file.c 84.5 KB
Newer Older
C
Chris Mason 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License v2 as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

C
Chris Mason 已提交
19 20 21 22 23 24 25 26
#include <linux/fs.h>
#include <linux/pagemap.h>
#include <linux/highmem.h>
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
27
#include <linux/falloc.h>
C
Chris Mason 已提交
28 29 30
#include <linux/swap.h>
#include <linux/writeback.h>
#include <linux/compat.h>
31
#include <linux/slab.h>
32
#include <linux/btrfs.h>
33
#include <linux/uio.h>
C
Chris Mason 已提交
34 35 36 37 38
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "print-tree.h"
39 40
#include "tree-log.h"
#include "locking.h"
J
Josef Bacik 已提交
41
#include "volumes.h"
J
Josef Bacik 已提交
42
#include "qgroup.h"
43
#include "compression.h"
C
Chris Mason 已提交
44

45
static struct kmem_cache *btrfs_inode_defrag_cachep;
C
Chris Mason 已提交
46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70
/*
 * when auto defrag is enabled we
 * queue up these defrag structs to remember which
 * inodes need defragging passes
 */
struct inode_defrag {
	struct rb_node rb_node;
	/* objectid */
	u64 ino;
	/*
	 * transid where the defrag was added, we search for
	 * extents newer than this
	 */
	u64 transid;

	/* root objectid */
	u64 root;

	/* last offset we were able to defrag */
	u64 last_offset;

	/* if we've wrapped around back to zero once already */
	int cycled;
};

71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
static int __compare_inode_defrag(struct inode_defrag *defrag1,
				  struct inode_defrag *defrag2)
{
	if (defrag1->root > defrag2->root)
		return 1;
	else if (defrag1->root < defrag2->root)
		return -1;
	else if (defrag1->ino > defrag2->ino)
		return 1;
	else if (defrag1->ino < defrag2->ino)
		return -1;
	else
		return 0;
}

C
Chris Mason 已提交
86 87 88 89 90 91 92 93 94
/* pop a record for an inode into the defrag tree.  The lock
 * must be held already
 *
 * If you're inserting a record for an older transid than an
 * existing record, the transid already in the tree is lowered
 *
 * If an existing record is found the defrag item you
 * pass in is freed
 */
95
static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
C
Chris Mason 已提交
96 97
				    struct inode_defrag *defrag)
{
98
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
C
Chris Mason 已提交
99 100 101
	struct inode_defrag *entry;
	struct rb_node **p;
	struct rb_node *parent = NULL;
102
	int ret;
C
Chris Mason 已提交
103

104
	p = &fs_info->defrag_inodes.rb_node;
C
Chris Mason 已提交
105 106 107 108
	while (*p) {
		parent = *p;
		entry = rb_entry(parent, struct inode_defrag, rb_node);

109 110
		ret = __compare_inode_defrag(defrag, entry);
		if (ret < 0)
C
Chris Mason 已提交
111
			p = &parent->rb_left;
112
		else if (ret > 0)
C
Chris Mason 已提交
113 114 115 116 117 118 119 120 121 122
			p = &parent->rb_right;
		else {
			/* if we're reinserting an entry for
			 * an old defrag run, make sure to
			 * lower the transid of our existing record
			 */
			if (defrag->transid < entry->transid)
				entry->transid = defrag->transid;
			if (defrag->last_offset > entry->last_offset)
				entry->last_offset = defrag->last_offset;
123
			return -EEXIST;
C
Chris Mason 已提交
124 125
		}
	}
126
	set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
C
Chris Mason 已提交
127
	rb_link_node(&defrag->rb_node, parent, p);
128
	rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
129 130
	return 0;
}
C
Chris Mason 已提交
131

132
static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
133
{
134
	if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
135 136
		return 0;

137
	if (btrfs_fs_closing(fs_info))
138
		return 0;
C
Chris Mason 已提交
139

140
	return 1;
C
Chris Mason 已提交
141 142 143 144 145 146 147
}

/*
 * insert a defrag record for this inode if auto defrag is
 * enabled
 */
int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
148
			   struct btrfs_inode *inode)
C
Chris Mason 已提交
149
{
150 151
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
	struct btrfs_root *root = inode->root;
C
Chris Mason 已提交
152 153
	struct inode_defrag *defrag;
	u64 transid;
154
	int ret;
C
Chris Mason 已提交
155

156
	if (!__need_auto_defrag(fs_info))
C
Chris Mason 已提交
157 158
		return 0;

159
	if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
C
Chris Mason 已提交
160 161 162 163 164
		return 0;

	if (trans)
		transid = trans->transid;
	else
165
		transid = inode->root->last_trans;
C
Chris Mason 已提交
166

167
	defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
C
Chris Mason 已提交
168 169 170
	if (!defrag)
		return -ENOMEM;

171
	defrag->ino = btrfs_ino(inode);
C
Chris Mason 已提交
172 173 174
	defrag->transid = transid;
	defrag->root = root->root_key.objectid;

175
	spin_lock(&fs_info->defrag_inodes_lock);
176
	if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
177 178 179 180 181 182 183 184 185
		/*
		 * If we set IN_DEFRAG flag and evict the inode from memory,
		 * and then re-read this inode, this new inode doesn't have
		 * IN_DEFRAG flag. At the case, we may find the existed defrag.
		 */
		ret = __btrfs_add_inode_defrag(inode, defrag);
		if (ret)
			kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	} else {
186
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
187
	}
188
	spin_unlock(&fs_info->defrag_inodes_lock);
189
	return 0;
C
Chris Mason 已提交
190 191 192
}

/*
193 194 195
 * Requeue the defrag object. If there is a defrag object that points to
 * the same inode in the tree, we will merge them together (by
 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
C
Chris Mason 已提交
196
 */
197
static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
198
				       struct inode_defrag *defrag)
199
{
200
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
201 202
	int ret;

203
	if (!__need_auto_defrag(fs_info))
204 205 206 207 208 209
		goto out;

	/*
	 * Here we don't check the IN_DEFRAG flag, because we need merge
	 * them together.
	 */
210
	spin_lock(&fs_info->defrag_inodes_lock);
211
	ret = __btrfs_add_inode_defrag(inode, defrag);
212
	spin_unlock(&fs_info->defrag_inodes_lock);
213 214 215 216 217 218 219
	if (ret)
		goto out;
	return;
out:
	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
}

C
Chris Mason 已提交
220
/*
221 222
 * pick the defragable inode that we want, if it doesn't exist, we will get
 * the next one.
C
Chris Mason 已提交
223
 */
224 225
static struct inode_defrag *
btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
C
Chris Mason 已提交
226 227
{
	struct inode_defrag *entry = NULL;
228
	struct inode_defrag tmp;
C
Chris Mason 已提交
229 230
	struct rb_node *p;
	struct rb_node *parent = NULL;
231 232 233 234
	int ret;

	tmp.ino = ino;
	tmp.root = root;
C
Chris Mason 已提交
235

236 237
	spin_lock(&fs_info->defrag_inodes_lock);
	p = fs_info->defrag_inodes.rb_node;
C
Chris Mason 已提交
238 239 240 241
	while (p) {
		parent = p;
		entry = rb_entry(parent, struct inode_defrag, rb_node);

242 243
		ret = __compare_inode_defrag(&tmp, entry);
		if (ret < 0)
C
Chris Mason 已提交
244
			p = parent->rb_left;
245
		else if (ret > 0)
C
Chris Mason 已提交
246 247
			p = parent->rb_right;
		else
248
			goto out;
C
Chris Mason 已提交
249 250
	}

251 252 253
	if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
		parent = rb_next(parent);
		if (parent)
C
Chris Mason 已提交
254
			entry = rb_entry(parent, struct inode_defrag, rb_node);
255 256
		else
			entry = NULL;
C
Chris Mason 已提交
257
	}
258 259 260 261 262
out:
	if (entry)
		rb_erase(parent, &fs_info->defrag_inodes);
	spin_unlock(&fs_info->defrag_inodes_lock);
	return entry;
C
Chris Mason 已提交
263 264
}

265
void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
C
Chris Mason 已提交
266 267
{
	struct inode_defrag *defrag;
268 269 270 271 272 273 274 275 276
	struct rb_node *node;

	spin_lock(&fs_info->defrag_inodes_lock);
	node = rb_first(&fs_info->defrag_inodes);
	while (node) {
		rb_erase(node, &fs_info->defrag_inodes);
		defrag = rb_entry(node, struct inode_defrag, rb_node);
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);

277
		cond_resched_lock(&fs_info->defrag_inodes_lock);
278 279 280 281 282 283 284 285 286 287 288

		node = rb_first(&fs_info->defrag_inodes);
	}
	spin_unlock(&fs_info->defrag_inodes_lock);
}

#define BTRFS_DEFRAG_BATCH	1024

static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
				    struct inode_defrag *defrag)
{
C
Chris Mason 已提交
289 290 291 292 293
	struct btrfs_root *inode_root;
	struct inode *inode;
	struct btrfs_key key;
	struct btrfs_ioctl_defrag_range_args range;
	int num_defrag;
294 295
	int index;
	int ret;
C
Chris Mason 已提交
296

297 298
	/* get the inode */
	key.objectid = defrag->root;
299
	key.type = BTRFS_ROOT_ITEM_KEY;
300
	key.offset = (u64)-1;
301 302 303

	index = srcu_read_lock(&fs_info->subvol_srcu);

304 305
	inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
	if (IS_ERR(inode_root)) {
306 307 308
		ret = PTR_ERR(inode_root);
		goto cleanup;
	}
309 310

	key.objectid = defrag->ino;
311
	key.type = BTRFS_INODE_ITEM_KEY;
312 313 314
	key.offset = 0;
	inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
	if (IS_ERR(inode)) {
315 316
		ret = PTR_ERR(inode);
		goto cleanup;
317
	}
318
	srcu_read_unlock(&fs_info->subvol_srcu, index);
319 320 321

	/* do a chunk of defrag */
	clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
C
Chris Mason 已提交
322 323
	memset(&range, 0, sizeof(range));
	range.len = (u64)-1;
324
	range.start = defrag->last_offset;
M
Miao Xie 已提交
325 326

	sb_start_write(fs_info->sb);
327 328
	num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
				       BTRFS_DEFRAG_BATCH);
M
Miao Xie 已提交
329
	sb_end_write(fs_info->sb);
330 331 332 333 334 335 336
	/*
	 * if we filled the whole defrag batch, there
	 * must be more work to do.  Queue this defrag
	 * again
	 */
	if (num_defrag == BTRFS_DEFRAG_BATCH) {
		defrag->last_offset = range.start;
337
		btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
338 339 340 341 342 343 344 345
	} else if (defrag->last_offset && !defrag->cycled) {
		/*
		 * we didn't fill our defrag batch, but
		 * we didn't start at zero.  Make sure we loop
		 * around to the start of the file.
		 */
		defrag->last_offset = 0;
		defrag->cycled = 1;
346
		btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
347 348 349 350 351 352
	} else {
		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	}

	iput(inode);
	return 0;
353 354 355 356
cleanup:
	srcu_read_unlock(&fs_info->subvol_srcu, index);
	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
	return ret;
357 358 359 360 361 362 363 364 365 366 367
}

/*
 * run through the list of inodes in the FS that need
 * defragging
 */
int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
{
	struct inode_defrag *defrag;
	u64 first_ino = 0;
	u64 root_objectid = 0;
C
Chris Mason 已提交
368 369

	atomic_inc(&fs_info->defrag_running);
370
	while (1) {
M
Miao Xie 已提交
371 372 373 374 375
		/* Pause the auto defragger. */
		if (test_bit(BTRFS_FS_STATE_REMOUNTING,
			     &fs_info->fs_state))
			break;

376
		if (!__need_auto_defrag(fs_info))
377
			break;
C
Chris Mason 已提交
378 379

		/* find an inode to defrag */
380 381
		defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
						 first_ino);
C
Chris Mason 已提交
382
		if (!defrag) {
383
			if (root_objectid || first_ino) {
384
				root_objectid = 0;
C
Chris Mason 已提交
385 386 387 388 389 390 391 392
				first_ino = 0;
				continue;
			} else {
				break;
			}
		}

		first_ino = defrag->ino + 1;
393
		root_objectid = defrag->root;
C
Chris Mason 已提交
394

395
		__btrfs_run_defrag_inode(fs_info, defrag);
C
Chris Mason 已提交
396 397 398 399 400 401 402 403 404 405
	}
	atomic_dec(&fs_info->defrag_running);

	/*
	 * during unmount, we use the transaction_wait queue to
	 * wait for the defragger to stop
	 */
	wake_up(&fs_info->transaction_wait);
	return 0;
}
C
Chris Mason 已提交
406

C
Chris Mason 已提交
407 408 409
/* simple helper to fault in pages and copy.  This should go away
 * and be replaced with calls into generic code.
 */
410
static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
411
					 struct page **prepared_pages,
412
					 struct iov_iter *i)
C
Chris Mason 已提交
413
{
414
	size_t copied = 0;
J
Josef Bacik 已提交
415
	size_t total_copied = 0;
416
	int pg = 0;
417
	int offset = pos & (PAGE_SIZE - 1);
C
Chris Mason 已提交
418

419
	while (write_bytes > 0) {
C
Chris Mason 已提交
420
		size_t count = min_t(size_t,
421
				     PAGE_SIZE - offset, write_bytes);
422
		struct page *page = prepared_pages[pg];
423 424 425 426
		/*
		 * Copy data from userspace to the current page
		 */
		copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
427

C
Chris Mason 已提交
428 429
		/* Flush processor's dcache for this page */
		flush_dcache_page(page);
430 431 432 433 434 435 436 437 438 439 440 441 442

		/*
		 * if we get a partial write, we can end up with
		 * partially up to date pages.  These add
		 * a lot of complexity, so make sure they don't
		 * happen by forcing this copy to be retried.
		 *
		 * The rest of the btrfs_file_write code will fall
		 * back to page at a time copies after we return 0.
		 */
		if (!PageUptodate(page) && copied < count)
			copied = 0;

443 444
		iov_iter_advance(i, copied);
		write_bytes -= copied;
445
		total_copied += copied;
C
Chris Mason 已提交
446

A
Al Viro 已提交
447
		/* Return to btrfs_file_write_iter to fault page */
J
Josef Bacik 已提交
448
		if (unlikely(copied == 0))
449
			break;
450

451
		if (copied < PAGE_SIZE - offset) {
452 453 454 455 456
			offset += copied;
		} else {
			pg++;
			offset = 0;
		}
C
Chris Mason 已提交
457
	}
458
	return total_copied;
C
Chris Mason 已提交
459 460
}

