inode.c 207.1 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.
 */

19
#include <linux/kernel.h>
20
#include <linux/bio.h>
C
Chris Mason 已提交
21
#include <linux/buffer_head.h>
S
Sage Weil 已提交
22
#include <linux/file.h>
C
Chris Mason 已提交
23 24 25 26 27 28 29 30 31 32 33 34
#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>
#include <linux/swap.h>
#include <linux/writeback.h>
#include <linux/statfs.h>
#include <linux/compat.h>
C
Chris Mason 已提交
35
#include <linux/bit_spinlock.h>
J
Josef Bacik 已提交
36
#include <linux/xattr.h>
J
Josef Bacik 已提交
37
#include <linux/posix_acl.h>
Y
Yan Zheng 已提交
38
#include <linux/falloc.h>
39
#include <linux/slab.h>
40
#include <linux/ratelimit.h>
41
#include <linux/mount.h>
C
Chris Mason 已提交
42
#include "compat.h"
C
Chris Mason 已提交
43 44 45 46 47 48
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
#include "btrfs_inode.h"
#include "ioctl.h"
#include "print-tree.h"
49
#include "ordered-data.h"
50
#include "xattr.h"
51
#include "tree-log.h"
52
#include "volumes.h"
C
Chris Mason 已提交
53
#include "compression.h"
54
#include "locking.h"
55
#include "free-space-cache.h"
56
#include "inode-map.h"
C
Chris Mason 已提交
57 58 59 60 61 62

struct btrfs_iget_args {
	u64 ino;
	struct btrfs_root *root;
};

63 64 65 66 67
static const struct inode_operations btrfs_dir_inode_operations;
static const struct inode_operations btrfs_symlink_inode_operations;
static const struct inode_operations btrfs_dir_ro_inode_operations;
static const struct inode_operations btrfs_special_inode_operations;
static const struct inode_operations btrfs_file_inode_operations;
68 69
static const struct address_space_operations btrfs_aops;
static const struct address_space_operations btrfs_symlink_aops;
70
static const struct file_operations btrfs_dir_file_operations;
71
static struct extent_io_ops btrfs_extent_io_ops;
C
Chris Mason 已提交
72 73 74 75 76

static struct kmem_cache *btrfs_inode_cachep;
struct kmem_cache *btrfs_trans_handle_cachep;
struct kmem_cache *btrfs_transaction_cachep;
struct kmem_cache *btrfs_path_cachep;
77
struct kmem_cache *btrfs_free_space_cachep;
C
Chris Mason 已提交
78 79 80 81 82 83 84 85 86 87 88 89

#define S_SHIFT 12
static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
	[S_IFREG >> S_SHIFT]	= BTRFS_FT_REG_FILE,
	[S_IFDIR >> S_SHIFT]	= BTRFS_FT_DIR,
	[S_IFCHR >> S_SHIFT]	= BTRFS_FT_CHRDEV,
	[S_IFBLK >> S_SHIFT]	= BTRFS_FT_BLKDEV,
	[S_IFIFO >> S_SHIFT]	= BTRFS_FT_FIFO,
	[S_IFSOCK >> S_SHIFT]	= BTRFS_FT_SOCK,
	[S_IFLNK >> S_SHIFT]	= BTRFS_FT_SYMLINK,
};

90 91
static int btrfs_setsize(struct inode *inode, loff_t newsize);
static int btrfs_truncate(struct inode *inode);
92
static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 94 95 96
static noinline int cow_file_range(struct inode *inode,
				   struct page *locked_page,
				   u64 start, u64 end, int *page_started,
				   unsigned long *nr_written, int unlock);
97 98
static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
				struct btrfs_root *root, struct inode *inode);
99

100
static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 102
				     struct inode *inode,  struct inode *dir,
				     const struct qstr *qstr)
J
Jim Owens 已提交
103 104 105
{
	int err;

106
	err = btrfs_init_acl(trans, inode, dir);
J
Jim Owens 已提交
107
	if (!err)
108
		err = btrfs_xattr_security_init(trans, inode, dir, qstr);
J
Jim Owens 已提交
109 110 111
	return err;
}

C
Chris Mason 已提交
112 113 114 115 116
/*
 * this does all the hard work for inserting an inline extent into
 * the btree.  The caller should have done a btrfs_drop_extents so that
 * no overlapping inline items exist in the btree
 */
C
Chris Mason 已提交
117
static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
C
Chris Mason 已提交
118 119
				struct btrfs_root *root, struct inode *inode,
				u64 start, size_t size, size_t compressed_size,
120
				int compress_type,
C
Chris Mason 已提交
121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
				struct page **compressed_pages)
{
	struct btrfs_key key;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct page *page = NULL;
	char *kaddr;
	unsigned long ptr;
	struct btrfs_file_extent_item *ei;
	int err = 0;
	int ret;
	size_t cur_size = size;
	size_t datasize;
	unsigned long offset;

136
	if (compressed_size && compressed_pages)
C
Chris Mason 已提交
137 138
		cur_size = compressed_size;

C
Chris Mason 已提交
139 140
	path = btrfs_alloc_path();
	if (!path)
C
Chris Mason 已提交
141 142
		return -ENOMEM;

143
	path->leave_spinning = 1;
C
Chris Mason 已提交
144

L
Li Zefan 已提交
145
	key.objectid = btrfs_ino(inode);
C
Chris Mason 已提交
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
	key.offset = start;
	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
	datasize = btrfs_file_extent_calc_inline_size(cur_size);

	inode_add_bytes(inode, size);
	ret = btrfs_insert_empty_item(trans, root, path, &key,
				      datasize);
	if (ret) {
		err = ret;
		goto fail;
	}
	leaf = path->nodes[0];
	ei = btrfs_item_ptr(leaf, path->slots[0],
			    struct btrfs_file_extent_item);
	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
	btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
	btrfs_set_file_extent_encryption(leaf, ei, 0);
	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
	btrfs_set_file_extent_ram_bytes(leaf, ei, size);
	ptr = btrfs_file_extent_inline_start(ei);

167
	if (compress_type != BTRFS_COMPRESS_NONE) {
C
Chris Mason 已提交
168 169
		struct page *cpage;
		int i = 0;
C
Chris Mason 已提交
170
		while (compressed_size > 0) {
C
Chris Mason 已提交
171
			cpage = compressed_pages[i];
172
			cur_size = min_t(unsigned long, compressed_size,
C
Chris Mason 已提交
173 174
				       PAGE_CACHE_SIZE);

175
			kaddr = kmap_atomic(cpage);
C
Chris Mason 已提交
176
			write_extent_buffer(leaf, kaddr, ptr, cur_size);
177
			kunmap_atomic(kaddr);
C
Chris Mason 已提交
178 179 180 181 182 183

			i++;
			ptr += cur_size;
			compressed_size -= cur_size;
		}
		btrfs_set_file_extent_compression(leaf, ei,
184
						  compress_type);
C
Chris Mason 已提交
185 186 187 188
	} else {
		page = find_get_page(inode->i_mapping,
				     start >> PAGE_CACHE_SHIFT);
		btrfs_set_file_extent_compression(leaf, ei, 0);
189
		kaddr = kmap_atomic(page);
C
Chris Mason 已提交
190 191
		offset = start & (PAGE_CACHE_SIZE - 1);
		write_extent_buffer(leaf, kaddr + offset, ptr, size);
192
		kunmap_atomic(kaddr);
C
Chris Mason 已提交
193 194 195 196 197
		page_cache_release(page);
	}
	btrfs_mark_buffer_dirty(leaf);
	btrfs_free_path(path);

198 199 200 201 202 203 204 205 206
	/*
	 * we're an inline extent, so nobody can
	 * extend the file past i_size without locking
	 * a page we already have locked.
	 *
	 * We must do any isize and inode updates
	 * before we unlock the pages.  Otherwise we
	 * could end up racing with unlink.
	 */
C
Chris Mason 已提交
207
	BTRFS_I(inode)->disk_i_size = inode->i_size;
208
	ret = btrfs_update_inode(trans, root, inode);
209

210
	return ret;
C
Chris Mason 已提交
211 212 213 214 215 216 217 218 219 220 221
fail:
	btrfs_free_path(path);
	return err;
}


/*
 * conditionally insert an inline extent into the file.  This
 * does the checks required to make sure the data is small enough
 * to fit as an inline extent.
 */
222
static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
C
Chris Mason 已提交
223 224
				 struct btrfs_root *root,
				 struct inode *inode, u64 start, u64 end,
225
				 size_t compressed_size, int compress_type,
C
Chris Mason 已提交
226 227 228 229 230 231 232 233 234 235 236 237 238 239
				 struct page **compressed_pages)
{
	u64 isize = i_size_read(inode);
	u64 actual_end = min(end + 1, isize);
	u64 inline_len = actual_end - start;
	u64 aligned_end = (end + root->sectorsize - 1) &
			~((u64)root->sectorsize - 1);
	u64 data_len = inline_len;
	int ret;

	if (compressed_size)
		data_len = compressed_size;

	if (start > 0 ||
C
Chris Mason 已提交
240
	    actual_end >= PAGE_CACHE_SIZE ||
C
Chris Mason 已提交
241 242 243 244 245 246 247 248
	    data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
	    (!compressed_size &&
	    (actual_end & (root->sectorsize - 1)) == 0) ||
	    end + 1 < isize ||
	    data_len > root->fs_info->max_inline) {
		return 1;
	}

