 BTRFS_I(inode)->ordered_data_close = 0;
1177 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1178 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1179 filemap_flush(inode->i_mapping);
1180 }
Sage Weil6bf13c02008-06-10 10:07:39 -04001181 if (filp->private_data)
1182 btrfs_ioctl_trans_end(filp);
Mingminge1b81e62008-05-27 10:55:43 -04001183 return 0;
1184}
1185
Chris Masond352ac62008-09-29 15:18:18 -04001186/*
1187 * fsync call for both files and directories. This logs the inode into
1188 * the tree log instead of forcing full commits whenever possible.
1189 *
1190 * It needs to call filemap_fdatawait so that all ordered extent updates are
1191 * in the metadata btree are up to date for copying to the log.
1192 *
1193 * It drops the inode mutex before doing the tree log commit. This is an
1194 * important optimization for directories because holding the mutex prevents
1195 * new operations on the dir while we write to disk.
1196 */
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001197int btrfs_sync_file(struct file *file, int datasync)
Chris Mason39279cc2007-06-12 06:35:45 -04001198{
Christoph Hellwig7ea80852010-05-26 17:53:25 +02001199 struct dentry *dentry = file->f_path.dentry;
Chris Mason39279cc2007-06-12 06:35:45 -04001200 struct inode *inode = dentry->d_inode;
1201 struct btrfs_root *root = BTRFS_I(inode)->root;
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001202 int ret = 0;
Chris Mason39279cc2007-06-12 06:35:45 -04001203 struct btrfs_trans_handle *trans;
1204
liubo1abe9b82011-03-24 11:18:59 +00001205 trace_btrfs_sync_file(file, datasync);
Chris Mason257c62e2009-10-13 13:21:08 -04001206
1207 /* we wait first, since the writeback may change the inode */
1208 root->log_batch++;
1209 /* the VFS called filemap_fdatawrite for us */
1210 btrfs_wait_ordered_range(inode, 0, (u64)-1);
1211 root->log_batch++;
1212
Chris Mason39279cc2007-06-12 06:35:45 -04001213 /*
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001214 * check the transaction that last modified this inode
1215 * and see if its already been committed
Chris Mason39279cc2007-06-12 06:35:45 -04001216 */
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001217 if (!BTRFS_I(inode)->last_trans)
1218 goto out;
Chris Masona2135012008-06-25 16:01:30 -04001219
Chris Mason257c62e2009-10-13 13:21:08 -04001220 /*
1221 * if the last transaction that changed this file was before
1222 * the current transaction, we can bail out now without any
1223 * syncing
1224 */
Josef Bacik15ee9bc2007-08-10 16:22:09 -04001225 mutex_lock(&root->fs_info->trans_mutex);
1226 if (BTRFS_I(inode)->last_trans <=
1227 root->fs_info->last_trans_committed) {
1228 BTRFS_I(inode)->last_trans = 0;
1229 mutex_unlock(&root->fs_info->trans_mutex);
1230 goto out;
1231 }
1232 mutex_unlock(&root->fs_info->trans_mutex);
1233
1234 /*
Chris Masona52d9a82007-08-27 16:49:44 -04001235 * ok we haven't committed the transaction yet, lets do a commit
1236 */
Dan Carpenter6f902af2010-05-29 09:49:07 +00001237 if (file->private_data)
Sage Weil6bf13c02008-06-10 10:07:39 -04001238 btrfs_ioctl_trans_end(file);
1239
Yan, Zhenga22285a2010-05-16 10:48:46 -04001240 trans = btrfs_start_transaction(root, 0);
1241 if (IS_ERR(trans)) {
1242 ret = PTR_ERR(trans);
Chris Mason39279cc2007-06-12 06:35:45 -04001243 goto out;
1244 }
Chris Masone02119d2008-09-05 16:13:11 -04001245
Chris Mason2cfbd502009-02-20 10:55:10 -05001246 ret = btrfs_log_dentry_safe(trans, root, dentry);
Chris Masond3977122009-01-05 21:25:51 -05001247 if (ret < 0)
Chris Masone02119d2008-09-05 16:13:11 -04001248 goto out;
Chris Mason49eb7e42008-09-11 15:53:12 -04001249
1250 /* we've logged all the items and now have a consistent
1251 * version of the file in the log. It is possible that
1252 * someone will come in and modify the file, but that's
1253 * fine because the log is consistent on disk, and we
1254 * have references to all of the file's extents
1255 *
1256 * It is possible that someone will come in and log the
1257 * file again, but that will end up using the synchronization
1258 * inside btrfs_sync_log to keep things safe.