C
Chris Mason 已提交
461 462 463
/*
 * unlocks pages after btrfs_file_write is done with them
 */
464
static void btrfs_drop_pages(struct page **pages, size_t num_pages)
C
Chris Mason 已提交
465 466 467
{
	size_t i;
	for (i = 0; i < num_pages; i++) {
C
Chris Mason 已提交
468 469
		/* page checked is some magic around finding pages that
		 * have been modified without going through btrfs_set_page_dirty
470 471 472
		 * clear it here. There should be no need to mark the pages
		 * accessed as prepare_pages should have marked them accessed
		 * in prepare_pages via find_or_create_page()
C
Chris Mason 已提交
473
		 */
C
Chris Mason 已提交
474
		ClearPageChecked(pages[i]);
C
Chris Mason 已提交
475
		unlock_page(pages[i]);
476
		put_page(pages[i]);
C
Chris Mason 已提交
477 478 479
	}
}

480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
					 const u64 start,
					 const u64 len,
					 struct extent_state **cached_state)
{
	u64 search_start = start;
	const u64 end = start + len - 1;

	while (search_start < end) {
		const u64 search_len = end - search_start + 1;
		struct extent_map *em;
		u64 em_len;
		int ret = 0;

		em = btrfs_get_extent(inode, NULL, 0, search_start,
				      search_len, 0);
		if (IS_ERR(em))
			return PTR_ERR(em);

		if (em->block_start != EXTENT_MAP_HOLE)
			goto next;

		em_len = em->len;
		if (em->start < search_start)
			em_len -= search_start - em->start;
		if (em_len > search_len)
			em_len = search_len;

		ret = set_extent_bit(&inode->io_tree, search_start,
				     search_start + em_len - 1,
				     EXTENT_DELALLOC_NEW,
				     NULL, cached_state, GFP_NOFS);
next:
		search_start = extent_map_end(em);
		free_extent_map(em);
		if (ret)
			return ret;
	}
	return 0;
}

C
Chris Mason 已提交
521 522 523 524 525 526 527 528
/*
 * after copy_from_user, pages need to be dirtied and we need to make
 * sure holes are created between the current EOF and the start of
 * any next extents (if required).
 *
 * this also makes the decision about creating an inline extent vs
 * doing real data extents, marking pages dirty and delalloc as required.
 */
529 530 531
int btrfs_dirty_pages(struct inode *inode, struct page **pages,
		      size_t num_pages, loff_t pos, size_t write_bytes,
		      struct extent_state **cached)
C
Chris Mason 已提交
532
{
533
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
C
Chris Mason 已提交
534
	int err = 0;
535
	int i;
536
	u64 num_bytes;
537 538 539 540
	u64 start_pos;
	u64 end_of_last_block;
	u64 end_pos = pos + write_bytes;
	loff_t isize = i_size_read(inode);
541
	unsigned int extra_bits = 0;
C
Chris Mason 已提交
542

543
	start_pos = pos & ~((u64) fs_info->sectorsize - 1);
544
	num_bytes = round_up(write_bytes + pos - start_pos,
545
			     fs_info->sectorsize);
C
Chris Mason 已提交
546

547
	end_of_last_block = start_pos + num_bytes - 1;
548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566

	if (!btrfs_is_free_space_inode(BTRFS_I(inode))) {
		if (start_pos >= isize &&
		    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) {
			/*
			 * There can't be any extents following eof in this case
			 * so just set the delalloc new bit for the range
			 * directly.
			 */
			extra_bits |= EXTENT_DELALLOC_NEW;
		} else {
			err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode),
							    start_pos,
							    num_bytes, cached);
			if (err)
				return err;
		}
	}

567
	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
568
					extra_bits, cached, 0);
J
Josef Bacik 已提交
569 570
	if (err)
		return err;
J
Josef Bacik 已提交
571

C
Chris Mason 已提交
572 573 574 575 576
	for (i = 0; i < num_pages; i++) {
		struct page *p = pages[i];
		SetPageUptodate(p);
		ClearPageChecked(p);
		set_page_dirty(p);
577
	}
J
Josef Bacik 已提交
578 579 580 581 582 583 584

	/*
	 * we've only changed i_size in ram, and we haven't updated
	 * the disk i_size.  There is no need to log the inode
	 * at this time.
	 */
	if (end_pos > isize)
585
		i_size_write(inode, end_pos);
586
	return 0;
C
Chris Mason 已提交
587 588
}

C
Chris Mason 已提交
589 590 591 592
/*
 * this drops all the extents in the cache that intersect the range
 * [start, end].  Existing extents are split as required.
 */
593
void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
594
			     int skip_pinned)
595 596
{
	struct extent_map *em;
597 598
	struct extent_map *split = NULL;
	struct extent_map *split2 = NULL;
599
	struct extent_map_tree *em_tree = &inode->extent_tree;
600
	u64 len = end - start + 1;
J
Josef Bacik 已提交
601
	u64 gen;
602 603
	int ret;
	int testend = 1;
604
	unsigned long flags;
C
Chris Mason 已提交
605
	int compressed = 0;
J
Josef Bacik 已提交
606
	bool modified;
607

608
	WARN_ON(end < start);
609
	if (end == (u64)-1) {
610
		len = (u64)-1;
611 612
		testend = 0;
	}
C
Chris Mason 已提交
613
	while (1) {
614 615
		int no_splits = 0;

J
Josef Bacik 已提交
616
		modified = false;
617
		if (!split)
618
			split = alloc_extent_map();
619
		if (!split2)
620
			split2 = alloc_extent_map();
621 622
		if (!split || !split2)
			no_splits = 1;
623

624
		write_lock(&em_tree->lock);
625
		em = lookup_extent_mapping(em_tree, start, len);
626
		if (!em) {
627
			write_unlock(&em_tree->lock);
628
			break;
629
		}
630
		flags = em->flags;
J
Josef Bacik 已提交
631
		gen = em->generation;
632
		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
633
			if (testend && em->start + em->len >= start + len) {
634
				free_extent_map(em);
C
Chris Mason 已提交
635
				write_unlock(&em_tree->lock);
636 637
				break;
			}
638 639
			start = em->start + em->len;
			if (testend)
640 641
				len = start + len - (em->start + em->len);
			free_extent_map(em);
C
Chris Mason 已提交
642
			write_unlock(&em_tree->lock);
643 644
			continue;
		}
C
Chris Mason 已提交
645
		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
646
		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
L
Liu Bo 已提交
647
		clear_bit(EXTENT_FLAG_LOGGING, &flags);
J
Josef Bacik 已提交
648
		modified = !list_empty(&em->list);
649 650
		if (no_splits)
			goto next;
651

652
		if (em->start < start) {
653 654
			split->start = em->start;
			split->len = start - em->start;
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674

			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
				split->orig_start = em->orig_start;
				split->block_start = em->block_start;

				if (compressed)
					split->block_len = em->block_len;
				else
					split->block_len = split->len;
				split->orig_block_len = max(split->block_len,
						em->orig_block_len);
				split->ram_bytes = em->ram_bytes;
			} else {
				split->orig_start = split->start;
				split->block_len = 0;
				split->block_start = em->block_start;
				split->orig_block_len = 0;
				split->ram_bytes = split->len;
			}

J
Josef Bacik 已提交
675
			split->generation = gen;
676
			split->bdev = em->bdev;
677
			split->flags = flags;
678
			split->compress_type = em->compress_type;
679
			replace_extent_mapping(em_tree, em, split, modified);
680 681 682 683
			free_extent_map(split);
			split = split2;
			split2 = NULL;
		}
684
		if (testend && em->start + em->len > start + len) {
685 686 687 688 689
			u64 diff = start + len - em->start;

			split->start = start + len;
			split->len = em->start + em->len - (start + len);
			split->bdev = em->bdev;
690
			split->flags = flags;
691
			split->compress_type = em->compress_type;
J
Josef Bacik 已提交
692
			split->generation = gen;
693 694 695

			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
				split->orig_block_len = max(em->block_len,
696
						    em->orig_block_len);
697

698 699 700 701 702 703 704 705 706 707 708
				split->ram_bytes = em->ram_bytes;
				if (compressed) {
					split->block_len = em->block_len;
					split->block_start = em->block_start;
					split->orig_start = em->orig_start;
				} else {
					split->block_len = split->len;
					split->block_start = em->block_start
						+ diff;
					split->orig_start = em->orig_start;
				}
C
Chris Mason 已提交
709
			} else {
710 711 712 713 714
				split->ram_bytes = split->len;
				split->orig_start = split->start;
				split->block_len = 0;
				split->block_start = em->block_start;
				split->orig_block_len = 0;
C
Chris Mason 已提交
715
			}
716

717 718 719 720 721 722 723 724
			if (extent_map_in_tree(em)) {
				replace_extent_mapping(em_tree, em, split,
						       modified);
			} else {
				ret = add_extent_mapping(em_tree, split,
							 modified);
				ASSERT(ret == 0); /* Logic error */
			}
725 726 727
			free_extent_map(split);
			split = NULL;
		}
728
next:
729 730
		if (extent_map_in_tree(em))
			remove_extent_mapping(em_tree, em);
731
		write_unlock(&em_tree->lock);
732

733 734 735 736 737
		/* once for us */
		free_extent_map(em);
		/* once for the tree*/
		free_extent_map(em);
	}
738 739 740 741
	if (split)
		free_extent_map(split);
	if (split2)
		free_extent_map(split2);
742 743
}

C
Chris Mason 已提交
744 745 746 747 748 749 750 751 752
/*
 * this is very complex, but the basic idea is to drop all extents
 * in the range start - end.  hint_block is filled in with a block number
 * that would be a good hint to the block allocator for this file.
 *
 * If an extent intersects the range but is not entirely inside the range
 * it is either truncated or split.  Anything entirely inside the range
 * is deleted from the tree.
 */
J
Josef Bacik 已提交
753 754 755
int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
			 struct btrfs_root *root, struct inode *inode,
			 struct btrfs_path *path, u64 start, u64 end,
756 757 758 759
			 u64 *drop_end, int drop_cache,
			 int replace_extent,
			 u32 extent_item_size,
			 int *key_inserted)
C
Chris Mason 已提交
760
{
761
	struct btrfs_fs_info *fs_info = root->fs_info;
762
	struct extent_buffer *leaf;
Y
Yan, Zheng 已提交
763
	struct btrfs_file_extent_item *fi;
764
	struct btrfs_key key;
Y
Yan, Zheng 已提交
765
	struct btrfs_key new_key;
766
	u64 ino = btrfs_ino(BTRFS_I(inode));
Y
Yan, Zheng 已提交
767 768 769 770 771
	u64 search_start = start;
	u64 disk_bytenr = 0;
	u64 num_bytes = 0;
	u64 extent_offset = 0;
	u64 extent_end = 0;
J
Josef Bacik 已提交
772
	u64 last_end = start;
Y
Yan, Zheng 已提交
773 774 775
	int del_nr = 0;
	int del_slot = 0;
	int extent_type;
C
Chris Mason 已提交
776
	int recow;
777
	int ret;
778
	int modify_tree = -1;
779
	int update_refs;
780
	int found = 0;
781
	int leafs_visited = 0;
C
Chris Mason 已提交
782

C
Chris Mason 已提交
783
	if (drop_cache)
784
		btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0);
785

786
	if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
787 788
		modify_tree = 0;

789
	update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
790
		       root == fs_info->tree_root);
C
Chris Mason 已提交
791
	while (1) {
C
Chris Mason 已提交
792
		recow = 0;
L
Li Zefan 已提交
793
		ret = btrfs_lookup_file_extent(trans, root, path, ino,
794
					       search_start, modify_tree);
C
Chris Mason 已提交
795
		if (ret < 0)
Y
Yan, Zheng 已提交
796 797 798 799
			break;
		if (ret > 0 && path->slots[0] > 0 && search_start == start) {
			leaf = path->nodes[0];
			btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
L
Li Zefan 已提交
800
			if (key.objectid == ino &&
Y
Yan, Zheng 已提交
801 802
			    key.type == BTRFS_EXTENT_DATA_KEY)
				path->slots[0]--;
C
Chris Mason 已提交
803
		}
Y
Yan, Zheng 已提交
804
		ret = 0;
805
		leafs_visited++;
806
next_slot:
807
		leaf = path->nodes[0];
Y
Yan, Zheng 已提交
808 809 810 811 812 813 814 815
		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
			BUG_ON(del_nr > 0);
			ret = btrfs_next_leaf(root, path);
			if (ret < 0)
				break;
			if (ret > 0) {
				ret = 0;
				break;
816
			}
817
			leafs_visited++;
Y
Yan, Zheng 已提交
818 819 820 821 822
			leaf = path->nodes[0];
			recow = 1;
		}

		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
823 824 825 826 827 828 829 830 831 832

		if (key.objectid > ino)
			break;
		if (WARN_ON_ONCE(key.objectid < ino) ||
		    key.type < BTRFS_EXTENT_DATA_KEY) {
			ASSERT(del_nr == 0);
			path->slots[0]++;
			goto next_slot;
		}
		if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
Y
Yan, Zheng 已提交
833 834 835 836 837 838 839 840 841 842 843 844 845 846 847
			break;

		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		extent_type = btrfs_file_extent_type(leaf, fi);

		if (extent_type == BTRFS_FILE_EXTENT_REG ||
		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
			num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
			extent_offset = btrfs_file_extent_offset(leaf, fi);
			extent_end = key.offset +
				btrfs_file_extent_num_bytes(leaf, fi);
		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
			extent_end = key.offset +
848 849
				btrfs_file_extent_inline_len(leaf,
						     path->slots[0], fi);
850
		} else {
851 852
			/* can't happen */
			BUG();
C
Chris Mason 已提交
853 854
		}

855 856 857 858 859 860 861 862 863
		/*
		 * Don't skip extent items representing 0 byte lengths. They
		 * used to be created (bug) if while punching holes we hit
		 * -ENOSPC condition. So if we find one here, just ensure we
		 * delete it, otherwise we would insert a new file extent item
		 * with the same key (offset) as that 0 bytes length file
		 * extent item in the call to setup_items_for_insert() later
		 * in this function.
		 */
J
Josef Bacik 已提交
864 865
		if (extent_end == key.offset && extent_end >= search_start) {
			last_end = extent_end;
866
			goto delete_extent_item;
J
Josef Bacik 已提交
867
		}
868

Y
Yan, Zheng 已提交
869 870
		if (extent_end <= search_start) {
			path->slots[0]++;
871
			goto next_slot;
C
Chris Mason 已提交
872 873
		}