249
	ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
250 251
	if (ret)
		return ret;
C
Chris Mason 已提交
252 253 254 255 256

	if (isize > actual_end)
		inline_len = min_t(u64, isize, actual_end);
	ret = insert_inline_extent(trans, root, inode, start,
				   inline_len, compressed_size,
257
				   compress_type, compressed_pages);
258
	if (ret && ret != -ENOSPC) {
259 260
		btrfs_abort_transaction(trans, root, ret);
		return ret;
261 262
	} else if (ret == -ENOSPC) {
		return 1;
263
	}
264

265
	btrfs_delalloc_release_metadata(inode, end + 1 - start);
C
Chris Mason 已提交
266
	btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
C
Chris Mason 已提交
267 268 269
	return 0;
}

270 271 272 273 274 275
struct async_extent {
	u64 start;
	u64 ram_size;
	u64 compressed_size;
	struct page **pages;
	unsigned long nr_pages;
276
	int compress_type;
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293
	struct list_head list;
};

struct async_cow {
	struct inode *inode;
	struct btrfs_root *root;
	struct page *locked_page;
	u64 start;
	u64 end;
	struct list_head extents;
	struct btrfs_work work;
};

static noinline int add_async_extent(struct async_cow *cow,
				     u64 start, u64 ram_size,
				     u64 compressed_size,
				     struct page **pages,
294 295
				     unsigned long nr_pages,
				     int compress_type)
296 297 298 299
{
	struct async_extent *async_extent;

	async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
300
	BUG_ON(!async_extent); /* -ENOMEM */
301 302 303 304 305
	async_extent->start = start;
	async_extent->ram_size = ram_size;
	async_extent->compressed_size = compressed_size;
	async_extent->pages = pages;
	async_extent->nr_pages = nr_pages;
306
	async_extent->compress_type = compress_type;
307 308 309 310
	list_add_tail(&async_extent->list, &cow->extents);
	return 0;
}

C
Chris Mason 已提交
311
/*
312 313 314
 * we create compressed extents in two phases.  The first
 * phase compresses a range of pages that have already been
 * locked (both pages and state bits are locked).
C
Chris Mason 已提交
315
 *
316 317 318 319 320
 * This is done inside an ordered work queue, and the compression
 * is spread across many cpus.  The actual IO submission is step
 * two, and the ordered work queue takes care of making sure that
 * happens in the same order things were put onto the queue by
 * writepages and friends.
C
Chris Mason 已提交
321
 *
322 323 324
 * If this code finds it can't get good compression, it puts an
 * entry onto the work queue to write the uncompressed bytes.  This
 * makes sure that both compressed inodes and uncompressed inodes
325 326
 * are written in the same order that the flusher thread sent them
 * down.
C
Chris Mason 已提交
327
 */
328 329 330 331 332
static noinline int compress_file_range(struct inode *inode,
					struct page *locked_page,
					u64 start, u64 end,
					struct async_cow *async_cow,
					int *num_added)
333 334 335
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
336 337
	u64 num_bytes;
	u64 blocksize = root->sectorsize;
C
Chris Mason 已提交
338
	u64 actual_end;
339
	u64 isize = i_size_read(inode);
340
	int ret = 0;
C
Chris Mason 已提交
341 342 343 344 345 346
	struct page **pages = NULL;
	unsigned long nr_pages;
	unsigned long nr_pages_ret = 0;
	unsigned long total_compressed = 0;
	unsigned long total_in = 0;
	unsigned long max_compressed = 128 * 1024;
347
	unsigned long max_uncompressed = 128 * 1024;
C
Chris Mason 已提交
348 349
	int i;
	int will_compress;
350
	int compress_type = root->fs_info->compress_type;
351

352 353 354
	/* if this is a small write inside eof, kick off a defrag */
	if ((end - start + 1) < 16 * 1024 &&
	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
C
Chris Mason 已提交
355 356
		btrfs_add_inode_defrag(NULL, inode);

357
	actual_end = min_t(u64, isize, end + 1);
C
Chris Mason 已提交
358 359 360 361
again:
	will_compress = 0;
	nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
	nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
362

363 364 365 366 367 368 369 370 371 372 373 374 375
	/*
	 * we don't want to send crud past the end of i_size through
	 * compression, that's just a waste of CPU time.  So, if the
	 * end of the file is before the start of our current
	 * requested range of bytes, we bail out to the uncompressed
	 * cleanup code that can deal with all of this.
	 *
	 * It isn't really the fastest way to fix things, but this is a
	 * very uncommon corner.
	 */
	if (actual_end <= start)
		goto cleanup_and_bail_uncompressed;

C
Chris Mason 已提交
376 377 378 379
	total_compressed = actual_end - start;

	/* we want to make sure that amount of ram required to uncompress
	 * an extent is reasonable, so we limit the total size in ram
380 381 382 383 384 385 386
	 * of a compressed extent to 128k.  This is a crucial number
	 * because it also controls how easily we can spread reads across
	 * cpus for decompression.
	 *
	 * We also want to make sure the amount of IO required to do
	 * a random read is reasonably small, so we limit the size of
	 * a compressed extent to 128k.
C
Chris Mason 已提交
387 388
	 */
	total_compressed = min(total_compressed, max_uncompressed);
389
	num_bytes = (end - start + blocksize) & ~(blocksize - 1);
390
	num_bytes = max(blocksize,  num_bytes);
C
Chris Mason 已提交
391 392
	total_in = 0;
	ret = 0;
393

394 395 396 397
	/*
	 * we do compression for mount -o compress and when the
	 * inode has not been flagged as nocompress.  This flag can
	 * change at any time if we discover bad compression ratios.
C
Chris Mason 已提交
398
	 */
399
	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
C
Chris Mason 已提交
400
	    (btrfs_test_opt(root, COMPRESS) ||
401 402
	     (BTRFS_I(inode)->force_compress) ||
	     (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
C
Chris Mason 已提交
403
		WARN_ON(pages);
404
		pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
405 406 407 408
		if (!pages) {
			/* just bail out to the uncompressed code */
			goto cont;
		}
C
Chris Mason 已提交
409

410 411 412 413 414 415 416 417 418 419
		if (BTRFS_I(inode)->force_compress)
			compress_type = BTRFS_I(inode)->force_compress;

		ret = btrfs_compress_pages(compress_type,
					   inode->i_mapping, start,
					   total_compressed, pages,
					   nr_pages, &nr_pages_ret,
					   &total_in,
					   &total_compressed,
					   max_compressed);
C
Chris Mason 已提交
420 421 422 423 424 425 426 427 428 429 430

		if (!ret) {
			unsigned long offset = total_compressed &
				(PAGE_CACHE_SIZE - 1);
			struct page *page = pages[nr_pages_ret - 1];
			char *kaddr;

			/* zero the tail end of the last page, we might be
			 * sending it down to disk
			 */
			if (offset) {
431
				kaddr = kmap_atomic(page);
C
Chris Mason 已提交
432 433
				memset(kaddr + offset, 0,
				       PAGE_CACHE_SIZE - offset);
434
				kunmap_atomic(kaddr);
C
Chris Mason 已提交
435 436 437 438
			}
			will_compress = 1;
		}
	}
439
cont:
C
Chris Mason 已提交
440
	if (start == 0) {
441
		trans = btrfs_join_transaction(root);
442 443 444 445 446
		if (IS_ERR(trans)) {
			ret = PTR_ERR(trans);
			trans = NULL;
			goto cleanup_and_out;
		}
447
		trans->block_rsv = &root->fs_info->delalloc_block_rsv;
448

C
Chris Mason 已提交
449
		/* lets try to make an inline extent */
450
		if (ret || total_in < (actual_end - start)) {
C
Chris Mason 已提交
451
			/* we didn't compress the entire range, try
452
			 * to make an uncompressed inline extent.
C
Chris Mason 已提交
453 454
			 */
			ret = cow_file_range_inline(trans, root, inode,
455
						    start, end, 0, 0, NULL);
C
Chris Mason 已提交
456
		} else {
457
			/* try making a compressed inline extent */
C
Chris Mason 已提交
458 459
			ret = cow_file_range_inline(trans, root, inode,
						    start, end,
460 461
						    total_compressed,
						    compress_type, pages);
C
Chris Mason 已提交
462
		}
463
		if (ret <= 0) {
464
			/*
465 466 467
			 * inline extent creation worked or returned error,
			 * we don't need to create any more async work items.
			 * Unlock and free up our temp pages.
468
			 */
C
Chris Mason 已提交
469
			extent_clear_unlock_delalloc(inode,
470 471 472
			     &BTRFS_I(inode)->io_tree,
			     start, end, NULL,
			     EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
473
			     EXTENT_CLEAR_DELALLOC |
474
			     EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
475 476

			btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
477 478
			goto free_pages_out;
		}
479
		btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
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
	}

	if (will_compress) {
		/*
		 * we aren't doing an inline extent round the compressed size
		 * up to a block size boundary so the allocator does sane
		 * things
		 */
		total_compressed = (total_compressed + blocksize - 1) &
			~(blocksize - 1);