1259 */
Chris Mason2cfbd502009-02-20 10:55:10 -05001260 mutex_unlock(&dentry->d_inode->i_mutex);
Chris Mason49eb7e42008-09-11 15:53:12 -04001261
Chris Mason257c62e2009-10-13 13:21:08 -04001262 if (ret != BTRFS_NO_LOG_SYNC) {
1263 if (ret > 0) {
Chris Mason12fcfd22009-03-24 10:24:20 -04001264 ret = btrfs_commit_transaction(trans, root);
Chris Mason257c62e2009-10-13 13:21:08 -04001265 } else {
1266 ret = btrfs_sync_log(trans, root);
1267 if (ret == 0)
1268 ret = btrfs_end_transaction(trans, root);
1269 else
1270 ret = btrfs_commit_transaction(trans, root);
1271 }
1272 } else {
1273 ret = btrfs_end_transaction(trans, root);
Chris Masone02119d2008-09-05 16:13:11 -04001274 }
Chris Mason2cfbd502009-02-20 10:55:10 -05001275 mutex_lock(&dentry->d_inode->i_mutex);
Chris Mason39279cc2007-06-12 06:35:45 -04001276out:
Roel Kluin014e4ac2010-01-29 10:42:11 +00001277 return ret > 0 ? -EIO : ret;
Chris Mason39279cc2007-06-12 06:35:45 -04001278}
1279
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001280static const struct vm_operations_struct btrfs_file_vm_ops = {
Chris Mason92fee662007-07-25 12:31:35 -04001281 .fault = filemap_fault,
Chris Mason9ebefb182007-06-15 13:50:00 -04001282 .page_mkwrite = btrfs_page_mkwrite,
1283};
1284
1285static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
1286{
Miao Xie058a4572010-05-20 07:21:50 +00001287 struct address_space *mapping = filp->f_mapping;
1288
1289 if (!mapping->a_ops->readpage)
1290 return -ENOEXEC;
1291
Chris Mason9ebefb182007-06-15 13:50:00 -04001292 file_accessed(filp);
Miao Xie058a4572010-05-20 07:21:50 +00001293 vma->vm_ops = &btrfs_file_vm_ops;
1294 vma->vm_flags |= VM_CAN_NONLINEAR;
1295
Chris Mason9ebefb182007-06-15 13:50:00 -04001296 return 0;
1297}
1298
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001299static long btrfs_fallocate(struct file *file, int mode,
1300 loff_t offset, loff_t len)
1301{
1302 struct inode *inode = file->f_path.dentry->d_inode;
1303 struct extent_state *cached_state = NULL;
1304 u64 cur_offset;
1305 u64 last_byte;
1306 u64 alloc_start;
1307 u64 alloc_end;
1308 u64 alloc_hint = 0;
1309 u64 locked_end;
1310 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1311 struct extent_map *em;
1312 int ret;
1313
1314 alloc_start = offset & ~mask;
1315 alloc_end = (offset + len + mask) & ~mask;
1316
1317 /* We only support the FALLOC_FL_KEEP_SIZE mode */
1318 if (mode & ~FALLOC_FL_KEEP_SIZE)
1319 return -EOPNOTSUPP;
1320
1321 /*
1322 * wait for ordered IO before we have any locks. We'll loop again
1323 * below with the locks held.