874
		found = 1;
Y
Yan, Zheng 已提交
875
		search_start = max(key.offset, start);
876 877
		if (recow || !modify_tree) {
			modify_tree = -1;
878
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
879
			continue;
C
Chris Mason 已提交
880
		}
Y
Yan Zheng 已提交
881

Y
Yan, Zheng 已提交
882 883 884 885 886 887
		/*
		 *     | - range to drop - |
		 *  | -------- extent -------- |
		 */
		if (start > key.offset && end < extent_end) {
			BUG_ON(del_nr > 0);
888
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
889
				ret = -EOPNOTSUPP;
890 891
				break;
			}
Y
Yan, Zheng 已提交
892 893 894 895 896 897

			memcpy(&new_key, &key, sizeof(new_key));
			new_key.offset = start;
			ret = btrfs_duplicate_item(trans, root, path,
						   &new_key);
			if (ret == -EAGAIN) {
898
				btrfs_release_path(path);
Y
Yan, Zheng 已提交
899
				continue;
Y
Yan Zheng 已提交
900
			}
Y
Yan, Zheng 已提交
901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918
			if (ret < 0)
				break;

			leaf = path->nodes[0];
			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
					    struct btrfs_file_extent_item);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);

			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);

			extent_offset += start - key.offset;
			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - start);
			btrfs_mark_buffer_dirty(leaf);

J
Josef Bacik 已提交
919
			if (update_refs && disk_bytenr > 0) {
920
				ret = btrfs_inc_extent_ref(trans, root,
Y
Yan, Zheng 已提交
921 922 923
						disk_bytenr, num_bytes, 0,
						root->root_key.objectid,
						new_key.objectid,
924
						start - extent_offset);
925
				BUG_ON(ret); /* -ENOMEM */
926
			}
Y
Yan, Zheng 已提交
927
			key.offset = start;
Y
Yan Zheng 已提交
928
		}
J
Josef Bacik 已提交
929 930 931 932 933 934
		/*
		 * From here on out we will have actually dropped something, so
		 * last_end can be updated.
		 */
		last_end = extent_end;

Y
Yan, Zheng 已提交
935 936 937 938 939
		/*
		 *  | ---- range to drop ----- |
		 *      | -------- extent -------- |
		 */
		if (start <= key.offset && end < extent_end) {
940
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
941
				ret = -EOPNOTSUPP;
942 943
				break;
			}
Y
Yan Zheng 已提交
944

Y
Yan, Zheng 已提交
945 946
			memcpy(&new_key, &key, sizeof(new_key));
			new_key.offset = end;
947
			btrfs_set_item_key_safe(fs_info, path, &new_key);
Y
Yan Zheng 已提交
948

Y
Yan, Zheng 已提交
949 950 951 952 953
			extent_offset += end - key.offset;
			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - end);
			btrfs_mark_buffer_dirty(leaf);
954
			if (update_refs && disk_bytenr > 0)
Y
Yan, Zheng 已提交
955 956
				inode_sub_bytes(inode, end - key.offset);
			break;
C
Chris Mason 已提交
957
		}
958

Y
Yan, Zheng 已提交
959 960 961 962 963 964 965
		search_start = extent_end;
		/*
		 *       | ---- range to drop ----- |
		 *  | -------- extent -------- |
		 */
		if (start > key.offset && end >= extent_end) {
			BUG_ON(del_nr > 0);
966
			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
967
				ret = -EOPNOTSUPP;
968 969
				break;
			}
970

Y
Yan, Zheng 已提交
971 972 973
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);
			btrfs_mark_buffer_dirty(leaf);
974
			if (update_refs && disk_bytenr > 0)
Y
Yan, Zheng 已提交
975 976 977
				inode_sub_bytes(inode, extent_end - start);
			if (end == extent_end)
				break;
C
Chris Mason 已提交
978

Y
Yan, Zheng 已提交
979 980
			path->slots[0]++;
			goto next_slot;
Z
Zheng Yan 已提交
981 982
		}

Y
Yan, Zheng 已提交
983 984 985 986 987
		/*
		 *  | ---- range to drop ----- |
		 *    | ------ extent ------ |
		 */
		if (start <= key.offset && end >= extent_end) {
988
delete_extent_item:
Y
Yan, Zheng 已提交
989 990 991 992 993 994 995
			if (del_nr == 0) {
				del_slot = path->slots[0];
				del_nr = 1;
			} else {
				BUG_ON(del_slot + del_nr != path->slots[0]);
				del_nr++;
			}
Z
Zheng Yan 已提交
996

J
Josef Bacik 已提交
997 998
			if (update_refs &&
			    extent_type == BTRFS_FILE_EXTENT_INLINE) {
999
				inode_sub_bytes(inode,
Y
Yan, Zheng 已提交
1000 1001
						extent_end - key.offset);
				extent_end = ALIGN(extent_end,
1002
						   fs_info->sectorsize);
J
Josef Bacik 已提交
1003
			} else if (update_refs && disk_bytenr > 0) {
1004
				ret = btrfs_free_extent(trans, root,
Y
Yan, Zheng 已提交
1005 1006
						disk_bytenr, num_bytes, 0,
						root->root_key.objectid,
1007
						key.objectid, key.offset -
1008
						extent_offset);
1009
				BUG_ON(ret); /* -ENOMEM */
Y
Yan, Zheng 已提交
1010 1011
				inode_sub_bytes(inode,
						extent_end - key.offset);
Z
Zheng Yan 已提交
1012 1013
			}

Y
Yan, Zheng 已提交
1014 1015 1016 1017 1018 1019 1020 1021 1022 1023
			if (end == extent_end)
				break;

			if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
				path->slots[0]++;
				goto next_slot;
			}

			ret = btrfs_del_items(trans, root, path, del_slot,
					      del_nr);
1024
			if (ret) {
1025
				btrfs_abort_transaction(trans, ret);
J
Josef Bacik 已提交
1026
				break;
1027
			}
Y
Yan, Zheng 已提交
1028 1029 1030 1031

			del_nr = 0;
			del_slot = 0;

1032
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
1033
			continue;
C
Chris Mason 已提交
1034
		}
Y
Yan, Zheng 已提交
1035 1036

		BUG_ON(1);
C
Chris Mason 已提交
1037
	}
Y
Yan, Zheng 已提交
1038

1039
	if (!ret && del_nr > 0) {
1040 1041 1042 1043
		/*
		 * Set path->slots[0] to first slot, so that after the delete
		 * if items are move off from our leaf to its immediate left or
		 * right neighbor leafs, we end up with a correct and adjusted
1044
		 * path->slots[0] for our insertion (if replace_extent != 0).
1045 1046
		 */
		path->slots[0] = del_slot;
Y
Yan, Zheng 已提交
1047
		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1048
		if (ret)
1049
			btrfs_abort_transaction(trans, ret);
1050
	}
1051

1052 1053 1054 1055 1056 1057 1058 1059 1060
	leaf = path->nodes[0];
	/*
	 * If btrfs_del_items() was called, it might have deleted a leaf, in
	 * which case it unlocked our path, so check path->locks[0] matches a
	 * write lock.
	 */
	if (!ret && replace_extent && leafs_visited == 1 &&
	    (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
	     path->locks[0] == BTRFS_WRITE_LOCK) &&
1061
	    btrfs_leaf_free_space(fs_info, leaf) >=
1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
	    sizeof(struct btrfs_item) + extent_item_size) {

		key.objectid = ino;
		key.type = BTRFS_EXTENT_DATA_KEY;
		key.offset = start;
		if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
			struct btrfs_key slot_key;

			btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
			if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
				path->slots[0]++;
1073
		}
1074 1075 1076 1077 1078 1079
		setup_items_for_insert(root, path, &key,
				       &extent_item_size,
				       extent_item_size,
				       sizeof(struct btrfs_item) +
				       extent_item_size, 1);
		*key_inserted = 1;
Y
Yan Zheng 已提交
1080
	}
Y
Yan, Zheng 已提交
1081

1082 1083
	if (!replace_extent || !(*key_inserted))
		btrfs_release_path(path);
J
Josef Bacik 已提交
1084
	if (drop_end)
J
Josef Bacik 已提交
1085
		*drop_end = found ? min(end, last_end) : end;
J
Josef Bacik 已提交
1086 1087 1088 1089 1090
	return ret;
}

int btrfs_drop_extents(struct btrfs_trans_handle *trans,
		       struct btrfs_root *root, struct inode *inode, u64 start,
1091
		       u64 end, int drop_cache)
J
Josef Bacik 已提交
1092 1093 1094 1095 1096 1097 1098
{
	struct btrfs_path *path;
	int ret;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;
J
Josef Bacik 已提交
1099
	ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1100
				   drop_cache, 0, 0, NULL);
Y
Yan, Zheng 已提交
1101
	btrfs_free_path(path);
C
Chris Mason 已提交
1102 1103 1104
	return ret;
}

Y
Yan Zheng 已提交
1105
static int extent_mergeable(struct extent_buffer *leaf, int slot,
1106 1107
			    u64 objectid, u64 bytenr, u64 orig_offset,
			    u64 *start, u64 *end)
Y
Yan Zheng 已提交
1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
{
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;
	u64 extent_end;

	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
		return 0;

	btrfs_item_key_to_cpu(leaf, &key, slot);
	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
		return 0;

	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1123
	    btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
Y
Yan Zheng 已提交
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145
	    btrfs_file_extent_compression(leaf, fi) ||
	    btrfs_file_extent_encryption(leaf, fi) ||
	    btrfs_file_extent_other_encoding(leaf, fi))
		return 0;

	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
	if ((*start && *start != key.offset) || (*end && *end != extent_end))
		return 0;

	*start = key.offset;
	*end = extent_end;
	return 1;
}

/*
 * Mark extent in the range start - end as written.
 *
 * This changes extent type from 'pre-allocated' to 'regular'. If only
 * part of extent is marked as written, the extent will be split into
 * two or three.
 */
int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1146
			      struct btrfs_inode *inode, u64 start, u64 end)
Y
Yan Zheng 已提交
1147
{
1148 1149
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
	struct btrfs_root *root = inode->root;
Y
Yan Zheng 已提交
1150 1151 1152 1153
	struct extent_buffer *leaf;
	struct btrfs_path *path;
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;
Y
Yan, Zheng 已提交
1154
	struct btrfs_key new_key;
Y
Yan Zheng 已提交
1155 1156 1157
	u64 bytenr;
	u64 num_bytes;
	u64 extent_end;
1158
	u64 orig_offset;
Y
Yan Zheng 已提交
1159 1160
	u64 other_start;
	u64 other_end;
Y
Yan, Zheng 已提交
1161 1162 1163
	u64 split;
	int del_nr = 0;
	int del_slot = 0;
1164
	int recow;
Y
Yan Zheng 已提交
1165
	int ret;
1166
	u64 ino = btrfs_ino(inode);
Y
Yan Zheng 已提交
1167 1168

	path = btrfs_alloc_path();
1169 1170
	if (!path)
		return -ENOMEM;
Y
Yan Zheng 已提交
1171
again:
1172
	recow = 0;
Y
Yan, Zheng 已提交
1173
	split = start;
L
Li Zefan 已提交
1174
	key.objectid = ino;
Y
Yan Zheng 已提交
1175
	key.type = BTRFS_EXTENT_DATA_KEY;
Y
Yan, Zheng 已提交
1176
	key.offset = split;
Y
Yan Zheng 已提交
1177 1178

	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1179 1180
	if (ret < 0)
		goto out;
Y
Yan Zheng 已提交
1181 1182 1183 1184 1185
	if (ret > 0 && path->slots[0] > 0)
		path->slots[0]--;

	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1186 1187 1188 1189 1190 1191
	if (key.objectid != ino ||
	    key.type != BTRFS_EXTENT_DATA_KEY) {
		ret = -EINVAL;
		btrfs_abort_transaction(trans, ret);
		goto out;
	}
Y
Yan Zheng 已提交
1192 1193
	fi = btrfs_item_ptr(leaf, path->slots[0],
			    struct btrfs_file_extent_item);
1194 1195 1196 1197 1198
	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
		ret = -EINVAL;
		btrfs_abort_transaction(trans, ret);
		goto out;
	}
Y
Yan Zheng 已提交
1199
	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1200 1201 1202 1203 1204
	if (key.offset > start || extent_end < end) {
		ret = -EINVAL;
		btrfs_abort_transaction(trans, ret);
		goto out;
	}
Y
Yan Zheng 已提交
1205 1206 1207

	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1208
	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1209 1210 1211 1212 1213 1214
	memcpy(&new_key, &key, sizeof(new_key));

	if (start == key.offset && end < extent_end) {
		other_start = 0;
		other_end = start;
		if (extent_mergeable(leaf, path->slots[0] - 1,
L
Li Zefan 已提交
1215
				     ino, bytenr, orig_offset,
1216 1217
				     &other_start, &other_end)) {
			new_key.offset = end;
1218
			btrfs_set_item_key_safe(fs_info, path, &new_key);
1219 1220
			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
1221 1222
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1223 1224 1225 1226 1227 1228
			btrfs_set_file_extent_num_bytes(leaf, fi,
							extent_end - end);
			btrfs_set_file_extent_offset(leaf, fi,
						     end - orig_offset);
			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
					    struct btrfs_file_extent_item);
1229 1230
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241
			btrfs_set_file_extent_num_bytes(leaf, fi,
							end - other_start);
			btrfs_mark_buffer_dirty(leaf);
			goto out;
		}
	}

	if (start > key.offset && end == extent_end) {
		other_start = end;
		other_end = 0;
		if (extent_mergeable(leaf, path->slots[0] + 1,
L
Li Zefan 已提交
1242
				     ino, bytenr, orig_offset,
1243 1244 1245 1246 1247
				     &other_start, &other_end)) {
			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
			btrfs_set_file_extent_num_bytes(leaf, fi,
							start - key.offset);
1248 1249
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1250 1251
			path->slots[0]++;
			new_key.offset = start;
1252
			btrfs_set_item_key_safe(fs_info, path, &new_key);
1253 1254 1255

			fi = btrfs_item_ptr(leaf, path->slots[0],
					    struct btrfs_file_extent_item);
1256 1257
			btrfs_set_file_extent_generation(leaf, fi,
							 trans->transid);
1258 1259 1260 1261 1262 1263 1264 1265
			btrfs_set_file_extent_num_bytes(leaf, fi,
							other_end - start);
			btrfs_set_file_extent_offset(leaf, fi,
						     start - orig_offset);
			btrfs_mark_buffer_dirty(leaf);
			goto out;
		}
	}
Y
Yan Zheng 已提交
1266