		/*
		 * one last check to make sure the compression is really a
		 * win, compare the page count read with the blocks on disk
		 */
		total_in = (total_in + PAGE_CACHE_SIZE - 1) &
			~(PAGE_CACHE_SIZE - 1);
		if (total_compressed >= total_in) {
			will_compress = 0;
		} else {
			num_bytes = total_in;
		}
	}
	if (!will_compress && pages) {
		/*
		 * the compression code ran but failed to make things smaller,
		 * free any pages it allocated and our page pointer array
		 */
		for (i = 0; i < nr_pages_ret; i++) {
C
Chris Mason 已提交
509
			WARN_ON(pages[i]->mapping);
C
Chris Mason 已提交
510 511 512 513 514 515 516 517
			page_cache_release(pages[i]);
		}
		kfree(pages);
		pages = NULL;
		total_compressed = 0;
		nr_pages_ret = 0;

		/* flag the file so we don't compress in the future */
C
Chris Mason 已提交
518 519
		if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
		    !(BTRFS_I(inode)->force_compress)) {
C
Chris Mason 已提交
520
			BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
C
Chris Mason 已提交
521
		}
C
Chris Mason 已提交
522
	}
523 524
	if (will_compress) {
		*num_added += 1;
C
Chris Mason 已提交
525

526 527 528 529 530
		/* the async work queues will take care of doing actual
		 * allocation on disk for these compressed pages,
		 * and will submit them to the elevator.
		 */
		add_async_extent(async_cow, start, num_bytes,
531 532
				 total_compressed, pages, nr_pages_ret,
				 compress_type);
533

534
		if (start + num_bytes < end) {
535 536 537 538 539 540
			start += num_bytes;
			pages = NULL;
			cond_resched();
			goto again;
		}
	} else {
541
cleanup_and_bail_uncompressed:
542 543 544 545 546 547 548 549 550 551 552 553
		/*
		 * No compression, but we still need to write the pages in
		 * the file we've been given so far.  redirty the locked
		 * page if it corresponds to our extent and set things up
		 * for the async work queue to run cow_file_range to do
		 * the normal delalloc dance
		 */
		if (page_offset(locked_page) >= start &&
		    page_offset(locked_page) <= end) {
			__set_page_dirty_nobuffers(locked_page);
			/* unlocked later on in the async handlers */
		}
554 555
		add_async_extent(async_cow, start, end - start + 1,
				 0, NULL, 0, BTRFS_COMPRESS_NONE);
556 557
		*num_added += 1;
	}
558

559
out:
560
	return ret;
561 562 563 564 565 566

free_pages_out:
	for (i = 0; i < nr_pages_ret; i++) {
		WARN_ON(pages[i]->mapping);
		page_cache_release(pages[i]);
	}
C
Chris Mason 已提交
567
	kfree(pages);
568 569

	goto out;
570 571 572 573 574 575 576 577 578 579 580 581 582 583

cleanup_and_out:
	extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
				     start, end, NULL,
				     EXTENT_CLEAR_UNLOCK_PAGE |
				     EXTENT_CLEAR_DIRTY |
				     EXTENT_CLEAR_DELALLOC |
				     EXTENT_SET_WRITEBACK |
				     EXTENT_END_WRITEBACK);
	if (!trans || IS_ERR(trans))
		btrfs_error(root->fs_info, ret, "Failed to join transaction");
	else
		btrfs_abort_transaction(trans, root, ret);
	goto free_pages_out;
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
}

/*
 * phase two of compressed writeback.  This is the ordered portion
 * of the code, which only gets called in the order the work was
 * queued.  We walk all the async extents created by compress_file_range
 * and send them down to the disk.
 */
static noinline int submit_compressed_extents(struct inode *inode,
					      struct async_cow *async_cow)
{
	struct async_extent *async_extent;
	u64 alloc_hint = 0;
	struct btrfs_trans_handle *trans;
	struct btrfs_key ins;
	struct extent_map *em;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	struct extent_io_tree *io_tree;
603
	int ret = 0;
604 605 606 607 608

	if (list_empty(&async_cow->extents))
		return 0;


C
Chris Mason 已提交
609
	while (!list_empty(&async_cow->extents)) {
610 611 612
		async_extent = list_entry(async_cow->extents.next,
					  struct async_extent, list);
		list_del(&async_extent->list);
C
Chris Mason 已提交
613

614 615
		io_tree = &BTRFS_I(inode)->io_tree;

616
retry:
617 618 619 620 621 622
		/* did the compression code fall back to uncompressed IO? */
		if (!async_extent->pages) {
			int page_started = 0;
			unsigned long nr_written = 0;

			lock_extent(io_tree, async_extent->start,
623
					 async_extent->start +
624
					 async_extent->ram_size - 1);
625 626

			/* allocate blocks */
627 628 629 630 631
			ret = cow_file_range(inode, async_cow->locked_page,
					     async_extent->start,
					     async_extent->start +
					     async_extent->ram_size - 1,
					     &page_started, &nr_written, 0);
632

633 634
			/* JDM XXX */

635 636 637 638 639 640
			/*
			 * if page_started, cow_file_range inserted an
			 * inline extent and took care of all the unlocking
			 * and IO for us.  Otherwise, we need to submit
			 * all those pages down to the drive.
			 */
641
			if (!page_started && !ret)
642 643
				extent_write_locked_range(io_tree,
						  inode, async_extent->start,
C
Chris Mason 已提交
644
						  async_extent->start +
645 646 647 648 649 650 651 652 653
						  async_extent->ram_size - 1,
						  btrfs_get_extent,
						  WB_SYNC_ALL);
			kfree(async_extent);
			cond_resched();
			continue;
		}

		lock_extent(io_tree, async_extent->start,
654
			    async_extent->start + async_extent->ram_size - 1);
655

656
		trans = btrfs_join_transaction(root);
657 658 659 660 661
		if (IS_ERR(trans)) {
			ret = PTR_ERR(trans);
		} else {
			trans->block_rsv = &root->fs_info->delalloc_block_rsv;
			ret = btrfs_reserve_extent(trans, root,
662 663
					   async_extent->compressed_size,
					   async_extent->compressed_size,
664
					   0, alloc_hint, &ins, 1);
665 666 667 668
			if (ret)
				btrfs_abort_transaction(trans, root, ret);
			btrfs_end_transaction(trans, root);
		}
669

670 671 672 673 674 675 676 677 678 679 680
		if (ret) {
			int i;
			for (i = 0; i < async_extent->nr_pages; i++) {
				WARN_ON(async_extent->pages[i]->mapping);
				page_cache_release(async_extent->pages[i]);
			}
			kfree(async_extent->pages);
			async_extent->nr_pages = 0;
			async_extent->pages = NULL;
			unlock_extent(io_tree, async_extent->start,
				      async_extent->start +
681
				      async_extent->ram_size - 1);
682 683 684
			if (ret == -ENOSPC)
				goto retry;
			goto out_free; /* JDM: Requeue? */
685 686
		}

687 688 689 690 691 692 693 694
		/*
		 * here we're doing allocation and writeback of the
		 * compressed pages
		 */
		btrfs_drop_extent_cache(inode, async_extent->start,
					async_extent->start +
					async_extent->ram_size - 1, 0);

695
		em = alloc_extent_map();
696
		BUG_ON(!em); /* -ENOMEM */
697 698
		em->start = async_extent->start;
		em->len = async_extent->ram_size;
699
		em->orig_start = em->start;
C
Chris Mason 已提交
700

701 702 703
		em->block_start = ins.objectid;
		em->block_len = ins.offset;
		em->bdev = root->fs_info->fs_devices->latest_bdev;
704
		em->compress_type = async_extent->compress_type;
705 706 707
		set_bit(EXTENT_FLAG_PINNED, &em->flags);
		set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);

C
Chris Mason 已提交
708
		while (1) {
709
			write_lock(&em_tree->lock);
710
			ret = add_extent_mapping(em_tree, em);
711
			write_unlock(&em_tree->lock);
712 713 714 715 716 717 718 719 720
			if (ret != -EEXIST) {
				free_extent_map(em);
				break;
			}
			btrfs_drop_extent_cache(inode, async_extent->start,
						async_extent->start +
						async_extent->ram_size - 1, 0);
		}

721 722 723 724 725 726 727
		ret = btrfs_add_ordered_extent_compress(inode,
						async_extent->start,
						ins.objectid,
						async_extent->ram_size,
						ins.offset,
						BTRFS_ORDERED_COMPRESSED,
						async_extent->compress_type);
728
		BUG_ON(ret); /* -ENOMEM */
729 730 731 732 733