1324 */
1325 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1326
1327 mutex_lock(&inode->i_mutex);
1328 ret = inode_newsize_ok(inode, alloc_end);
1329 if (ret)
1330 goto out;
1331
1332 if (alloc_start > inode->i_size) {
Josef Bacika41ad392011-01-31 15:30:16 -05001333 ret = btrfs_cont_expand(inode, i_size_read(inode),
1334 alloc_start);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001335 if (ret)
1336 goto out;
1337 }
1338
1339 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1340 if (ret)
1341 goto out;
1342
1343 locked_end = alloc_end - 1;
1344 while (1) {
1345 struct btrfs_ordered_extent *ordered;
1346
1347 /* the extent lock is ordered inside the running
1348 * transaction
1349 */
1350 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1351 locked_end, 0, &cached_state, GFP_NOFS);
1352 ordered = btrfs_lookup_first_ordered_extent(inode,
1353 alloc_end - 1);
1354 if (ordered &&
1355 ordered->file_offset + ordered->len > alloc_start &&
1356 ordered->file_offset < alloc_end) {
1357 btrfs_put_ordered_extent(ordered);
1358 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1359 alloc_start, locked_end,
1360 &cached_state, GFP_NOFS);
1361 /*
1362 * we can't wait on the range with the transaction
1363 * running or with the extent lock held
1364 */
1365 btrfs_wait_ordered_range(inode, alloc_start,
1366 alloc_end - alloc_start);
1367 } else {
1368 if (ordered)
1369 btrfs_put_ordered_extent(ordered);
1370 break;
1371 }
1372 }
1373
1374 cur_offset = alloc_start;
1375 while (1) {
1376 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1377 alloc_end - cur_offset, 0);
1378 BUG_ON(IS_ERR(em) || !em);
1379 last_byte = min(extent_map_end(em), alloc_end);
1380 last_byte = (last_byte + mask) & ~mask;
1381 if (em->block_start == EXTENT_MAP_HOLE ||
1382 (cur_offset >= inode->i_size &&
1383 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1384 ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1385 last_byte - cur_offset,
1386 1 << inode->i_blkbits,
1387 offset + len,
1388 &alloc_hint);
1389 if (ret < 0) {
1390 free_extent_map(em);
1391 break;
1392 }
1393 }
1394 free_extent_map(em);
1395
1396 cur_offset = last_byte;
1397 if (cur_offset >= alloc_end) {
1398 ret = 0;
1399 break;
1400 }
1401 }
1402 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1403 &cached_state, GFP_NOFS);
1404
1405 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1406out:
1407 mutex_unlock(&inode->i_mutex);
1408 return ret;
1409}
1410
Alexey Dobriyan828c0952009-10-01 15:43:56 -07001411const struct file_operations btrfs_file_operations = {
Chris Mason39279cc2007-06-12 06:35:45 -04001412 .llseek = generic_file_llseek,
1413 .read = do_sync_read,
Miao Xie4a0010712010-06-07 03:38:51 +00001414 .write = do_sync_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001415 .aio_read = generic_file_aio_read,
Chris Masone9906a92007-12-14 12:56:58 -05001416 .splice_read = generic_file_splice_read,
Josef Bacik11c65dc2010-05-23 11:07:21 -04001417 .aio_write = btrfs_file_aio_write,
Chris Mason9ebefb182007-06-15 13:50:00 -04001418 .mmap = btrfs_file_mmap,
Chris Mason39279cc2007-06-12 06:35:45 -04001419 .open = generic_file_open,
Mingminge1b81e62008-05-27 10:55:43 -04001420 .release = btrfs_release_file,
Chris Mason39279cc2007-06-12 06:35:45 -04001421 .fsync = btrfs_sync_file,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001422 .fallocate = btrfs_fallocate,
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001423 .unlocked_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001424#ifdef CONFIG_COMPAT
Christoph Hellwig34287aa2007-09-14 10:22:47 -04001425 .compat_ioctl = btrfs_ioctl,
Chris Mason39279cc2007-06-12 06:35:45 -04001426#endif
1427};