Y
Yan, Zheng 已提交
1267 1268 1269 1270 1271 1272 1273
	while (start > key.offset || end < extent_end) {
		if (key.offset == start)
			split = end;

		new_key.offset = split;
		ret = btrfs_duplicate_item(trans, root, path, &new_key);
		if (ret == -EAGAIN) {
1274
			btrfs_release_path(path);
Y
Yan, Zheng 已提交
1275
			goto again;
Y
Yan Zheng 已提交
1276
		}
1277
		if (ret < 0) {
1278
			btrfs_abort_transaction(trans, ret);
1279 1280
			goto out;
		}
Y
Yan Zheng 已提交
1281

Y
Yan, Zheng 已提交
1282 1283
		leaf = path->nodes[0];
		fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
Y
Yan Zheng 已提交
1284
				    struct btrfs_file_extent_item);
1285
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan Zheng 已提交
1286
		btrfs_set_file_extent_num_bytes(leaf, fi,
Y
Yan, Zheng 已提交
1287 1288 1289 1290 1291
						split - key.offset);

		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);

1292
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan, Zheng 已提交
1293 1294 1295
		btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
		btrfs_set_file_extent_num_bytes(leaf, fi,
						extent_end - split);
Y
Yan Zheng 已提交
1296 1297
		btrfs_mark_buffer_dirty(leaf);

1298
		ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes,
1299
					   0, root->root_key.objectid,
1300
					   ino, orig_offset);
1301 1302 1303 1304
		if (ret) {
			btrfs_abort_transaction(trans, ret);
			goto out;
		}
Y
Yan Zheng 已提交
1305

Y
Yan, Zheng 已提交
1306 1307 1308
		if (split == start) {
			key.offset = start;
		} else {
1309 1310 1311 1312 1313
			if (start != key.offset) {
				ret = -EINVAL;
				btrfs_abort_transaction(trans, ret);
				goto out;
			}
Y
Yan Zheng 已提交
1314
			path->slots[0]--;
Y
Yan, Zheng 已提交
1315
			extent_end = end;
Y
Yan Zheng 已提交
1316
		}
1317
		recow = 1;
Y
Yan Zheng 已提交
1318 1319
	}

Y
Yan, Zheng 已提交
1320 1321
	other_start = end;
	other_end = 0;
1322
	if (extent_mergeable(leaf, path->slots[0] + 1,
L
Li Zefan 已提交
1323
			     ino, bytenr, orig_offset,
1324 1325
			     &other_start, &other_end)) {
		if (recow) {
1326
			btrfs_release_path(path);
1327 1328
			goto again;
		}
Y
Yan, Zheng 已提交
1329 1330 1331
		extent_end = other_end;
		del_slot = path->slots[0] + 1;
		del_nr++;
1332
		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
Y
Yan, Zheng 已提交
1333
					0, root->root_key.objectid,
1334
					ino, orig_offset);
1335 1336 1337 1338
		if (ret) {
			btrfs_abort_transaction(trans, ret);
			goto out;
		}
Y
Yan Zheng 已提交
1339
	}
Y
Yan, Zheng 已提交
1340 1341
	other_start = 0;
	other_end = start;
1342
	if (extent_mergeable(leaf, path->slots[0] - 1,
L
Li Zefan 已提交
1343
			     ino, bytenr, orig_offset,
1344 1345
			     &other_start, &other_end)) {
		if (recow) {
1346
			btrfs_release_path(path);
1347 1348
			goto again;
		}
Y
Yan, Zheng 已提交
1349 1350 1351
		key.offset = other_start;
		del_slot = path->slots[0];
		del_nr++;
1352
		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
Y
Yan, Zheng 已提交
1353
					0, root->root_key.objectid,
1354
					ino, orig_offset);
1355 1356 1357 1358
		if (ret) {
			btrfs_abort_transaction(trans, ret);
			goto out;
		}
Y
Yan, Zheng 已提交
1359 1360
	}
	if (del_nr == 0) {
1361 1362
		fi = btrfs_item_ptr(leaf, path->slots[0],
			   struct btrfs_file_extent_item);
Y
Yan, Zheng 已提交
1363 1364
		btrfs_set_file_extent_type(leaf, fi,
					   BTRFS_FILE_EXTENT_REG);
1365
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
Y
Yan, Zheng 已提交
1366
		btrfs_mark_buffer_dirty(leaf);
1367
	} else {
1368 1369
		fi = btrfs_item_ptr(leaf, del_slot - 1,
			   struct btrfs_file_extent_item);
1370 1371
		btrfs_set_file_extent_type(leaf, fi,
					   BTRFS_FILE_EXTENT_REG);
1372
		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1373 1374 1375
		btrfs_set_file_extent_num_bytes(leaf, fi,
						extent_end - key.offset);
		btrfs_mark_buffer_dirty(leaf);
Y
Yan, Zheng 已提交
1376

1377
		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1378
		if (ret < 0) {
1379
			btrfs_abort_transaction(trans, ret);
1380 1381
			goto out;
		}
1382
	}
Y
Yan, Zheng 已提交
1383
out:
Y
Yan Zheng 已提交
1384 1385 1386 1387
	btrfs_free_path(path);
	return 0;
}

1388 1389 1390 1391
/*
 * on error we return an unlocked page and the error value
 * on success we return a locked page and 0
 */
1392 1393
static int prepare_uptodate_page(struct inode *inode,
				 struct page *page, u64 pos,
1394
				 bool force_uptodate)
1395 1396 1397
{
	int ret = 0;

1398
	if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1399
	    !PageUptodate(page)) {
1400 1401 1402 1403 1404 1405 1406 1407
		ret = btrfs_readpage(NULL, page);
		if (ret)
			return ret;
		lock_page(page);
		if (!PageUptodate(page)) {
			unlock_page(page);
			return -EIO;
		}
1408 1409 1410 1411
		if (page->mapping != inode->i_mapping) {
			unlock_page(page);
			return -EAGAIN;
		}
1412 1413 1414 1415
	}
	return 0;
}

C
Chris Mason 已提交
1416
/*
1417
 * this just gets pages into the page cache and locks them down.
C
Chris Mason 已提交
1418
 */
1419 1420 1421
static noinline int prepare_pages(struct inode *inode, struct page **pages,
				  size_t num_pages, loff_t pos,
				  size_t write_bytes, bool force_uptodate)
C
Chris Mason 已提交
1422 1423
{
	int i;
1424
	unsigned long index = pos >> PAGE_SHIFT;
1425
	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
1426
	int err = 0;
1427
	int faili;
1428

C
Chris Mason 已提交
1429
	for (i = 0; i < num_pages; i++) {
1430
again:
1431
		pages[i] = find_or_create_page(inode->i_mapping, index + i,
1432
					       mask | __GFP_WRITE);
C
Chris Mason 已提交
1433
		if (!pages[i]) {
1434 1435 1436 1437 1438 1439
			faili = i - 1;
			err = -ENOMEM;
			goto fail;
		}

		if (i == 0)
1440
			err = prepare_uptodate_page(inode, pages[i], pos,
1441
						    force_uptodate);
1442 1443
		if (!err && i == num_pages - 1)
			err = prepare_uptodate_page(inode, pages[i],
1444
						    pos + write_bytes, false);
1445
		if (err) {
1446
			put_page(pages[i]);
1447 1448 1449 1450
			if (err == -EAGAIN) {
				err = 0;
				goto again;
			}
1451 1452
			faili = i - 1;
			goto fail;
C
Chris Mason 已提交
1453
		}
C
Chris Mason 已提交
1454
		wait_on_page_writeback(pages[i]);
C
Chris Mason 已提交
1455
	}
1456 1457 1458 1459 1460

	return 0;
fail:
	while (faili >= 0) {
		unlock_page(pages[faili]);
1461
		put_page(pages[faili]);
1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478
		faili--;
	}
	return err;

}

/*
 * This function locks the extent and properly waits for data=ordered extents
 * to finish before allowing the pages to be modified if need.
 *
 * The return value:
 * 1 - the extent is locked
 * 0 - the extent is not locked, and everything is OK
 * -EAGAIN - need re-prepare the pages
 * the other < 0 number - Something wrong happens
 */
static noinline int
1479
lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1480
				size_t num_pages, loff_t pos,
1481
				size_t write_bytes,
1482 1483 1484
				u64 *lockstart, u64 *lockend,
				struct extent_state **cached_state)
{
1485
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1486 1487 1488 1489 1490
	u64 start_pos;
	u64 last_pos;
	int i;
	int ret = 0;

1491
	start_pos = round_down(pos, fs_info->sectorsize);
1492
	last_pos = start_pos
1493
		+ round_up(pos + write_bytes - start_pos,
1494
			   fs_info->sectorsize) - 1;
1495

1496
	if (start_pos < inode->vfs_inode.i_size) {
1497
		struct btrfs_ordered_extent *ordered;
1498

1499 1500
		lock_extent_bits(&inode->io_tree, start_pos, last_pos,
				cached_state);
1501 1502
		ordered = btrfs_lookup_ordered_range(inode, start_pos,
						     last_pos - start_pos + 1);
1503 1504
		if (ordered &&
		    ordered->file_offset + ordered->len > start_pos &&
1505
		    ordered->file_offset <= last_pos) {
1506 1507
			unlock_extent_cached(&inode->io_tree, start_pos,
					last_pos, cached_state, GFP_NOFS);
1508 1509
			for (i = 0; i < num_pages; i++) {
				unlock_page(pages[i]);
1510
				put_page(pages[i]);
1511
			}
1512 1513
			btrfs_start_ordered_extent(&inode->vfs_inode,
					ordered, 1);
1514 1515
			btrfs_put_ordered_extent(ordered);
			return -EAGAIN;
1516 1517 1518
		}
		if (ordered)
			btrfs_put_ordered_extent(ordered);
1519 1520 1521
		clear_extent_bit(&inode->io_tree, start_pos, last_pos,
				 EXTENT_DIRTY | EXTENT_DELALLOC |
				 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
1522
				 0, 0, cached_state);
1523 1524 1525
		*lockstart = start_pos;
		*lockend = last_pos;
		ret = 1;
1526
	}
1527

1528
	for (i = 0; i < num_pages; i++) {
1529 1530
		if (clear_page_dirty_for_io(pages[i]))
			account_page_redirty(pages[i]);
1531 1532 1533
		set_page_extent_mapped(pages[i]);
		WARN_ON(!PageLocked(pages[i]));
	}
1534

1535
	return ret;
C
Chris Mason 已提交
1536 1537
}

1538
static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1539 1540
				    size_t *write_bytes)
{
1541 1542
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
	struct btrfs_root *root = inode->root;
1543 1544 1545 1546 1547
	struct btrfs_ordered_extent *ordered;
	u64 lockstart, lockend;
	u64 num_bytes;
	int ret;

1548
	ret = btrfs_start_write_no_snapshotting(root);
1549 1550 1551
	if (!ret)
		return -ENOSPC;

1552
	lockstart = round_down(pos, fs_info->sectorsize);
1553
	lockend = round_up(pos + *write_bytes,
1554
			   fs_info->sectorsize) - 1;
1555 1556

	while (1) {
1557
		lock_extent(&inode->io_tree, lockstart, lockend);
1558 1559 1560 1561 1562
		ordered = btrfs_lookup_ordered_range(inode, lockstart,
						     lockend - lockstart + 1);
		if (!ordered) {
			break;
		}
1563 1564
		unlock_extent(&inode->io_tree, lockstart, lockend);
		btrfs_start_ordered_extent(&inode->vfs_inode, ordered, 1);
1565 1566 1567 1568
		btrfs_put_ordered_extent(ordered);
	}

	num_bytes = lockend - lockstart + 1;
1569 1570
	ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
			NULL, NULL, NULL);
1571 1572
	if (ret <= 0) {
		ret = 0;
1573
		btrfs_end_write_no_snapshotting(root);
1574
	} else {
1575 1576
		*write_bytes = min_t(size_t, *write_bytes ,
				     num_bytes - pos + lockstart);
1577 1578
	}

1579
	unlock_extent(&inode->io_tree, lockstart, lockend);
1580 1581 1582 1583

	return ret;
}

J
Josef Bacik 已提交
1584 1585 1586
static noinline ssize_t __btrfs_buffered_write(struct file *file,
					       struct iov_iter *i,
					       loff_t pos)
1587
{
A
Al Viro 已提交
1588
	struct inode *inode = file_inode(file);
1589
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1590 1591
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct page **pages = NULL;
1592
	struct extent_state *cached_state = NULL;
1593
	struct extent_changeset *data_reserved = NULL;
1594
	u64 release_bytes = 0;
1595 1596
	u64 lockstart;
	u64 lockend;
J
Josef Bacik 已提交
1597 1598
	size_t num_written = 0;
	int nrptrs;
1599
	int ret = 0;
1600
	bool only_release_metadata = false;
1601
	bool force_page_uptodate = false;
1602

1603 1604
	nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
			PAGE_SIZE / (sizeof(struct page *)));
1605 1606
	nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
	nrptrs = max(nrptrs, 8);
1607
	pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
J
Josef Bacik 已提交
1608 1609
	if (!pages)
		return -ENOMEM;
1610

J
Josef Bacik 已提交
1611
	while (iov_iter_count(i) > 0) {
1612
		size_t offset = pos & (PAGE_SIZE - 1);
1613
		size_t sector_offset;
J
Josef Bacik 已提交
1614
		size_t write_bytes = min(iov_iter_count(i),
1615
					 nrptrs * (size_t)PAGE_SIZE -
1616
					 offset);
1617
		size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1618
						PAGE_SIZE);
1619
		size_t reserve_bytes;
J
Josef Bacik 已提交
1620 1621
		size_t dirty_pages;
		size_t copied;
1622 1623
		size_t dirty_sectors;
		size_t num_sectors;
1624
		int extents_locked;
C
Chris Mason 已提交
1625

1626
		WARN_ON(num_pages > nrptrs);
1627

1628 1629 1630 1631
		/*
		 * Fault pages before locking them in prepare_pages
		 * to avoid recursive lock
		 */
J
Josef Bacik 已提交
1632
		if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1633
			ret = -EFAULT;
J
Josef Bacik 已提交
1634
			break;
1635 1636
		}

1637
		sector_offset = pos & (fs_info->sectorsize - 1);
1638
		reserve_bytes = round_up(write_bytes + sector_offset,
1639
				fs_info->sectorsize);
1640