		/*
		 * clear dirty, set writeback and unlock the pages.
		 */
		extent_clear_unlock_delalloc(inode,
734 735 736 737 738 739
				&BTRFS_I(inode)->io_tree,
				async_extent->start,
				async_extent->start +
				async_extent->ram_size - 1,
				NULL, EXTENT_CLEAR_UNLOCK_PAGE |
				EXTENT_CLEAR_UNLOCK |
740
				EXTENT_CLEAR_DELALLOC |
741
				EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
742 743

		ret = btrfs_submit_compressed_write(inode,
C
Chris Mason 已提交
744 745 746 747 748
				    async_extent->start,
				    async_extent->ram_size,
				    ins.objectid,
				    ins.offset, async_extent->pages,
				    async_extent->nr_pages);
749

750
		BUG_ON(ret); /* -ENOMEM */
751 752 753 754
		alloc_hint = ins.objectid + ins.offset;
		kfree(async_extent);
		cond_resched();
	}
755 756 757 758 759 760
	ret = 0;
out:
	return ret;
out_free:
	kfree(async_extent);
	goto out;
761 762
}

763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794
static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
				      u64 num_bytes)
{
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	struct extent_map *em;
	u64 alloc_hint = 0;

	read_lock(&em_tree->lock);
	em = search_extent_mapping(em_tree, start, num_bytes);
	if (em) {
		/*
		 * if block start isn't an actual block number then find the
		 * first block in this inode and use that as a hint.  If that
		 * block is also bogus then just don't worry about it.
		 */
		if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
			free_extent_map(em);
			em = search_extent_mapping(em_tree, 0, 0);
			if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
				alloc_hint = em->block_start;
			if (em)
				free_extent_map(em);
		} else {
			alloc_hint = em->block_start;
			free_extent_map(em);
		}
	}
	read_unlock(&em_tree->lock);

	return alloc_hint;
}

795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826
/*
 * when extent_io.c finds a delayed allocation range in the file,
 * the call backs end up in this code.  The basic idea is to
 * allocate extents on disk for the range, and create ordered data structs
 * in ram to track those extents.
 *
 * locked_page is the page that writepage had locked already.  We use
 * it to make sure we don't do extra locks or unlocks.
 *
 * *page_started is set to one if we unlock locked_page and do everything
 * required to start IO on it.  It may be clean and already done with
 * IO when we return.
 */
static noinline int cow_file_range(struct inode *inode,
				   struct page *locked_page,
				   u64 start, u64 end, int *page_started,
				   unsigned long *nr_written,
				   int unlock)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
	u64 alloc_hint = 0;
	u64 num_bytes;
	unsigned long ram_size;
	u64 disk_num_bytes;
	u64 cur_alloc_size;
	u64 blocksize = root->sectorsize;
	struct btrfs_key ins;
	struct extent_map *em;
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	int ret = 0;

827
	BUG_ON(btrfs_is_free_space_inode(inode));
828
	trans = btrfs_join_transaction(root);
829 830 831
	if (IS_ERR(trans)) {
		extent_clear_unlock_delalloc(inode,
			     &BTRFS_I(inode)->io_tree,
832
			     start, end, locked_page,
833 834 835 836 837 838 839 840
			     EXTENT_CLEAR_UNLOCK_PAGE |
			     EXTENT_CLEAR_UNLOCK |
			     EXTENT_CLEAR_DELALLOC |
			     EXTENT_CLEAR_DIRTY |
			     EXTENT_SET_WRITEBACK |
			     EXTENT_END_WRITEBACK);
		return PTR_ERR(trans);
	}
841
	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
842 843 844 845 846 847

	num_bytes = (end - start + blocksize) & ~(blocksize - 1);
	num_bytes = max(blocksize,  num_bytes);
	disk_num_bytes = num_bytes;
	ret = 0;

C
Chris Mason 已提交
848
	/* if this is a small write inside eof, kick off defrag */
849 850
	if (num_bytes < 64 * 1024 &&
	    (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
C
Chris Mason 已提交
851 852
		btrfs_add_inode_defrag(trans, inode);

853 854 855
	if (start == 0) {
		/* lets try to make an inline extent */
		ret = cow_file_range_inline(trans, root, inode,
856
					    start, end, 0, 0, NULL);
857 858
		if (ret == 0) {
			extent_clear_unlock_delalloc(inode,
859 860 861 862 863 864 865 866
				     &BTRFS_I(inode)->io_tree,
				     start, end, NULL,
				     EXTENT_CLEAR_UNLOCK_PAGE |
				     EXTENT_CLEAR_UNLOCK |
				     EXTENT_CLEAR_DELALLOC |
				     EXTENT_CLEAR_DIRTY |
				     EXTENT_SET_WRITEBACK |
				     EXTENT_END_WRITEBACK);
867

868 869 870 871
			*nr_written = *nr_written +
			     (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
			*page_started = 1;
			goto out;
872 873 874
		} else if (ret < 0) {
			btrfs_abort_transaction(trans, root, ret);
			goto out_unlock;
875 876 877 878
		}
	}

	BUG_ON(disk_num_bytes >
879
	       btrfs_super_total_bytes(root->fs_info->super_copy));
880

881
	alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
882 883
	btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);

C
Chris Mason 已提交
884
	while (disk_num_bytes > 0) {
885 886
		unsigned long op;

887
		cur_alloc_size = disk_num_bytes;
888
		ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
889
					   root->sectorsize, 0, alloc_hint,
890
					   &ins, 1);
891 892 893 894
		if (ret < 0) {
			btrfs_abort_transaction(trans, root, ret);
			goto out_unlock;
		}
C
Chris Mason 已提交
895

896
		em = alloc_extent_map();
897
		BUG_ON(!em); /* -ENOMEM */
898
		em->start = start;
899
		em->orig_start = em->start;
900 901
		ram_size = ins.offset;
		em->len = ins.offset;
C
Chris Mason 已提交
902

903
		em->block_start = ins.objectid;
C
Chris Mason 已提交
904
		em->block_len = ins.offset;
905
		em->bdev = root->fs_info->fs_devices->latest_bdev;
906
		set_bit(EXTENT_FLAG_PINNED, &em->flags);
C
Chris Mason 已提交
907

C
Chris Mason 已提交
908
		while (1) {
909
			write_lock(&em_tree->lock);
910
			ret = add_extent_mapping(em_tree, em);
911
			write_unlock(&em_tree->lock);
912 913 914 915 916
			if (ret != -EEXIST) {
				free_extent_map(em);
				break;
			}
			btrfs_drop_extent_cache(inode, start,
C
Chris Mason 已提交
917
						start + ram_size - 1, 0);
918 919
		}

920
		cur_alloc_size = ins.offset;
921
		ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
922
					       ram_size, cur_alloc_size, 0);
923
		BUG_ON(ret); /* -ENOMEM */
C
Chris Mason 已提交
924

925 926 927 928
		if (root->root_key.objectid ==
		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
			ret = btrfs_reloc_clone_csums(inode, start,
						      cur_alloc_size);
929 930 931 932
			if (ret) {
				btrfs_abort_transaction(trans, root, ret);
				goto out_unlock;
			}
933 934
		}

C
Chris Mason 已提交
935
		if (disk_num_bytes < cur_alloc_size)
936
			break;
C
Chris Mason 已提交
937

C
Chris Mason 已提交
938 939 940
		/* we're not doing compressed IO, don't unlock the first
		 * page (which the caller expects to stay locked), don't
		 * clear any dirty bits and don't set any writeback bits
941 942 943
		 *
		 * Do set the Private2 bit so we know this page was properly
		 * setup for writepage
C
Chris Mason 已提交
944
		 */
945 946 947 948
		op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
		op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
			EXTENT_SET_PRIVATE2;

C
Chris Mason 已提交
949 950
		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
					     start, start + ram_size - 1,
951
					     locked_page, op);
C
Chris Mason 已提交
952
		disk_num_bytes -= cur_alloc_size;
953 954 955
		num_bytes -= cur_alloc_size;
		alloc_hint = ins.objectid + ins.offset;
		start += cur_alloc_size;
956
	}
957
	ret = 0;
958
out:
959
	btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
960

961
	return ret;
962 963 964
out_unlock:
	extent_clear_unlock_delalloc(inode,
		     &BTRFS_I(inode)->io_tree,
965
		     start, end, locked_page,
966 967 968 969 970 971 972 973
		     EXTENT_CLEAR_UNLOCK_PAGE |
		     EXTENT_CLEAR_UNLOCK |
		     EXTENT_CLEAR_DELALLOC |
		     EXTENT_CLEAR_DIRTY |
		     EXTENT_SET_WRITEBACK |
		     EXTENT_END_WRITEBACK);

	goto out;
974
}
C
Chris Mason 已提交
975

976 977 978 979 980 981 982 983 984 985 986 987
/*
 * work queue call back to started compression on a file and pages
 */
static noinline void async_cow_start(struct btrfs_work *work)
{
	struct async_cow *async_cow;
	int num_added = 0;
	async_cow = container_of(work, struct async_cow, work);

	compress_file_range(async_cow->inode, async_cow->locked_page,
			    async_cow->start, async_cow->end, async_cow,
			    &num_added);
988
	if (num_added == 0) {
989
		btrfs_add_delayed_iput(async_cow->inode);
990
		async_cow->inode = NULL;
991
	}
992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008
}