1641 1642 1643
		extent_changeset_release(data_reserved);
		ret = btrfs_check_data_free_space(inode, &data_reserved, pos,
						  write_bytes);
1644 1645 1646
		if (ret < 0) {
			if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
						      BTRFS_INODE_PREALLOC)) &&
1647 1648
			    check_can_nocow(BTRFS_I(inode), pos,
					&write_bytes) > 0) {
1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
				/*
				 * For nodata cow case, no need to reserve
				 * data space.
				 */
				only_release_metadata = true;
				/*
				 * our prealloc extent may be smaller than
				 * write_bytes, so scale down.
				 */
				num_pages = DIV_ROUND_UP(write_bytes + offset,
							 PAGE_SIZE);
				reserve_bytes = round_up(write_bytes +
							 sector_offset,
1662
							 fs_info->sectorsize);
1663 1664 1665 1666
			} else {
				break;
			}
		}
1667

J
Josef Bacik 已提交
1668
		WARN_ON(reserve_bytes == 0);
1669 1670
		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
				reserve_bytes);
1671 1672
		if (ret) {
			if (!only_release_metadata)
1673 1674 1675
				btrfs_free_reserved_data_space(inode,
						data_reserved, pos,
						write_bytes);
1676
			else
1677
				btrfs_end_write_no_snapshotting(root);
1678 1679 1680 1681
			break;
		}

		release_bytes = reserve_bytes;
1682
again:
1683 1684 1685 1686 1687
		/*
		 * This is going to setup the pages array with the number of
		 * pages we want, so we don't really need to worry about the
		 * contents of pages from loop to loop
		 */
1688 1689
		ret = prepare_pages(inode, pages, num_pages,
				    pos, write_bytes,
1690
				    force_page_uptodate);
J
Josef Bacik 已提交
1691 1692 1693
		if (ret) {
			btrfs_delalloc_release_extents(BTRFS_I(inode),
						       reserve_bytes);
J
Josef Bacik 已提交
1694
			break;
J
Josef Bacik 已提交
1695
		}
C
Chris Mason 已提交
1696

1697 1698
		extents_locked = lock_and_cleanup_extent_if_need(
				BTRFS_I(inode), pages,
1699 1700
				num_pages, pos, write_bytes, &lockstart,
				&lockend, &cached_state);
1701 1702
		if (extents_locked < 0) {
			if (extents_locked == -EAGAIN)
1703
				goto again;
J
Josef Bacik 已提交
1704 1705
			btrfs_delalloc_release_extents(BTRFS_I(inode),
						       reserve_bytes);
1706
			ret = extents_locked;
1707 1708 1709
			break;
		}

1710
		copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1711

1712
		num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1713
		dirty_sectors = round_up(copied + sector_offset,
1714 1715
					fs_info->sectorsize);
		dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1716

1717 1718 1719 1720 1721 1722 1723
		/*
		 * if we have trouble faulting in the pages, fall
		 * back to one page at a time
		 */
		if (copied < write_bytes)
			nrptrs = 1;

1724 1725
		if (copied == 0) {
			force_page_uptodate = true;
1726
			dirty_sectors = 0;
1727
			dirty_pages = 0;
1728 1729
		} else {
			force_page_uptodate = false;
1730
			dirty_pages = DIV_ROUND_UP(copied + offset,
1731
						   PAGE_SIZE);
1732
		}
1733

1734
		if (num_sectors > dirty_sectors) {
1735 1736
			/* release everything except the sectors we dirtied */
			release_bytes -= dirty_sectors <<
1737
						fs_info->sb->s_blocksize_bits;
1738
			if (only_release_metadata) {
1739
				btrfs_delalloc_release_metadata(BTRFS_I(inode),
1740
								release_bytes);
1741 1742 1743
			} else {
				u64 __pos;

1744
				__pos = round_down(pos,
1745
						   fs_info->sectorsize) +
1746
					(dirty_pages << PAGE_SHIFT);
1747 1748 1749
				btrfs_delalloc_release_space(inode,
						data_reserved, __pos,
						release_bytes);
1750
			}
1751 1752
		}

1753
		release_bytes = round_up(copied + sector_offset,
1754
					fs_info->sectorsize);
1755 1756

		if (copied > 0)
1757 1758
			ret = btrfs_dirty_pages(inode, pages, dirty_pages,
						pos, copied, NULL);
1759
		if (extents_locked)
1760 1761 1762
			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
					     lockstart, lockend, &cached_state,
					     GFP_NOFS);
J
Josef Bacik 已提交
1763
		btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1764 1765
		if (ret) {
			btrfs_drop_pages(pages, num_pages);
1766
			break;
1767
		}
C
Chris Mason 已提交
1768

1769
		release_bytes = 0;
1770
		if (only_release_metadata)
1771
			btrfs_end_write_no_snapshotting(root);
1772

1773
		if (only_release_metadata && copied > 0) {
1774
			lockstart = round_down(pos,
1775
					       fs_info->sectorsize);
1776
			lockend = round_up(pos + copied,
1777
					   fs_info->sectorsize) - 1;
1778 1779 1780 1781 1782 1783 1784

			set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
				       lockend, EXTENT_NORESERVE, NULL,
				       NULL, GFP_NOFS);
			only_release_metadata = false;
		}

1785 1786
		btrfs_drop_pages(pages, num_pages);

J
Josef Bacik 已提交
1787 1788
		cond_resched();

1789
		balance_dirty_pages_ratelimited(inode->i_mapping);
1790
		if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1)
1791
			btrfs_btree_balance_dirty(fs_info);
1792

1793 1794
		pos += copied;
		num_written += copied;
J
Josef Bacik 已提交
1795
	}
C
Chris Mason 已提交
1796

J
Josef Bacik 已提交
1797 1798
	kfree(pages);

1799
	if (release_bytes) {
1800
		if (only_release_metadata) {
1801
			btrfs_end_write_no_snapshotting(root);
1802 1803
			btrfs_delalloc_release_metadata(BTRFS_I(inode),
					release_bytes);
1804
		} else {
1805 1806 1807
			btrfs_delalloc_release_space(inode, data_reserved,
					round_down(pos, fs_info->sectorsize),
					release_bytes);
1808
		}
1809 1810
	}

1811
	extent_changeset_free(data_reserved);
J
Josef Bacik 已提交
1812 1813 1814
	return num_written ? num_written : ret;
}

1815
static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
J
Josef Bacik 已提交
1816 1817
{
	struct file *file = iocb->ki_filp;
1818
	struct inode *inode = file_inode(file);
1819
	loff_t pos = iocb->ki_pos;
J
Josef Bacik 已提交
1820 1821 1822 1823 1824
	ssize_t written;
	ssize_t written_buffered;
	loff_t endbyte;
	int err;

1825
	written = generic_file_direct_write(iocb, from);
J
Josef Bacik 已提交
1826

A
Al Viro 已提交
1827
	if (written < 0 || !iov_iter_count(from))
J
Josef Bacik 已提交
1828 1829 1830
		return written;

	pos += written;
1831
	written_buffered = __btrfs_buffered_write(file, from, pos);
J
Josef Bacik 已提交
1832 1833 1834
	if (written_buffered < 0) {
		err = written_buffered;
		goto out;
C
Chris Mason 已提交
1835
	}
1836 1837 1838 1839
	/*
	 * Ensure all data is persisted. We want the next direct IO read to be
	 * able to read what was just written.
	 */
J
Josef Bacik 已提交
1840
	endbyte = pos + written_buffered - 1;
1841
	err = btrfs_fdatawrite_range(inode, pos, endbyte);
1842 1843
	if (err)
		goto out;
1844
	err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
J
Josef Bacik 已提交
1845 1846 1847
	if (err)
		goto out;
	written += written_buffered;
1848
	iocb->ki_pos = pos + written_buffered;
1849 1850
	invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
				 endbyte >> PAGE_SHIFT);
C
Chris Mason 已提交
1851
out:
J
Josef Bacik 已提交
1852 1853
	return written ? written : err;
}
1854

1855 1856 1857 1858 1859 1860 1861
static void update_time_for_write(struct inode *inode)
{
	struct timespec now;

	if (IS_NOCMTIME(inode))
		return;

1862
	now = current_time(inode);
1863 1864 1865 1866 1867 1868 1869 1870 1871 1872
	if (!timespec_equal(&inode->i_mtime, &now))
		inode->i_mtime = now;

	if (!timespec_equal(&inode->i_ctime, &now))
		inode->i_ctime = now;

	if (IS_I_VERSION(inode))
		inode_inc_iversion(inode);
}

A
Al Viro 已提交
1873 1874
static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
				    struct iov_iter *from)
J
Josef Bacik 已提交
1875 1876
{
	struct file *file = iocb->ki_filp;
A
Al Viro 已提交
1877
	struct inode *inode = file_inode(file);
1878
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
J
Josef Bacik 已提交
1879
	struct btrfs_root *root = BTRFS_I(inode)->root;
1880
	u64 start_pos;
1881
	u64 end_pos;
J
Josef Bacik 已提交
1882
	ssize_t num_written = 0;
1883
	bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
1884
	ssize_t err;
1885
	loff_t pos;
G
Goldwyn Rodrigues 已提交
1886
	size_t count = iov_iter_count(from);
1887 1888
	loff_t oldsize;
	int clean_page = 0;
J
Josef Bacik 已提交
1889

1890 1891 1892 1893
	if (!(iocb->ki_flags & IOCB_DIRECT) &&
	    (iocb->ki_flags & IOCB_NOWAIT))
		return -EOPNOTSUPP;

1894 1895
	if (!inode_trylock(inode)) {
		if (iocb->ki_flags & IOCB_NOWAIT)
G
Goldwyn Rodrigues 已提交
1896
			return -EAGAIN;
1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907
		inode_lock(inode);
	}

	err = generic_write_checks(iocb, from);
	if (err <= 0) {
		inode_unlock(inode);
		return err;
	}

	pos = iocb->ki_pos;
	if (iocb->ki_flags & IOCB_NOWAIT) {
G
Goldwyn Rodrigues 已提交
1908 1909 1910 1911 1912 1913 1914 1915 1916 1917
		/*
		 * We will allocate space in case nodatacow is not set,
		 * so bail
		 */
		if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
					      BTRFS_INODE_PREALLOC)) ||
		    check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) {
			inode_unlock(inode);
			return -EAGAIN;
		}
J
Josef Bacik 已提交
1918 1919
	}

1920
	current->backing_dev_info = inode_to_bdi(inode);
1921
	err = file_remove_privs(file);
J
Josef Bacik 已提交
1922
	if (err) {
A
Al Viro 已提交
1923
		inode_unlock(inode);
J
Josef Bacik 已提交
1924 1925 1926 1927 1928 1929 1930 1931 1932
		goto out;
	}

	/*
	 * If BTRFS flips readonly due to some impossible error
	 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
	 * although we have opened a file as writable, we have
	 * to stop this write operation to ensure FS consistency.
	 */
1933
	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
A
Al Viro 已提交
1934
		inode_unlock(inode);
J
Josef Bacik 已提交
1935 1936 1937 1938
		err = -EROFS;
		goto out;
	}

1939 1940 1941 1942 1943 1944 1945
	/*
	 * We reserve space for updating the inode when we reserve space for the
	 * extent we are going to write, so we will enospc out there.  We don't
	 * need to start yet another transaction to update the inode as we will
	 * update the inode when we finish writing whatever data we write.
	 */
	update_time_for_write(inode);
J
Josef Bacik 已提交
1946

1947
	start_pos = round_down(pos, fs_info->sectorsize);
1948 1949
	oldsize = i_size_read(inode);
	if (start_pos > oldsize) {
1950
		/* Expand hole size to cover write data, preventing empty gap */
1951
		end_pos = round_up(pos + count,
1952
				   fs_info->sectorsize);
1953
		err = btrfs_cont_expand(inode, oldsize, end_pos);
1954
		if (err) {
A
Al Viro 已提交
1955
			inode_unlock(inode);
1956 1957
			goto out;
		}
1958
		if (start_pos > round_up(oldsize, fs_info->sectorsize))
1959
			clean_page = 1;
1960 1961
	}

1962 1963 1964
	if (sync)
		atomic_inc(&BTRFS_I(inode)->sync_writers);

1965
	if (iocb->ki_flags & IOCB_DIRECT) {
1966
		num_written = __btrfs_direct_write(iocb, from);
J
Josef Bacik 已提交
1967
	} else {
A
Al Viro 已提交
1968
		num_written = __btrfs_buffered_write(file, from, pos);
J
Josef Bacik 已提交
1969
		if (num_written > 0)
1970
			iocb->ki_pos = pos + num_written;
1971 1972 1973
		if (clean_page)
			pagecache_isize_extended(inode, oldsize,
						i_size_read(inode));
J
Josef Bacik 已提交
1974 1975
	}

A
Al Viro 已提交
1976
	inode_unlock(inode);
1977

1978
	/*
1979 1980
	 * We also have to set last_sub_trans to the current log transid,
	 * otherwise subsequent syncs to a file that's been synced in this
1981
	 * transaction will appear to have already occurred.
1982
	 */
1983
	spin_lock(&BTRFS_I(inode)->lock);
1984
	BTRFS_I(inode)->last_sub_trans = root->log_transid;
1985
	spin_unlock(&BTRFS_I(inode)->lock);
1986 1987
	if (num_written > 0)
		num_written = generic_write_sync(iocb, num_written);
1988

1989 1990
	if (sync)
		atomic_dec(&BTRFS_I(inode)->sync_writers);
1991
out:
C
Chris Mason 已提交
1992 1993 1994 1995
	current->backing_dev_info = NULL;
	return num_written ? num_written : err;
}

C
Chris Mason 已提交
1996
int btrfs_release_file(struct inode *inode, struct file *filp)
1997
{
1998 1999 2000
	struct btrfs_file_private *private = filp->private_data;

	if (private && private->trans)
S
Sage Weil 已提交
2001
		btrfs_ioctl_trans_end(filp);
2002 2003 2004 2005 2006
	if (private && private->filldir_buf)
		kfree(private->filldir_buf);
	kfree(private);
	filp->private_data = NULL;

2007 2008 2009 2010 2011 2012 2013 2014 2015
	/*
	 * ordered_data_close is set by settattr when we are about to truncate
	 * a file from a non-zero size to a zero size.  This tries to
	 * flush down new bytes that may have been written if the
	 * application were using truncate to replace a file in place.
	 */
	if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
			       &BTRFS_I(inode)->runtime_flags))
			filemap_flush(inode->i_mapping);
2016 2017 2018
	return 0;
}

2019 2020 2021 2022 2023
static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
{
	int ret;

	atomic_inc(&BTRFS_I(inode)->sync_writers);
2024
	ret = btrfs_fdatawrite_range(inode, start, end);
2025 2026 2027 2028 2029
	atomic_dec(&BTRFS_I(inode)->sync_writers);

	return ret;
}

C
Chris Mason 已提交
2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040
/*
 * fsync call for both files and directories.  This logs the inode into
 * the tree log instead of forcing full commits whenever possible.
 *
 * It needs to call filemap_fdatawait so that all ordered extent updates are
 * in the metadata btree are up to date for copying to the log.
 *
 * It drops the inode mutex before doing the tree log commit.  This is an
 * important optimization for directories because holding the mutex prevents
 * new operations on the dir while we write to disk.
 */
2041
int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
C
Chris Mason 已提交
2042
{
2043
	struct dentry *dentry = file_dentry(file);
2044
	struct inode *inode = d_inode(dentry);
2045
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
C
Chris Mason 已提交
2046 2047
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
2048
	struct btrfs_log_ctx ctx;
2049
	int ret = 0, err;
2050
	bool full_sync = false;
2051
	u64 len;
C
Chris Mason 已提交
2052

2053 2054 2055 2056 2057
	/*
	 * The range length can be represented by u64, we have to do the typecasts
	 * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync()
	 */
	len = (u64)end - (u64)start + 1;
2058
	trace_btrfs_sync_file(file, datasync);
2059

2060 2061
	btrfs_init_log_ctx(&ctx, inode);

2062 2063 2064
	/*
	 * We write the dirty pages in the range and wait until they complete
	 * out of the ->i_mutex. If so, we can flush the dirty pages by
2065 2066
	 * multi-task, and make the performance up.  See
	 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2067
	 */
2068
	ret = start_ordered_ops(inode, start, end);
2069
	if (ret)
2070
		goto out;
2071

A
Al Viro 已提交
2072
	inode_lock(inode);
M
Miao Xie 已提交
2073
	atomic_inc(&root->log_batch);
2074 2075
	full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			     &BTRFS_I(inode)->runtime_flags);
2076 2077 2078 2079
	/*
	 * We might have have had more pages made dirty after calling
	 * start_ordered_ops and before acquiring the inode's i_mutex.
	 */
2080
	if (full_sync) {
2081 2082 2083 2084 2085 2086
		/*
		 * For a full sync, we need to make sure any ordered operations
		 * start and finish before we start logging the inode, so that
		 * all extents are persisted and the respective file extent
		 * items are in the fs/subvol btree.
		 */
2087
		ret = btrfs_wait_ordered_range(inode, start, len);
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123
	} else {
		/*
		 * Start any new ordered operations before starting to log the
		 * inode. We will wait for them to finish in btrfs_sync_log().
		 *
		 * Right before acquiring the inode's mutex, we might have new
		 * writes dirtying pages, which won't immediately start the
		 * respective ordered operations - that is done through the
		 * fill_delalloc callbacks invoked from the writepage and
		 * writepages address space operations. So make sure we start
		 * all ordered operations before starting to log our inode. Not
		 * doing this means that while logging the inode, writeback
		 * could start and invoke writepage/writepages, which would call
		 * the fill_delalloc callbacks (cow_file_range,
		 * submit_compressed_extents). These callbacks add first an
		 * extent map to the modified list of extents and then create
		 * the respective ordered operation, which means in
		 * tree-log.c:btrfs_log_inode() we might capture all existing
		 * ordered operations (with btrfs_get_logged_extents()) before
		 * the fill_delalloc callback adds its ordered operation, and by
		 * the time we visit the modified list of extent maps (with
		 * btrfs_log_changed_extents()), we see and process the extent
		 * map they created. We then use the extent map to construct a
		 * file extent item for logging without waiting for the
		 * respective ordered operation to finish - this file extent
		 * item points to a disk location that might not have yet been
		 * written to, containing random data - so after a crash a log
		 * replay will make our inode have file extent items that point
		 * to disk locations containing invalid data, as we returned
		 * success to userspace without waiting for the respective
		 * ordered operation to finish, because it wasn't captured by
		 * btrfs_get_logged_extents().
		 */
		ret = start_ordered_ops(inode, start, end);
	}
	if (ret) {
A
Al Viro 已提交
2124
		inode_unlock(inode);
2125
		goto out;
2126
	}
M
Miao Xie 已提交
2127
	atomic_inc(&root->log_batch);
2128

C
Chris Mason 已提交
2129
	/*
2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155
	 * If the last transaction that changed this file was before the current
	 * transaction and we have the full sync flag set in our inode, we can
	 * bail out now without any syncing.
	 *
	 * Note that we can't bail out if the full sync flag isn't set. This is
	 * because when the full sync flag is set we start all ordered extents
	 * and wait for them to fully complete - when they complete they update
	 * the inode's last_trans field through:
	 *
	 *     btrfs_finish_ordered_io() ->
	 *         btrfs_update_inode_fallback() ->
	 *             btrfs_update_inode() ->
	 *                 btrfs_set_inode_last_trans()
	 *
	 * So we are sure that last_trans is up to date and can do this check to
	 * bail out safely. For the fast path, when the full sync flag is not
	 * set in our inode, we can not do it because we start only our ordered
	 * extents and don't wait for them to complete (that is when
	 * btrfs_finish_ordered_io runs), so here at this point their last_trans
	 * value might be less than or equals to fs_info->last_trans_committed,
	 * and setting a speculative last_trans for an inode when a buffered
	 * write is made (such as fs_info->generation + 1 for example) would not
	 * be reliable since after setting the value and before fsync is called
	 * any number of transactions can start and commit (transaction kthread
	 * commits the current transaction periodically), and a transaction
	 * commit does not start nor waits for ordered extents to complete.
2156
	 */
J
Josef Bacik 已提交
2157
	smp_mb();
2158
	if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2159
	    (full_sync && BTRFS_I(inode)->last_trans <=
2160
	     fs_info->last_trans_committed) ||
2161 2162
	    (!btrfs_have_ordered_extents_in_range(inode, start, len) &&
	     BTRFS_I(inode)->last_trans
2163
	     <= fs_info->last_trans_committed)) {
J
Josef Bacik 已提交
2164
		/*
2165
		 * We've had everything committed since the last time we were
J
Josef Bacik 已提交
2166 2167 2168 2169 2170
		 * modified so clear this flag in case it was set for whatever
		 * reason, it's no longer relevant.
		 */
		clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			  &BTRFS_I(inode)->runtime_flags);
2171 2172 2173 2174
		/*
		 * An ordered extent might have started before and completed
		 * already with io errors, in which case the inode was not
		 * updated and we end up here. So check the inode's mapping
2175 2176
		 * for any errors that might have happened since we last
		 * checked called fsync.
2177
		 */
2178
		ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
A
Al Viro 已提交
2179
		inode_unlock(inode);
2180 2181 2182 2183
		goto out;
	}

	/*
2184 2185
	 * ok we haven't committed the transaction yet, lets do a commit
	 */
2186
	if (file->private_data)
S
Sage Weil 已提交
2187 2188
		btrfs_ioctl_trans_end(file);

2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199
	/*
	 * We use start here because we will need to wait on the IO to complete
	 * in btrfs_sync_log, which could require joining a transaction (for
	 * example checking cross references in the nocow path).  If we use join
	 * here we could get into a situation where we're waiting on IO to
	 * happen that is blocked on a transaction trying to commit.  With start
	 * we inc the extwriter counter, so we wait for all extwriters to exit
	 * before we start blocking join'ers.  This comment is to keep somebody
	 * from thinking they are super smart and changing this to
	 * btrfs_join_transaction *cough*Josef*cough*.
	 */
2200 2201 2202
	trans = btrfs_start_transaction(root, 0);
	if (IS_ERR(trans)) {
		ret = PTR_ERR(trans);
A
Al Viro 已提交
2203
		inode_unlock(inode);
C
Chris Mason 已提交
2204 2205
		goto out;
	}
2206
	trans->sync = true;
2207

2208
	ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx);
2209
	if (ret < 0) {
2210 2211
		/* Fallthrough and commit/free transaction. */
		ret = 1;
2212
	}
C
Chris Mason 已提交
2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223

	/* we've logged all the items and now have a consistent
	 * version of the file in the log.  It is possible that
	 * someone will come in and modify the file, but that's
	 * fine because the log is consistent on disk, and we
	 * have references to all of the file's extents
	 *
	 * It is possible that someone will come in and log the
	 * file again, but that will end up using the synchronization
	 * inside btrfs_sync_log to keep things safe.
	 */
A
Al Viro 已提交
2224
	inode_unlock(inode);
C
Chris Mason 已提交
2225

2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239
	/*
	 * If any of the ordered extents had an error, just return it to user
	 * space, so that the application knows some writes didn't succeed and
	 * can take proper action (retry for e.g.). Blindly committing the
	 * transaction in this case, would fool userspace that everything was
	 * successful. And we also want to make sure our log doesn't contain
	 * file extent items pointing to extents that weren't fully written to -
	 * just like in the non fast fsync path, where we check for the ordered
	 * operation's error flag before writing to the log tree and return -EIO
	 * if any of them had this flag set (btrfs_wait_ordered_range) -
	 * therefore we need to check for errors in the ordered operations,
	 * which are indicated by ctx.io_err.
	 */
	if (ctx.io_err) {
2240
		btrfs_end_transaction(trans);
2241 2242 2243 2244
		ret = ctx.io_err;
		goto out;
	}

2245
	if (ret != BTRFS_NO_LOG_SYNC) {
2246
		if (!ret) {
2247
			ret = btrfs_sync_log(trans, root, &ctx);
2248
			if (!ret) {
2249
				ret = btrfs_end_transaction(trans);
2250
				goto out;
2251
			}
2252
		}
2253
		if (!full_sync) {
2254
			ret = btrfs_wait_ordered_range(inode, start, len);
2255
			if (ret) {
2256
				btrfs_end_transaction(trans);
2257
				goto out;
2258
			}
2259
		}
2260
		ret = btrfs_commit_transaction(trans);
2261
	} else {
2262
		ret = btrfs_end_transaction(trans);
2263
	}
C
Chris Mason 已提交
2264
out:
2265
	ASSERT(list_empty(&ctx.list));
2266 2267 2268
	err = file_check_and_advance_wb_err(file);
	if (!ret)
		ret = err;
2269
	return ret > 0 ? -EIO : ret;
C
Chris Mason 已提交
2270 2271
}

2272
static const struct vm_operations_struct btrfs_file_vm_ops = {
2273
	.fault		= filemap_fault,
2274
	.map_pages	= filemap_map_pages,
C
Chris Mason 已提交
2275 2276 2277 2278 2279
	.page_mkwrite	= btrfs_page_mkwrite,
};

static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
{
M
Miao Xie 已提交
2280 2281 2282 2283 2284
	struct address_space *mapping = filp->f_mapping;

	if (!mapping->a_ops->readpage)
		return -ENOEXEC;

C
Chris Mason 已提交
2285
	file_accessed(filp);
M
Miao Xie 已提交
2286 2287
	vma->vm_ops = &btrfs_file_vm_ops;

C
Chris Mason 已提交
2288 2289 2290
	return 0;
}

2291
static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
J
Josef Bacik 已提交
2292 2293 2294 2295 2296 2297 2298 2299 2300
			  int slot, u64 start, u64 end)
{
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;

	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
		return 0;

	btrfs_item_key_to_cpu(leaf, &key, slot);
2301
	if (key.objectid != btrfs_ino(inode) ||
J
Josef Bacik 已提交
2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319
	    key.type != BTRFS_EXTENT_DATA_KEY)
		return 0;

	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);

	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
		return 0;

	if (btrfs_file_extent_disk_bytenr(leaf, fi))
		return 0;

	if (key.offset == end)
		return 1;
	if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
		return 1;
	return 0;
}

2320 2321 2322
static int fill_holes(struct btrfs_trans_handle *trans,
		struct btrfs_inode *inode,
		struct btrfs_path *path, u64 offset, u64 end)
J
Josef Bacik 已提交
2323
{
2324 2325
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
	struct btrfs_root *root = inode->root;
J
Josef Bacik 已提交
2326 2327 2328
	struct extent_buffer *leaf;
	struct btrfs_file_extent_item *fi;
	struct extent_map *hole_em;
2329
	struct extent_map_tree *em_tree = &inode->extent_tree;
J
Josef Bacik 已提交
2330 2331 2332
	struct btrfs_key key;
	int ret;

2333
	if (btrfs_fs_incompat(fs_info, NO_HOLES))
2334 2335
		goto out;

2336
	key.objectid = btrfs_ino(inode);
J
Josef Bacik 已提交
2337 2338 2339 2340
	key.type = BTRFS_EXTENT_DATA_KEY;
	key.offset = offset;

	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2341 2342 2343 2344 2345 2346 2347
	if (ret <= 0) {
		/*
		 * We should have dropped this offset, so if we find it then
		 * something has gone horribly wrong.
		 */
		if (ret == 0)
			ret = -EINVAL;
J
Josef Bacik 已提交
2348
		return ret;
2349
	}
J
Josef Bacik 已提交
2350 2351

	leaf = path->nodes[0];
2352
	if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
J
Josef Bacik 已提交
2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366
		u64 num_bytes;

		path->slots[0]--;
		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
			end - offset;
		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_offset(leaf, fi, 0);
		btrfs_mark_buffer_dirty(leaf);
		goto out;
	}

2367
	if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
J
Josef Bacik 已提交
2368 2369 2370
		u64 num_bytes;

		key.offset = offset;
2371
		btrfs_set_item_key_safe(fs_info, path, &key);
J
Josef Bacik 已提交
2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383
		fi = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_file_extent_item);
		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
			offset;
		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
		btrfs_set_file_extent_offset(leaf, fi, 0);
		btrfs_mark_buffer_dirty(leaf);
		goto out;
	}
	btrfs_release_path(path);

2384
	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2385
			offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
J
Josef Bacik 已提交
2386 2387 2388 2389 2390 2391 2392 2393 2394
	if (ret)
		return ret;

out:
	btrfs_release_path(path);

	hole_em = alloc_extent_map();
	if (!hole_em) {
		btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2395
		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
J
Josef Bacik 已提交
2396 2397 2398
	} else {
		hole_em->start = offset;
		hole_em->len = end - offset;
J
Josef Bacik 已提交
2399
		hole_em->ram_bytes = hole_em->len;
J
Josef Bacik 已提交
2400 2401 2402 2403
		hole_em->orig_start = offset;

		hole_em->block_start = EXTENT_MAP_HOLE;
		hole_em->block_len = 0;
2404
		hole_em->orig_block_len = 0;
2405
		hole_em->bdev = fs_info->fs_devices->latest_bdev;
J
Josef Bacik 已提交
2406 2407 2408 2409 2410 2411
		hole_em->compress_type = BTRFS_COMPRESS_NONE;
		hole_em->generation = trans->transid;

		do {
			btrfs_drop_extent_cache(inode, offset, end - 1, 0);
			write_lock(&em_tree->lock);
J
Josef Bacik 已提交
2412
			ret = add_extent_mapping(em_tree, hole_em, 1);
J
Josef Bacik 已提交
2413 2414 2415 2416 2417
			write_unlock(&em_tree->lock);
		} while (ret == -EEXIST);
		free_extent_map(hole_em);
		if (ret)
			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2418
					&inode->runtime_flags);
J
Josef Bacik 已提交
2419 2420 2421 2422 2423
	}

	return 0;
}

2424 2425 2426 2427 2428 2429 2430 2431
/*
 * Find a hole extent on given inode and change start/len to the end of hole
 * extent.(hole/vacuum extent whose em->start <= start &&
 *	   em->start + em->len > start)
 * When a hole extent is found, return 1 and modify start/len.
 */
static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
{
2432
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2433 2434 2435
	struct extent_map *em;
	int ret = 0;