/*
 * work queue call back to submit previously compressed pages
 */
static noinline void async_cow_submit(struct btrfs_work *work)
{
	struct async_cow *async_cow;
	struct btrfs_root *root;
	unsigned long nr_pages;

	async_cow = container_of(work, struct async_cow, work);

	root = async_cow->root;
	nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
		PAGE_CACHE_SHIFT;

1009
	if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1010
	    5 * 1024 * 1024 &&
1011 1012 1013
	    waitqueue_active(&root->fs_info->async_submit_wait))
		wake_up(&root->fs_info->async_submit_wait);

C
Chris Mason 已提交
1014
	if (async_cow->inode)
1015 1016
		submit_compressed_extents(async_cow->inode, async_cow);
}
C
Chris Mason 已提交
1017

1018 1019 1020 1021
static noinline void async_cow_free(struct btrfs_work *work)
{
	struct async_cow *async_cow;
	async_cow = container_of(work, struct async_cow, work);
1022
	if (async_cow->inode)
1023
		btrfs_add_delayed_iput(async_cow->inode);
1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
	kfree(async_cow);
}

static int cow_file_range_async(struct inode *inode, struct page *locked_page,
				u64 start, u64 end, int *page_started,
				unsigned long *nr_written)
{
	struct async_cow *async_cow;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	unsigned long nr_pages;
	u64 cur_end;
1035
	int limit = 10 * 1024 * 1024;
1036

1037 1038
	clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
			 1, 0, NULL, GFP_NOFS);
C
Chris Mason 已提交
1039
	while (start < end) {
1040
		async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1041
		BUG_ON(!async_cow); /* -ENOMEM */
1042
		async_cow->inode = igrab(inode);
1043 1044 1045 1046
		async_cow->root = root;
		async_cow->locked_page = locked_page;
		async_cow->start = start;

1047
		if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
			cur_end = end;
		else
			cur_end = min(end, start + 512 * 1024 - 1);

		async_cow->end = cur_end;
		INIT_LIST_HEAD(&async_cow->extents);

		async_cow->work.func = async_cow_start;
		async_cow->work.ordered_func = async_cow_submit;
		async_cow->work.ordered_free = async_cow_free;
		async_cow->work.flags = 0;

		nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
			PAGE_CACHE_SHIFT;
		atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);

		btrfs_queue_worker(&root->fs_info->delalloc_workers,
				   &async_cow->work);

		if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
			wait_event(root->fs_info->async_submit_wait,
			   (atomic_read(&root->fs_info->async_delalloc_pages) <
			    limit));
		}

C
Chris Mason 已提交
1073
		while (atomic_read(&root->fs_info->async_submit_draining) &&
1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
		      atomic_read(&root->fs_info->async_delalloc_pages)) {
			wait_event(root->fs_info->async_submit_wait,
			  (atomic_read(&root->fs_info->async_delalloc_pages) ==
			   0));
		}

		*nr_written += nr_pages;
		start = cur_end + 1;
	}
	*page_started = 1;
	return 0;
1085 1086
}

C
Chris Mason 已提交
1087
static noinline int csum_exist_in_range(struct btrfs_root *root,
1088 1089 1090 1091 1092 1093
					u64 bytenr, u64 num_bytes)
{
	int ret;
	struct btrfs_ordered_sum *sums;
	LIST_HEAD(list);

Y
Yan Zheng 已提交
1094
	ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
A
Arne Jansen 已提交
1095
				       bytenr + num_bytes - 1, &list, 0);
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106
	if (ret == 0 && list_empty(&list))
		return 0;

	while (!list_empty(&list)) {
		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
		list_del(&sums->list);
		kfree(sums);
	}
	return 1;
}

C
Chris Mason 已提交
1107 1108 1109 1110 1111 1112 1113
/*
 * when nowcow writeback call back.  This checks for snapshots or COW copies
 * of the extents that exist in the file, and COWs the file as required.
 *
 * If no cow copies or snapshots exist, we write directly to the existing
 * blocks on disk
 */
1114 1115
static noinline int run_delalloc_nocow(struct inode *inode,
				       struct page *locked_page,
1116 1117
			      u64 start, u64 end, int *page_started, int force,
			      unsigned long *nr_written)
1118 1119
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
1120
	struct btrfs_trans_handle *trans;
1121 1122
	struct extent_buffer *leaf;
	struct btrfs_path *path;
Y
Yan Zheng 已提交
1123
	struct btrfs_file_extent_item *fi;
1124
	struct btrfs_key found_key;
Y
Yan Zheng 已提交
1125 1126 1127
	u64 cow_start;
	u64 cur_offset;
	u64 extent_end;
1128
	u64 extent_offset;
Y
Yan Zheng 已提交
1129 1130 1131
	u64 disk_bytenr;
	u64 num_bytes;
	int extent_type;
1132
	int ret, err;
Y
Yan Zheng 已提交
1133
	int type;
Y
Yan Zheng 已提交
1134 1135
	int nocow;
	int check_prev = 1;
1136
	bool nolock;
L
Li Zefan 已提交
1137
	u64 ino = btrfs_ino(inode);
1138 1139

	path = btrfs_alloc_path();
1140 1141 1142 1143 1144 1145 1146 1147 1148 1149
	if (!path) {
		extent_clear_unlock_delalloc(inode,
			     &BTRFS_I(inode)->io_tree,
			     start, end, locked_page,
			     EXTENT_CLEAR_UNLOCK_PAGE |
			     EXTENT_CLEAR_UNLOCK |
			     EXTENT_CLEAR_DELALLOC |
			     EXTENT_CLEAR_DIRTY |
			     EXTENT_SET_WRITEBACK |
			     EXTENT_END_WRITEBACK);
1150
		return -ENOMEM;
1151
	}
1152

1153
	nolock = btrfs_is_free_space_inode(inode);
1154 1155

	if (nolock)
1156
		trans = btrfs_join_transaction_nolock(root);
1157
	else
1158
		trans = btrfs_join_transaction(root);
C
Chris Mason 已提交
1159

1160
	if (IS_ERR(trans)) {
1161 1162 1163 1164 1165 1166 1167 1168 1169
		extent_clear_unlock_delalloc(inode,
			     &BTRFS_I(inode)->io_tree,
			     start, end, locked_page,
			     EXTENT_CLEAR_UNLOCK_PAGE |
			     EXTENT_CLEAR_UNLOCK |
			     EXTENT_CLEAR_DELALLOC |
			     EXTENT_CLEAR_DIRTY |
			     EXTENT_SET_WRITEBACK |
			     EXTENT_END_WRITEBACK);
1170 1171 1172 1173
		btrfs_free_path(path);
		return PTR_ERR(trans);
	}

1174
	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1175

Y
Yan Zheng 已提交
1176 1177 1178
	cow_start = (u64)-1;
	cur_offset = start;
	while (1) {
L
Li Zefan 已提交
1179
		ret = btrfs_lookup_file_extent(trans, root, path, ino,
Y
Yan Zheng 已提交
1180
					       cur_offset, 0);
1181 1182 1183 1184
		if (ret < 0) {
			btrfs_abort_transaction(trans, root, ret);
			goto error;
		}
Y
Yan Zheng 已提交
1185 1186 1187 1188
		if (ret > 0 && path->slots[0] > 0 && check_prev) {
			leaf = path->nodes[0];
			btrfs_item_key_to_cpu(leaf, &found_key,
					      path->slots[0] - 1);
L
Li Zefan 已提交
1189
			if (found_key.objectid == ino &&
Y
Yan Zheng 已提交
1190 1191 1192 1193 1194 1195 1196 1197
			    found_key.type == BTRFS_EXTENT_DATA_KEY)
				path->slots[0]--;
		}
		check_prev = 0;
next_slot:
		leaf = path->nodes[0];
		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
			ret = btrfs_next_leaf(root, path);
1198 1199 1200 1201
			if (ret < 0) {
				btrfs_abort_transaction(trans, root, ret);
				goto error;
			}
Y
Yan Zheng 已提交
1202 1203 1204 1205
			if (ret > 0)
				break;
			leaf = path->nodes[0];
		}
1206

Y
Yan Zheng 已提交
1207 1208
		nocow = 0;
		disk_bytenr = 0;
1209
		num_bytes = 0;
Y
Yan Zheng 已提交
1210 1211
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);