2436 2437 2438
	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
			      round_down(*start, fs_info->sectorsize),
			      round_up(*len, fs_info->sectorsize), 0);
2439 2440
	if (IS_ERR(em))
		return PTR_ERR(em);
2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452

	/* Hole or vacuum extent(only exists in no-hole mode) */
	if (em->block_start == EXTENT_MAP_HOLE) {
		ret = 1;
		*len = em->start + em->len > *start + *len ?
		       0 : *start + *len - em->start - em->len;
		*start = em->start + em->len;
	}
	free_extent_map(em);
	return ret;
}

J
Josef Bacik 已提交
2453 2454
static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
{
2455
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
J
Josef Bacik 已提交
2456 2457 2458 2459 2460
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct extent_state *cached_state = NULL;
	struct btrfs_path *path;
	struct btrfs_block_rsv *rsv;
	struct btrfs_trans_handle *trans;
2461 2462 2463 2464 2465 2466
	u64 lockstart;
	u64 lockend;
	u64 tail_start;
	u64 tail_len;
	u64 orig_start = offset;
	u64 cur_offset;
C
Chris Mason 已提交
2467
	u64 min_size = btrfs_calc_trans_metadata_size(fs_info, 1);
J
Josef Bacik 已提交
2468 2469 2470
	u64 drop_end;
	int ret = 0;
	int err = 0;
2471
	unsigned int rsv_count;
2472
	bool same_block;
2473
	bool no_holes = btrfs_fs_incompat(fs_info, NO_HOLES);
2474
	u64 ino_size;
2475
	bool truncated_block = false;
2476
	bool updated_inode = false;
J
Josef Bacik 已提交
2477

2478 2479 2480
	ret = btrfs_wait_ordered_range(inode, offset, len);
	if (ret)
		return ret;
J
Josef Bacik 已提交
2481

A
Al Viro 已提交
2482
	inode_lock(inode);
2483
	ino_size = round_up(inode->i_size, fs_info->sectorsize);
2484 2485 2486 2487 2488 2489 2490 2491 2492
	ret = find_first_non_hole(inode, &offset, &len);
	if (ret < 0)
		goto out_only_mutex;
	if (ret && !len) {
		/* Already in a large hole */
		ret = 0;
		goto out_only_mutex;
	}

2493
	lockstart = round_up(offset, btrfs_inode_sectorsize(inode));
2494
	lockend = round_down(offset + len,
2495
			     btrfs_inode_sectorsize(inode)) - 1;
2496 2497
	same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
		== (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2498
	/*
2499
	 * We needn't truncate any block which is beyond the end of the file
2500 2501
	 * because we are sure there is no data there.
	 */
J
Josef Bacik 已提交
2502
	/*
2503 2504
	 * Only do this if we are in the same block and we aren't doing the
	 * entire block.
J
Josef Bacik 已提交
2505
	 */
2506
	if (same_block && len < fs_info->sectorsize) {
2507
		if (offset < ino_size) {
2508 2509
			truncated_block = true;
			ret = btrfs_truncate_block(inode, offset, len, 0);
2510 2511 2512
		} else {
			ret = 0;
		}
2513
		goto out_only_mutex;
J
Josef Bacik 已提交
2514 2515
	}

2516
	/* zero back part of the first block */
2517
	if (offset < ino_size) {
2518 2519
		truncated_block = true;
		ret = btrfs_truncate_block(inode, offset, 0, 0);
2520
		if (ret) {
A
Al Viro 已提交
2521
			inode_unlock(inode);
2522 2523
			return ret;
		}
J
Josef Bacik 已提交
2524 2525
	}

2526 2527
	/* Check the aligned pages after the first unaligned page,
	 * if offset != orig_start, which means the first unaligned page
2528
	 * including several following pages are already in holes,
2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553
	 * the extra check can be skipped */
	if (offset == orig_start) {
		/* after truncate page, check hole again */
		len = offset + len - lockstart;
		offset = lockstart;
		ret = find_first_non_hole(inode, &offset, &len);
		if (ret < 0)
			goto out_only_mutex;
		if (ret && !len) {
			ret = 0;
			goto out_only_mutex;
		}
		lockstart = offset;
	}

	/* Check the tail unaligned part is in a hole */
	tail_start = lockend + 1;
	tail_len = offset + len - tail_start;
	if (tail_len) {
		ret = find_first_non_hole(inode, &tail_start, &tail_len);
		if (unlikely(ret < 0))
			goto out_only_mutex;
		if (!ret) {
			/* zero the front end of the last page */
			if (tail_start + tail_len < ino_size) {
2554 2555 2556 2557
				truncated_block = true;
				ret = btrfs_truncate_block(inode,
							tail_start + tail_len,
							0, 1);
2558 2559
				if (ret)
					goto out_only_mutex;
2560
			}
M
Miao Xie 已提交
2561
		}
J
Josef Bacik 已提交
2562 2563 2564
	}

	if (lockend < lockstart) {
2565 2566
		ret = 0;
		goto out_only_mutex;
J
Josef Bacik 已提交
2567 2568 2569 2570 2571 2572 2573 2574
	}

	while (1) {
		struct btrfs_ordered_extent *ordered;

		truncate_pagecache_range(inode, lockstart, lockend);

		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2575
				 &cached_state);
J
Josef Bacik 已提交
2576 2577 2578 2579 2580 2581 2582 2583
		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);

		/*
		 * We need to make sure we have no ordered extents in this range
		 * and nobody raced in and read a page in this range, if we did
		 * we need to try again.
		 */
		if ((!ordered ||
2584
		    (ordered->file_offset + ordered->len <= lockstart ||
J
Josef Bacik 已提交
2585
		     ordered->file_offset > lockend)) &&
2586
		     !btrfs_page_exists_in_range(inode, lockstart, lockend)) {
J
Josef Bacik 已提交
2587 2588 2589 2590 2591 2592 2593 2594
			if (ordered)
				btrfs_put_ordered_extent(ordered);
			break;
		}
		if (ordered)
			btrfs_put_ordered_extent(ordered);
		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
				     lockend, &cached_state, GFP_NOFS);
2595 2596 2597
		ret = btrfs_wait_ordered_range(inode, lockstart,
					       lockend - lockstart + 1);
		if (ret) {
A
Al Viro 已提交
2598
			inode_unlock(inode);
2599 2600
			return ret;
		}
J
Josef Bacik 已提交
2601 2602 2603 2604 2605 2606 2607 2608
	}

	path = btrfs_alloc_path();
	if (!path) {
		ret = -ENOMEM;
		goto out;
	}

2609
	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
J
Josef Bacik 已提交
2610 2611 2612 2613
	if (!rsv) {
		ret = -ENOMEM;
		goto out_free;
	}
C
Chris Mason 已提交
2614
	rsv->size = btrfs_calc_trans_metadata_size(fs_info, 1);
J
Josef Bacik 已提交
2615 2616 2617 2618 2619
	rsv->failfast = 1;

	/*
	 * 1 - update the inode
	 * 1 - removing the extents in the range
2620
	 * 1 - adding the hole extent if no_holes isn't set
J
Josef Bacik 已提交
2621
	 */
2622 2623
	rsv_count = no_holes ? 2 : 3;
	trans = btrfs_start_transaction(root, rsv_count);
J
Josef Bacik 已提交
2624 2625 2626 2627 2628
	if (IS_ERR(trans)) {
		err = PTR_ERR(trans);
		goto out_free;
	}

2629
	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2630
				      min_size, 0);
J
Josef Bacik 已提交
2631 2632 2633
	BUG_ON(ret);
	trans->block_rsv = rsv;

2634 2635
	cur_offset = lockstart;
	len = lockend - cur_offset;
J
Josef Bacik 已提交
2636 2637 2638
	while (cur_offset < lockend) {
		ret = __btrfs_drop_extents(trans, root, inode, path,
					   cur_offset, lockend + 1,
2639
					   &drop_end, 1, 0, 0, NULL);
J
Josef Bacik 已提交
2640 2641 2642
		if (ret != -ENOSPC)
			break;

2643
		trans->block_rsv = &fs_info->trans_block_rsv;
J
Josef Bacik 已提交
2644

J
Josef Bacik 已提交
2645
		if (cur_offset < drop_end && cur_offset < ino_size) {
2646 2647
			ret = fill_holes(trans, BTRFS_I(inode), path,
					cur_offset, drop_end);
2648
			if (ret) {
2649 2650 2651 2652 2653 2654 2655
				/*
				 * If we failed then we didn't insert our hole
				 * entries for the area we dropped, so now the
				 * fs is corrupted, so we must abort the
				 * transaction.
				 */
				btrfs_abort_transaction(trans, ret);
2656 2657 2658
				err = ret;
				break;
			}
J
Josef Bacik 已提交
2659 2660 2661 2662 2663 2664 2665 2666 2667 2668
		}

		cur_offset = drop_end;

		ret = btrfs_update_inode(trans, root, inode);
		if (ret) {
			err = ret;
			break;
		}

2669
		btrfs_end_transaction(trans);
2670
		btrfs_btree_balance_dirty(fs_info);
J
Josef Bacik 已提交
2671

2672
		trans = btrfs_start_transaction(root, rsv_count);
J
Josef Bacik 已提交
2673 2674 2675 2676 2677 2678
		if (IS_ERR(trans)) {
			ret = PTR_ERR(trans);
			trans = NULL;
			break;
		}

2679
		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2680
					      rsv, min_size, 0);
J
Josef Bacik 已提交
2681 2682
		BUG_ON(ret);	/* shouldn't happen */
		trans->block_rsv = rsv;
2683 2684 2685 2686 2687 2688 2689 2690

		ret = find_first_non_hole(inode, &cur_offset, &len);
		if (unlikely(ret < 0))
			break;
		if (ret && !len) {
			ret = 0;
			break;
		}
J
Josef Bacik 已提交
2691 2692 2693 2694 2695 2696 2697
	}

	if (ret) {
		err = ret;
		goto out_trans;
	}

2698
	trans->block_rsv = &fs_info->trans_block_rsv;
2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711
	/*
	 * If we are using the NO_HOLES feature we might have had already an
	 * hole that overlaps a part of the region [lockstart, lockend] and
	 * ends at (or beyond) lockend. Since we have no file extent items to
	 * represent holes, drop_end can be less than lockend and so we must
	 * make sure we have an extent map representing the existing hole (the
	 * call to __btrfs_drop_extents() might have dropped the existing extent
	 * map representing the existing hole), otherwise the fast fsync path
	 * will not record the existence of the hole region
	 * [existing_hole_start, lockend].
	 */
	if (drop_end <= lockend)
		drop_end = lockend + 1;
2712 2713 2714 2715 2716 2717
	/*
	 * Don't insert file hole extent item if it's for a range beyond eof
	 * (because it's useless) or if it represents a 0 bytes range (when
	 * cur_offset == drop_end).
	 */
	if (cur_offset < ino_size && cur_offset < drop_end) {
2718 2719
		ret = fill_holes(trans, BTRFS_I(inode), path,
				cur_offset, drop_end);
2720
		if (ret) {
2721 2722
			/* Same comment as above. */
			btrfs_abort_transaction(trans, ret);
2723 2724 2725
			err = ret;
			goto out_trans;
		}
J
Josef Bacik 已提交
2726 2727 2728 2729 2730 2731
	}

out_trans:
	if (!trans)
		goto out_free;

2732
	inode_inc_iversion(inode);
2733
	inode->i_mtime = inode->i_ctime = current_time(inode);
2734

2735
	trans->block_rsv = &fs_info->trans_block_rsv;
J
Josef Bacik 已提交
2736
	ret = btrfs_update_inode(trans, root, inode);
2737
	updated_inode = true;
2738
	btrfs_end_transaction(trans);
2739
	btrfs_btree_balance_dirty(fs_info);
J
Josef Bacik 已提交
2740 2741
out_free:
	btrfs_free_path(path);
2742
	btrfs_free_block_rsv(fs_info, rsv);
J
Josef Bacik 已提交
2743 2744 2745
out:
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
			     &cached_state, GFP_NOFS);
2746
out_only_mutex:
2747
	if (!updated_inode && truncated_block && !ret && !err) {
2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759
		/*
		 * If we only end up zeroing part of a page, we still need to
		 * update the inode item, so that all the time fields are
		 * updated as well as the necessary btrfs inode in memory fields
		 * for detecting, at fsync time, if the inode isn't yet in the
		 * log tree or it's there but not up to date.
		 */
		trans = btrfs_start_transaction(root, 1);
		if (IS_ERR(trans)) {
			err = PTR_ERR(trans);
		} else {
			err = btrfs_update_inode(trans, root, inode);
2760
			ret = btrfs_end_transaction(trans);
2761 2762
		}
	}
A
Al Viro 已提交
2763
	inode_unlock(inode);
J
Josef Bacik 已提交
2764 2765 2766 2767 2768
	if (ret && !err)
		err = ret;
	return err;
}

2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799
/* Helper structure to record which range is already reserved */
struct falloc_range {
	struct list_head list;
	u64 start;
	u64 len;
};

/*
 * Helper function to add falloc range
 *
 * Caller should have locked the larger range of extent containing
 * [start, len)
 */
static int add_falloc_range(struct list_head *head, u64 start, u64 len)
{
	struct falloc_range *prev = NULL;
	struct falloc_range *range = NULL;

	if (list_empty(head))
		goto insert;

	/*
	 * As fallocate iterate by bytenr order, we only need to check
	 * the last range.
	 */
	prev = list_entry(head->prev, struct falloc_range, list);
	if (prev->start + prev->len == start) {
		prev->len += len;
		return 0;
	}
insert:
D
David Sterba 已提交
2800
	range = kmalloc(sizeof(*range), GFP_KERNEL);
2801 2802 2803 2804 2805 2806 2807 2808
	if (!range)
		return -ENOMEM;
	range->start = start;
	range->len = len;
	list_add_tail(&range->list, head);
	return 0;
}