L
Li Zefan 已提交
1212
		if (found_key.objectid > ino ||
Y
Yan Zheng 已提交
1213 1214 1215 1216 1217 1218
		    found_key.type > BTRFS_EXTENT_DATA_KEY ||
		    found_key.offset > end)
			break;

		if (found_key.offset > cur_offset) {
			extent_end = found_key.offset;
1219
			extent_type = 0;
Y
Yan Zheng 已提交
1220 1221 1222 1223 1224 1225 1226
			goto out_check;
		}

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

Y
Yan Zheng 已提交
1227 1228
		if (extent_type == BTRFS_FILE_EXTENT_REG ||
		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
Y
Yan Zheng 已提交
1229
			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1230
			extent_offset = btrfs_file_extent_offset(leaf, fi);
Y
Yan Zheng 已提交
1231 1232 1233 1234 1235 1236
			extent_end = found_key.offset +
				btrfs_file_extent_num_bytes(leaf, fi);
			if (extent_end <= start) {
				path->slots[0]++;
				goto next_slot;
			}
1237 1238
			if (disk_bytenr == 0)
				goto out_check;
Y
Yan Zheng 已提交
1239 1240 1241 1242
			if (btrfs_file_extent_compression(leaf, fi) ||
			    btrfs_file_extent_encryption(leaf, fi) ||
			    btrfs_file_extent_other_encoding(leaf, fi))
				goto out_check;
Y
Yan Zheng 已提交
1243 1244
			if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
				goto out_check;
1245
			if (btrfs_extent_readonly(root, disk_bytenr))
Y
Yan Zheng 已提交
1246
				goto out_check;
L
Li Zefan 已提交
1247
			if (btrfs_cross_ref_exist(trans, root, ino,
1248 1249
						  found_key.offset -
						  extent_offset, disk_bytenr))
1250
				goto out_check;
1251
			disk_bytenr += extent_offset;
1252 1253 1254 1255 1256 1257 1258 1259 1260
			disk_bytenr += cur_offset - found_key.offset;
			num_bytes = min(end + 1, extent_end) - cur_offset;
			/*
			 * force cow if csum exists in the range.
			 * this ensure that csum for a given extent are
			 * either valid or do not exist.
			 */
			if (csum_exist_in_range(root, disk_bytenr, num_bytes))
				goto out_check;
Y
Yan Zheng 已提交
1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281
			nocow = 1;
		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
			extent_end = found_key.offset +
				btrfs_file_extent_inline_len(leaf, fi);
			extent_end = ALIGN(extent_end, root->sectorsize);
		} else {
			BUG_ON(1);
		}
out_check:
		if (extent_end <= start) {
			path->slots[0]++;
			goto next_slot;
		}
		if (!nocow) {
			if (cow_start == (u64)-1)
				cow_start = cur_offset;
			cur_offset = extent_end;
			if (cur_offset > end)
				break;
			path->slots[0]++;
			goto next_slot;
1282 1283
		}

1284
		btrfs_release_path(path);
Y
Yan Zheng 已提交
1285 1286
		if (cow_start != (u64)-1) {
			ret = cow_file_range(inode, locked_page, cow_start,
1287 1288
					found_key.offset - 1, page_started,
					nr_written, 1);
1289 1290 1291 1292
			if (ret) {
				btrfs_abort_transaction(trans, root, ret);
				goto error;
			}
Y
Yan Zheng 已提交
1293
			cow_start = (u64)-1;
1294
		}
Y
Yan Zheng 已提交
1295

Y
Yan Zheng 已提交
1296 1297 1298 1299
		if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
			struct extent_map *em;
			struct extent_map_tree *em_tree;
			em_tree = &BTRFS_I(inode)->extent_tree;
1300
			em = alloc_extent_map();
1301
			BUG_ON(!em); /* -ENOMEM */
Y
Yan Zheng 已提交
1302
			em->start = cur_offset;
1303
			em->orig_start = em->start;
Y
Yan Zheng 已提交
1304 1305 1306 1307 1308
			em->len = num_bytes;
			em->block_len = num_bytes;
			em->block_start = disk_bytenr;
			em->bdev = root->fs_info->fs_devices->latest_bdev;
			set_bit(EXTENT_FLAG_PINNED, &em->flags);
1309
			set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
Y
Yan Zheng 已提交
1310
			while (1) {
1311
				write_lock(&em_tree->lock);
Y
Yan Zheng 已提交
1312
				ret = add_extent_mapping(em_tree, em);
1313
				write_unlock(&em_tree->lock);
Y
Yan Zheng 已提交
1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
				if (ret != -EEXIST) {
					free_extent_map(em);
					break;
				}
				btrfs_drop_extent_cache(inode, em->start,
						em->start + em->len - 1, 0);
			}
			type = BTRFS_ORDERED_PREALLOC;
		} else {
			type = BTRFS_ORDERED_NOCOW;
		}
Y
Yan Zheng 已提交
1325 1326

		ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
Y
Yan Zheng 已提交
1327
					       num_bytes, num_bytes, type);
1328
		BUG_ON(ret); /* -ENOMEM */
1329

1330 1331 1332 1333
		if (root->root_key.objectid ==
		    BTRFS_DATA_RELOC_TREE_OBJECTID) {
			ret = btrfs_reloc_clone_csums(inode, cur_offset,
						      num_bytes);
1334 1335 1336 1337
			if (ret) {
				btrfs_abort_transaction(trans, root, ret);
				goto error;
			}
1338 1339
		}

Y
Yan Zheng 已提交
1340
		extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1341 1342 1343 1344
				cur_offset, cur_offset + num_bytes - 1,
				locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
				EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
				EXTENT_SET_PRIVATE2);
Y
Yan Zheng 已提交
1345 1346 1347
		cur_offset = extent_end;
		if (cur_offset > end)
			break;
1348
	}
1349
	btrfs_release_path(path);
Y
Yan Zheng 已提交
1350

1351
	if (cur_offset <= end && cow_start == (u64)-1) {
Y
Yan Zheng 已提交
1352
		cow_start = cur_offset;
1353 1354 1355
		cur_offset = end;
	}

Y
Yan Zheng 已提交
1356 1357
	if (cow_start != (u64)-1) {
		ret = cow_file_range(inode, locked_page, cow_start, end,
1358
				     page_started, nr_written, 1);
1359 1360 1361 1362
		if (ret) {
			btrfs_abort_transaction(trans, root, ret);
			goto error;
		}
Y
Yan Zheng 已提交
1363 1364
	}

1365
error:
J
Josef Bacik 已提交
1366
	if (nolock) {
1367
		err = btrfs_end_transaction_nolock(trans, root);
J
Josef Bacik 已提交
1368
	} else {
1369
		err = btrfs_end_transaction(trans, root);
J
Josef Bacik 已提交
1370
	}
1371 1372 1373
	if (!ret)
		ret = err;

1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384
	if (ret && cur_offset < end)
		extent_clear_unlock_delalloc(inode,
			     &BTRFS_I(inode)->io_tree,
			     cur_offset, end, locked_page,
			     EXTENT_CLEAR_UNLOCK_PAGE |
			     EXTENT_CLEAR_UNLOCK |
			     EXTENT_CLEAR_DELALLOC |
			     EXTENT_CLEAR_DIRTY |
			     EXTENT_SET_WRITEBACK |
			     EXTENT_END_WRITEBACK);

1385
	btrfs_free_path(path);
1386
	return ret;
1387 1388
}

C
Chris Mason 已提交
1389 1390 1391
/*
 * extent_io.c call back to do delayed allocation processing
 */
C
Chris Mason 已提交
1392
static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1393 1394
			      u64 start, u64 end, int *page_started,
			      unsigned long *nr_written)
1395 1396
{
	int ret;
1397
	struct btrfs_root *root = BTRFS_I(inode)->root;
1398

1399
	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
C
Chris Mason 已提交
1400
		ret = run_delalloc_nocow(inode, locked_page, start, end,
C
Chris Mason 已提交
1401
					 page_started, 1, nr_written);
1402
	} else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
Y
Yan Zheng 已提交
1403
		ret = run_delalloc_nocow(inode, locked_page, start, end,
C
Chris Mason 已提交
1404
					 page_started, 0, nr_written);
1405 1406 1407
	} else if (!btrfs_test_opt(root, COMPRESS) &&
		   !(BTRFS_I(inode)->force_compress) &&
		   !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1408 1409
		ret = cow_file_range(inode, locked_page, start, end,
				      page_started, nr_written, 1);
1410 1411 1412
	} else {
		set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
			&BTRFS_I(inode)->runtime_flags);
1413
		ret = cow_file_range_async(inode, locked_page, start, end,
C
Chris Mason 已提交
1414
					   page_started, nr_written);
1415
	}
1416 1417 1418
	return ret;
}

1419 1420
static void btrfs_split_extent_hook(struct inode *inode,
				    struct extent_state *orig, u64 split)
J
Josef Bacik 已提交
1421
{
1422
	/* not delalloc, ignore it */
J
Josef Bacik 已提交
1423
	if (!(orig->state & EXTENT_DELALLOC))
1424
		return;
J
Josef Bacik 已提交
1425