2809 2810 2811
static long btrfs_fallocate(struct file *file, int mode,
			    loff_t offset, loff_t len)
{
A
Al Viro 已提交
2812
	struct inode *inode = file_inode(file);
2813
	struct extent_state *cached_state = NULL;
2814
	struct extent_changeset *data_reserved = NULL;
2815 2816 2817
	struct falloc_range *range;
	struct falloc_range *tmp;
	struct list_head reserve_list;
2818 2819 2820 2821 2822 2823
	u64 cur_offset;
	u64 last_byte;
	u64 alloc_start;
	u64 alloc_end;
	u64 alloc_hint = 0;
	u64 locked_end;
2824
	u64 actual_end = 0;
2825
	struct extent_map *em;
2826
	int blocksize = btrfs_inode_sectorsize(inode);
2827 2828
	int ret;

2829 2830
	alloc_start = round_down(offset, blocksize);
	alloc_end = round_up(offset + len, blocksize);
2831
	cur_offset = alloc_start;
2832

J
Josef Bacik 已提交
2833 2834
	/* Make sure we aren't being give some crap mode */
	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2835 2836
		return -EOPNOTSUPP;

J
Josef Bacik 已提交
2837 2838 2839
	if (mode & FALLOC_FL_PUNCH_HOLE)
		return btrfs_punch_hole(inode, offset, len);

2840
	/*
2841 2842 2843
	 * Only trigger disk allocation, don't trigger qgroup reserve
	 *
	 * For qgroup space, it will be checked later.
2844
	 */
2845 2846
	ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
			alloc_end - alloc_start);
2847
	if (ret < 0)
2848 2849
		return ret;

A
Al Viro 已提交
2850
	inode_lock(inode);
2851 2852 2853 2854 2855 2856

	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
		ret = inode_newsize_ok(inode, offset + len);
		if (ret)
			goto out;
	}
2857

2858 2859 2860 2861 2862 2863 2864
	/*
	 * TODO: Move these two operations after we have checked
	 * accurate reserved space, or fallocate can still fail but
	 * with page truncated or size expanded.
	 *
	 * But that's a minor problem and won't do much harm BTW.
	 */
2865
	if (alloc_start > inode->i_size) {
2866 2867
		ret = btrfs_cont_expand(inode, i_size_read(inode),
					alloc_start);
2868 2869
		if (ret)
			goto out;
2870
	} else if (offset + len > inode->i_size) {
2871 2872
		/*
		 * If we are fallocating from the end of the file onward we
2873 2874
		 * need to zero out the end of the block if i_size lands in the
		 * middle of a block.
2875
		 */
2876
		ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
2877 2878
		if (ret)
			goto out;
2879 2880
	}

2881 2882 2883 2884
	/*
	 * wait for ordered IO before we have any locks.  We'll loop again
	 * below with the locks held.
	 */
2885 2886 2887 2888
	ret = btrfs_wait_ordered_range(inode, alloc_start,
				       alloc_end - alloc_start);
	if (ret)
		goto out;
2889

2890 2891 2892 2893 2894 2895 2896 2897
	locked_end = alloc_end - 1;
	while (1) {
		struct btrfs_ordered_extent *ordered;

		/* the extent lock is ordered inside the running
		 * transaction
		 */
		lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
2898
				 locked_end, &cached_state);
2899 2900
		ordered = btrfs_lookup_first_ordered_extent(inode, locked_end);

2901 2902 2903 2904 2905 2906
		if (ordered &&
		    ordered->file_offset + ordered->len > alloc_start &&
		    ordered->file_offset < alloc_end) {
			btrfs_put_ordered_extent(ordered);
			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
					     alloc_start, locked_end,
D
David Sterba 已提交
2907
					     &cached_state, GFP_KERNEL);
2908 2909 2910 2911
			/*
			 * we can't wait on the range with the transaction
			 * running or with the extent lock held
			 */
2912 2913 2914 2915
			ret = btrfs_wait_ordered_range(inode, alloc_start,
						       alloc_end - alloc_start);
			if (ret)
				goto out;
2916 2917 2918 2919 2920 2921 2922
		} else {
			if (ordered)
				btrfs_put_ordered_extent(ordered);
			break;
		}
	}

2923 2924
	/* First, check if we exceed the qgroup limit */
	INIT_LIST_HEAD(&reserve_list);
2925
	while (cur_offset < alloc_end) {
2926
		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
2927
				      alloc_end - cur_offset, 0);
2928 2929
		if (IS_ERR(em)) {
			ret = PTR_ERR(em);
2930 2931
			break;
		}
2932
		last_byte = min(extent_map_end(em), alloc_end);
2933
		actual_end = min_t(u64, extent_map_end(em), offset + len);
2934
		last_byte = ALIGN(last_byte, blocksize);
2935 2936 2937
		if (em->block_start == EXTENT_MAP_HOLE ||
		    (cur_offset >= inode->i_size &&
		     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
2938 2939 2940 2941 2942
			ret = add_falloc_range(&reserve_list, cur_offset,
					       last_byte - cur_offset);
			if (ret < 0) {
				free_extent_map(em);
				break;
2943
			}
2944 2945
			ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
					cur_offset, last_byte - cur_offset);
2946 2947
			if (ret < 0) {
				free_extent_map(em);
2948
				break;
2949
			}
2950 2951 2952 2953 2954 2955
		} else {
			/*
			 * Do not need to reserve unwritten extent for this
			 * range, free reserved data space first, otherwise
			 * it'll result in false ENOSPC error.
			 */
2956 2957
			btrfs_free_reserved_data_space(inode, data_reserved,
					cur_offset, last_byte - cur_offset);
2958 2959 2960
		}
		free_extent_map(em);
		cur_offset = last_byte;
2961 2962 2963 2964 2965 2966 2967 2968 2969 2970
	}

	/*
	 * If ret is still 0, means we're OK to fallocate.
	 * Or just cleanup the list and exit.
	 */
	list_for_each_entry_safe(range, tmp, &reserve_list, list) {
		if (!ret)
			ret = btrfs_prealloc_file_range(inode, mode,
					range->start,
F
Fabian Frederick 已提交
2971
					range->len, i_blocksize(inode),
2972
					offset + len, &alloc_hint);
2973
		else
2974 2975 2976
			btrfs_free_reserved_data_space(inode,
					data_reserved, range->start,
					range->len);
2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996
		list_del(&range->list);
		kfree(range);
	}
	if (ret < 0)
		goto out_unlock;

	if (actual_end > inode->i_size &&
	    !(mode & FALLOC_FL_KEEP_SIZE)) {
		struct btrfs_trans_handle *trans;
		struct btrfs_root *root = BTRFS_I(inode)->root;

		/*
		 * We didn't need to allocate any more space, but we
		 * still extended the size of the file so we need to
		 * update i_size and the inode item.
		 */
		trans = btrfs_start_transaction(root, 1);
		if (IS_ERR(trans)) {
			ret = PTR_ERR(trans);
		} else {
2997
			inode->i_ctime = current_time(inode);
2998 2999 3000 3001
			i_size_write(inode, actual_end);
			btrfs_ordered_update_i_size(inode, actual_end, NULL);
			ret = btrfs_update_inode(trans, root, inode);
			if (ret)
3002
				btrfs_end_transaction(trans);
3003
			else
3004
				ret = btrfs_end_transaction(trans);
3005 3006
		}
	}
3007
out_unlock:
3008
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
D
David Sterba 已提交
3009
			     &cached_state, GFP_KERNEL);
3010
out:
A
Al Viro 已提交
3011
	inode_unlock(inode);
3012
	/* Let go of our reservation. */
3013
	if (ret != 0)
3014 3015
		btrfs_free_reserved_data_space(inode, data_reserved,
				alloc_start, alloc_end - cur_offset);
3016
	extent_changeset_free(data_reserved);
3017 3018 3019
	return ret;
}

3020
static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
J
Josef Bacik 已提交
3021
{
3022
	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3023
	struct extent_map *em = NULL;
J
Josef Bacik 已提交
3024
	struct extent_state *cached_state = NULL;
L
Liu Bo 已提交
3025 3026 3027 3028
	u64 lockstart;
	u64 lockend;
	u64 start;
	u64 len;
J
Josef Bacik 已提交
3029 3030
	int ret = 0;

L
Liu Bo 已提交
3031 3032 3033 3034 3035 3036 3037 3038 3039
	if (inode->i_size == 0)
		return -ENXIO;

	/*
	 * *offset can be negative, in this case we start finding DATA/HOLE from
	 * the very start of the file.
	 */
	start = max_t(loff_t, 0, *offset);

3040
	lockstart = round_down(start, fs_info->sectorsize);
3041
	lockend = round_up(i_size_read(inode),
3042
			   fs_info->sectorsize);
J
Josef Bacik 已提交
3043
	if (lockend <= lockstart)
3044
		lockend = lockstart + fs_info->sectorsize;
L
Liu Bo 已提交
3045
	lockend--;
J
Josef Bacik 已提交
3046 3047
	len = lockend - lockstart + 1;

3048
	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3049
			 &cached_state);
J
Josef Bacik 已提交
3050

3051
	while (start < inode->i_size) {
3052 3053
		em = btrfs_get_extent_fiemap(BTRFS_I(inode), NULL, 0,
				start, len, 0);
J
Josef Bacik 已提交
3054
		if (IS_ERR(em)) {
3055
			ret = PTR_ERR(em);
3056
			em = NULL;
J
Josef Bacik 已提交
3057 3058 3059
			break;
		}

3060 3061 3062 3063 3064 3065 3066 3067
		if (whence == SEEK_HOLE &&
		    (em->block_start == EXTENT_MAP_HOLE ||
		     test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
			break;
		else if (whence == SEEK_DATA &&
			   (em->block_start != EXTENT_MAP_HOLE &&
			    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
			break;
J
Josef Bacik 已提交
3068 3069 3070

		start = em->start + em->len;
		free_extent_map(em);
3071
		em = NULL;
J
Josef Bacik 已提交
3072 3073
		cond_resched();
	}
3074 3075 3076 3077 3078 3079 3080
	free_extent_map(em);
	if (!ret) {
		if (whence == SEEK_DATA && start >= inode->i_size)
			ret = -ENXIO;
		else
			*offset = min_t(loff_t, start, inode->i_size);
	}
J
Josef Bacik 已提交
3081 3082 3083 3084 3085
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
			     &cached_state, GFP_NOFS);
	return ret;
}

3086
static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
J
Josef Bacik 已提交
3087 3088 3089 3090
{
	struct inode *inode = file->f_mapping->host;
	int ret;

A
Al Viro 已提交
3091
	inode_lock(inode);
3092
	switch (whence) {
J
Josef Bacik 已提交
3093 3094
	case SEEK_END:
	case SEEK_CUR:
3095
		offset = generic_file_llseek(file, offset, whence);
J
Josef Bacik 已提交
3096 3097 3098
		goto out;
	case SEEK_DATA:
	case SEEK_HOLE:
3099
		if (offset >= i_size_read(inode)) {
A
Al Viro 已提交
3100
			inode_unlock(inode);
3101 3102 3103
			return -ENXIO;
		}

3104
		ret = find_desired_extent(inode, &offset, whence);
J
Josef Bacik 已提交
3105
		if (ret) {
A
Al Viro 已提交
3106
			inode_unlock(inode);
J
Josef Bacik 已提交
3107 3108 3109 3110
			return ret;
		}
	}

J
Jie Liu 已提交
3111
	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
J
Josef Bacik 已提交
3112
out:
A
Al Viro 已提交
3113
	inode_unlock(inode);
J
Josef Bacik 已提交
3114 3115 3116
	return offset;
}

G
Goldwyn Rodrigues 已提交
3117 3118
static int btrfs_file_open(struct inode *inode, struct file *filp)
{
3119
	filp->f_mode |= FMODE_NOWAIT;
G
Goldwyn Rodrigues 已提交
3120 3121 3122
	return generic_file_open(inode, filp);
}

3123
const struct file_operations btrfs_file_operations = {
J
Josef Bacik 已提交
3124
	.llseek		= btrfs_file_llseek,
3125
	.read_iter      = generic_file_read_iter,
C
Chris Mason 已提交
3126
	.splice_read	= generic_file_splice_read,
A
Al Viro 已提交
3127
	.write_iter	= btrfs_file_write_iter,
C
Chris Mason 已提交
3128
	.mmap		= btrfs_file_mmap,
G
Goldwyn Rodrigues 已提交
3129
	.open		= btrfs_file_open,
3130
	.release	= btrfs_release_file,
C
Chris Mason 已提交
3131
	.fsync		= btrfs_sync_file,
3132
	.fallocate	= btrfs_fallocate,
C
Christoph Hellwig 已提交
3133
	.unlocked_ioctl	= btrfs_ioctl,
C
Chris Mason 已提交
3134
#ifdef CONFIG_COMPAT
3135
	.compat_ioctl	= btrfs_compat_ioctl,
C
Chris Mason 已提交
3136
#endif
3137
	.clone_file_range = btrfs_clone_file_range,
3138
	.dedupe_file_range = btrfs_dedupe_file_range,
C
Chris Mason 已提交
3139
};
3140 3141 3142

void btrfs_auto_defrag_exit(void)
{
3143
	kmem_cache_destroy(btrfs_inode_defrag_cachep);
3144 3145
}

3146
int __init btrfs_auto_defrag_init(void)
3147 3148 3149
{
	btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
					sizeof(struct inode_defrag), 0,
3150
					SLAB_MEM_SPREAD,
3151 3152 3153 3154 3155 3156
					NULL);
	if (!btrfs_inode_defrag_cachep)
		return -ENOMEM;

	return 0;
}
3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182

int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
{
	int ret;

	/*
	 * So with compression we will find and lock a dirty page and clear the
	 * first one as dirty, setup an async extent, and immediately return
	 * with the entire range locked but with nobody actually marked with
	 * writeback.  So we can't just filemap_write_and_wait_range() and
	 * expect it to work since it will just kick off a thread to do the
	 * actual work.  So we need to call filemap_fdatawrite_range _again_
	 * since it will wait on the page lock, which won't be unlocked until
	 * after the pages have been marked as writeback and so we're good to go
	 * from there.  We have to do this otherwise we'll miss the ordered
	 * extents and that results in badness.  Please Josef, do not think you
	 * know better and pull this out at some point in the future, it is
	 * right and you are wrong.
	 */
	ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
	if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
			     &BTRFS_I(inode)->runtime_flags))
		ret = filemap_fdatawrite_range(inode->i_mapping, start, end);

	return ret;
}