1426 1427 1428
	spin_lock(&BTRFS_I(inode)->lock);
	BTRFS_I(inode)->outstanding_extents++;
	spin_unlock(&BTRFS_I(inode)->lock);
J
Josef Bacik 已提交
1429 1430 1431 1432 1433 1434 1435 1436
}

/*
 * extent_io.c merge_extent_hook, used to track merged delayed allocation
 * extents so we can keep track of new extents that are just merged onto old
 * extents, such as when we are doing sequential writes, so we can properly
 * account for the metadata space we'll need.
 */
1437 1438 1439
static void btrfs_merge_extent_hook(struct inode *inode,
				    struct extent_state *new,
				    struct extent_state *other)
J
Josef Bacik 已提交
1440 1441 1442
{
	/* not delalloc, ignore it */
	if (!(other->state & EXTENT_DELALLOC))
1443
		return;
J
Josef Bacik 已提交
1444

1445 1446 1447
	spin_lock(&BTRFS_I(inode)->lock);
	BTRFS_I(inode)->outstanding_extents--;
	spin_unlock(&BTRFS_I(inode)->lock);
J
Josef Bacik 已提交
1448 1449
}

C
Chris Mason 已提交
1450 1451 1452 1453 1454
/*
 * extent_io.c set_bit_hook, used to track delayed allocation
 * bytes in this file, and to maintain the list of inodes that
 * have pending delalloc work to be done.
 */
1455 1456
static void btrfs_set_bit_hook(struct inode *inode,
			       struct extent_state *state, int *bits)
1457
{
J
Josef Bacik 已提交
1458

1459 1460
	/*
	 * set_bit and clear bit hooks normally require _irqsave/restore
1461
	 * but in this case, we are only testing for the DELALLOC
1462 1463
	 * bit, which is only set or cleared with irqs on
	 */
1464
	if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1465
		struct btrfs_root *root = BTRFS_I(inode)->root;
1466
		u64 len = state->end + 1 - state->start;
1467
		bool do_list = !btrfs_is_free_space_inode(inode);
J
Josef Bacik 已提交
1468

1469
		if (*bits & EXTENT_FIRST_DELALLOC) {
1470
			*bits &= ~EXTENT_FIRST_DELALLOC;
1471 1472 1473 1474 1475
		} else {
			spin_lock(&BTRFS_I(inode)->lock);
			BTRFS_I(inode)->outstanding_extents++;
			spin_unlock(&BTRFS_I(inode)->lock);
		}
1476

1477
		spin_lock(&root->fs_info->delalloc_lock);
1478 1479
		BTRFS_I(inode)->delalloc_bytes += len;
		root->fs_info->delalloc_bytes += len;
J
Josef Bacik 已提交
1480
		if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1481 1482 1483
			list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
				      &root->fs_info->delalloc_inodes);
		}
1484
		spin_unlock(&root->fs_info->delalloc_lock);
1485 1486 1487
	}
}

C
Chris Mason 已提交
1488 1489 1490
/*
 * extent_io.c clear_bit_hook, see set_bit_hook for why
 */
1491 1492
static void btrfs_clear_bit_hook(struct inode *inode,
				 struct extent_state *state, int *bits)
1493
{
1494 1495
	/*
	 * set_bit and clear bit hooks normally require _irqsave/restore
1496
	 * but in this case, we are only testing for the DELALLOC
1497 1498
	 * bit, which is only set or cleared with irqs on
	 */
1499
	if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1500
		struct btrfs_root *root = BTRFS_I(inode)->root;
1501
		u64 len = state->end + 1 - state->start;
1502
		bool do_list = !btrfs_is_free_space_inode(inode);
1503

1504
		if (*bits & EXTENT_FIRST_DELALLOC) {
1505
			*bits &= ~EXTENT_FIRST_DELALLOC;
1506 1507 1508 1509 1510
		} else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
			spin_lock(&BTRFS_I(inode)->lock);
			BTRFS_I(inode)->outstanding_extents--;
			spin_unlock(&BTRFS_I(inode)->lock);
		}
1511 1512 1513 1514

		if (*bits & EXTENT_DO_ACCOUNTING)
			btrfs_delalloc_release_metadata(inode, len);

J
Josef Bacik 已提交
1515 1516
		if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
		    && do_list)
1517
			btrfs_free_reserved_data_space(inode, len);
J
Josef Bacik 已提交
1518

1519
		spin_lock(&root->fs_info->delalloc_lock);
1520 1521 1522
		root->fs_info->delalloc_bytes -= len;
		BTRFS_I(inode)->delalloc_bytes -= len;

J
Josef Bacik 已提交
1523
		if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1524 1525 1526
		    !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
			list_del_init(&BTRFS_I(inode)->delalloc_inodes);
		}
1527
		spin_unlock(&root->fs_info->delalloc_lock);
1528 1529 1530
	}
}

C
Chris Mason 已提交
1531 1532 1533 1534
/*
 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
 * we don't create bios that span stripes or chunks
 */
1535
int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
C
Chris Mason 已提交
1536 1537
			 size_t size, struct bio *bio,
			 unsigned long bio_flags)
1538 1539 1540
{
	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
	struct btrfs_mapping_tree *map_tree;
1541
	u64 logical = (u64)bio->bi_sector << 9;
1542 1543 1544 1545
	u64 length = 0;
	u64 map_length;
	int ret;

1546 1547 1548
	if (bio_flags & EXTENT_BIO_COMPRESSED)
		return 0;

1549
	length = bio->bi_size;
1550 1551
	map_tree = &root->fs_info->mapping_tree;
	map_length = length;
1552
	ret = btrfs_map_block(map_tree, READ, logical,
1553
			      &map_length, NULL, 0);
1554 1555
	/* Will always return 0 or 1 with map_multi == NULL */
	BUG_ON(ret < 0);
C
Chris Mason 已提交
1556
	if (map_length < length + size)
1557
		return 1;
1558
	return 0;
1559 1560
}

C
Chris Mason 已提交
1561 1562 1563 1564 1565 1566 1567 1568
/*
 * in order to insert checksums into the metadata in large chunks,
 * we wait until bio submission time.   All the pages in the bio are
 * checksummed and sums are attached onto the ordered extent record.
 *
 * At IO completion time the cums attached on the ordered extent record
 * are inserted into the btree
 */
C
Chris Mason 已提交
1569 1570
static int __btrfs_submit_bio_start(struct inode *inode, int rw,
				    struct bio *bio, int mirror_num,
1571 1572
				    unsigned long bio_flags,
				    u64 bio_offset)
1573 1574 1575
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	int ret = 0;
1576

1577
	ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1578
	BUG_ON(ret); /* -ENOMEM */
C
Chris Mason 已提交
1579 1580
	return 0;
}
1581

C
Chris Mason 已提交
1582 1583 1584 1585 1586 1587 1588 1589
/*
 * in order to insert checksums into the metadata in large chunks,
 * we wait until bio submission time.   All the pages in the bio are
 * checksummed and sums are attached onto the ordered extent record.
 *
 * At IO completion time the cums attached on the ordered extent record
 * are inserted into the btree
 */
1590
static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1591 1592
			  int mirror_num, unsigned long bio_flags,
			  u64 bio_offset)
C
Chris Mason 已提交
1593 1594
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
1595
	return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1596 1597
}

C
Chris Mason 已提交
1598
/*
1599 1600
 * extent_io.c submission hook. This does the right thing for csum calculation
 * on write, or reading the csums from the tree before a read
C
Chris Mason 已提交
1601
 */
1602
static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1603 1604
			  int mirror_num, unsigned long bio_flags,
			  u64 bio_offset)
1605 1606 1607
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	int ret = 0;
1608
	int skip_sum;
1609
	int metadata = 0;
1610

1611
	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1612

1613
	if (btrfs_is_free_space_inode(inode))
1614 1615
		metadata = 2;

1616
	if (!(rw & REQ_WRITE)) {
1617 1618 1619 1620
		ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
		if (ret)
			return ret;

1621
		if (bio_flags & EXTENT_BIO_COMPRESSED) {
C
Chris Mason 已提交
1622 1623
			return btrfs_submit_compressed_read(inode, bio,
						    mirror_num, bio_flags);
1624 1625 1626 1627 1628
		} else if (!skip_sum) {
			ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
			if (ret)
				return ret;
		}
1629
		goto mapit;
1630
	} else if (!skip_sum) {
1631 1632 1633
		/* csum items have already been cloned */
		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
			goto mapit;
1634 1635
		/* we're doing a write, do the async checksumming */
		return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1636
				   inode, rw, bio, mirror_num,
1637 1638
				   bio_flags, bio_offset,
				   __btrfs_submit_bio_start,
C
Chris Mason 已提交
1639
				   __btrfs_submit_bio_done);
1640 1641
	}

1642
mapit:
1643
	return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1644
}
C
Chris Mason 已提交
1645

C
Chris Mason 已提交
1646 1647 1648 1649
/*
 * given a list of ordered sums record them in the inode.  This happens
 * at IO completion time based on sums calculated at bio submission time.
 */
1650
static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1651 1652 1653 1654 1655
			     struct inode *inode, u64 file_offset,
			     struct list_head *list)
{
	struct btrfs_ordered_sum *sum;

Q
Qinghuang Feng 已提交
1656
	list_for_each_entry(sum, list, list) {
1657 1658
		btrfs_csum_file_blocks(trans,
		       BTRFS_I(inode)->root->fs_info->csum_root, sum);
1659 1660 1661 1662
	}
	return 0;
}

1663 1664
int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
			      struct extent_state **cached_state)
1665
{
C
Chris Mason 已提交
1666
	if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1667
		WARN_ON(1);
1668
	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1669
				   cached_state, GFP_NOFS);
1670 1671
}

C
Chris Mason 已提交
1672
/* see btrfs_writepage_start_hook for details on why this is required */
1673 1674 1675 1676 1677
struct btrfs_writepage_fixup {
	struct page *page;
	struct btrfs_work work;
};

1678
static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1679 1680 1681
{
	struct btrfs_writepage_fixup *fixup;
	struct btrfs_ordered_extent *ordered;
1682
	struct extent_state *cached_state = NULL;
1683 1684 1685 1686
	struct page *page;
	struct inode *inode;
	u64 page_start;
	u64 page_end;
1687
	int ret;
1688 1689 1690

	fixup = container_of(work, struct btrfs_writepage_fixup, work);
	page = fixup->page;
C
Chris Mason 已提交
1691
again:
1692 1693 1694 1695 1696 1697 1698 1699 1700 1701
	lock_page(page);
	if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
		ClearPageChecked(page);
		goto out_page;
	}

	inode = page->mapping->host;
	page_start = page_offset(page);
	page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;

1702
	lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1703
			 &cached_state);
C
Chris Mason 已提交
1704 1705

	/* already ordered? We're done */
1706
	if (PagePrivate2(page))
1707
		goto out;
C
Chris Mason 已提交
1708 1709 1710

	ordered = btrfs_lookup_ordered_extent(inode, page_start);
	if (ordered) {
1711 1712
		unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
				     page_end, &cached_state, GFP_NOFS);
C
Chris Mason 已提交
1713 1714
		unlock_page(page);
		btrfs_start_ordered_extent(inode, ordered, 1);
1715
		btrfs_put_ordered_extent(ordered);
C
Chris Mason 已提交
1716 1717
		goto again;
	}
1718

1719 1720 1721 1722 1723 1724 1725 1726
	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
	if (ret) {
		mapping_set_error(page->mapping, ret);
		end_extent_writepage(page, ret, page_start, page_end);
		ClearPageChecked(page);
		goto out;
	 }

1727
	btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1728
	ClearPageChecked(page);
1729
	set_page_dirty(page);
1730
out:
1731 1732
	unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
			     &cached_state, GFP_NOFS);
1733 1734 1735
out_page:
	unlock_page(page);
	page_cache_release(page);
1736
	kfree(fixup);
1737 1738 1739 1740 1741 1742 1743 1744
}

/*
 * There are a few paths in the higher layers of the kernel that directly
 * set the page dirty bit without asking the filesystem if it is a
 * good idea.  This causes problems because we want to make sure COW
 * properly happens and the data=ordered rules are followed.
 *
C
Chris Mason 已提交
1745
 * In our case any range that doesn't have the ORDERED bit set
1746 1747 1748 1749
 * hasn't been properly setup for IO.  We kick off an async process
 * to fix it up.  The async helper will wait for ordered extents, set
 * the delalloc bit and make it safe to write the page.
 */
1750
static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1751 1752 1753 1754 1755
{
	struct inode *inode = page->mapping->host;
	struct btrfs_writepage_fixup *fixup;
	struct btrfs_root *root = BTRFS_I(inode)->root;

1756 1757
	/* this page is properly in the ordered list */
	if (TestClearPagePrivate2(page))
1758 1759 1760 1761 1762 1763 1764 1765
		return 0;

	if (PageChecked(page))
		return -EAGAIN;

	fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
	if (!fixup)
		return -EAGAIN;
1766

1767 1768 1769 1770 1771
	SetPageChecked(page);
	page_cache_get(page);
	fixup->work.func = btrfs_writepage_fixup_worker;
	fixup->page = page;
	btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1772
	return -EBUSY;
1773 1774
}

Y
Yan Zheng 已提交
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789
static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
				       struct inode *inode, u64 file_pos,
				       u64 disk_bytenr, u64 disk_num_bytes,
				       u64 num_bytes, u64 ram_bytes,
				       u8 compression, u8 encryption,
				       u16 other_encoding, int extent_type)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_file_extent_item *fi;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_key ins;
	int ret;

	path = btrfs_alloc_path();
1790 1791
	if (!path)
		return -ENOMEM;
Y
Yan Zheng 已提交
1792

1793
	path->leave_spinning = 1;
C
Chris Mason 已提交
1794 1795 1796 1797 1798 1799 1800 1801 1802 1803

	/*
	 * we may be replacing one extent in the tree with another.
	 * The new extent is pinned in the extent map, and we don't want
	 * to drop it from the cache until it is completely in the btree.
	 *
	 * So, tell btrfs_drop_extents to leave this extent in the cache.
	 * the caller is expected to unpin it and allow it to be merged
	 * with the others.
	 */
J
Josef Bacik 已提交
1804
	ret = btrfs_drop_extents(trans, root, inode, file_pos,
1805
				 file_pos + num_bytes, 0);
1806 1807
	if (ret)
		goto out;
Y
Yan Zheng 已提交
1808

L
Li Zefan 已提交
1809
	ins.objectid = btrfs_ino(inode);
Y
Yan Zheng 已提交
1810 1811 1812
	ins.offset = file_pos;
	ins.type = BTRFS_EXTENT_DATA_KEY;
	ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1813 1814
	if (ret)
		goto out;
Y
Yan Zheng 已提交
1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827
	leaf = path->nodes[0];
	fi = btrfs_item_ptr(leaf, path->slots[0],
			    struct btrfs_file_extent_item);
	btrfs_set_file_extent_generation(leaf, fi, trans->transid);
	btrfs_set_file_extent_type(leaf, fi, extent_type);
	btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
	btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
	btrfs_set_file_extent_offset(leaf, fi, 0);
	btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
	btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
	btrfs_set_file_extent_compression(leaf, fi, compression);
	btrfs_set_file_extent_encryption(leaf, fi, encryption);
	btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1828 1829 1830 1831

	btrfs_unlock_up_safe(path, 1);
	btrfs_set_lock_blocking(leaf);

Y
Yan Zheng 已提交
1832 1833 1834 1835 1836 1837 1838
	btrfs_mark_buffer_dirty(leaf);

	inode_add_bytes(inode, num_bytes);

	ins.objectid = disk_bytenr;
	ins.offset = disk_num_bytes;
	ins.type = BTRFS_EXTENT_ITEM_KEY;
1839 1840
	ret = btrfs_alloc_reserved_file_extent(trans, root,
					root->root_key.objectid,
L
Li Zefan 已提交
1841
					btrfs_ino(inode), file_pos, &ins);
1842
out:
Y
Yan Zheng 已提交
1843
	btrfs_free_path(path);
1844

1845
	return ret;
Y
Yan Zheng 已提交
1846 1847
}

1848 1849 1850 1851 1852 1853
/*
 * helper function for btrfs_finish_ordered_io, this
 * just reads in some of the csum leaves to prime them into ram
 * before we start the transaction.  It limits the amount of btree
 * reads required while inside the transaction.
 */
C
Chris Mason 已提交
1854 1855 1856 1857
/* as ordered data IO finishes, this gets called so we can finish
 * an ordered extent if the range of bytes in the file it covers are
 * fully written.
 */
1858
static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1859
{
1860
	struct inode *inode = ordered_extent->inode;
1861
	struct btrfs_root *root = BTRFS_I(inode)->root;
1862
	struct btrfs_trans_handle *trans = NULL;
1863
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1864
	struct extent_state *cached_state = NULL;
1865
	int compress_type = 0;
1866
	int ret;
1867
	bool nolock;
1868

1869
	nolock = btrfs_is_free_space_inode(inode);
J
Josef Bacik 已提交
1870

1871 1872 1873 1874 1875
	if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
		ret = -EIO;
		goto out;
	}

1876
	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1877
		BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1878 1879
		ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
		if (!ret) {
J
Josef Bacik 已提交
1880
			if (nolock)
1881
				trans = btrfs_join_transaction_nolock(root);
J
Josef Bacik 已提交
1882
			else
1883
				trans = btrfs_join_transaction(root);
1884 1885 1886 1887 1888
			if (IS_ERR(trans)) {
				ret = PTR_ERR(trans);
				trans = NULL;
				goto out;
			}
1889
			trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1890
			ret = btrfs_update_inode_fallback(trans, root, inode);
1891 1892
			if (ret) /* -ENOMEM or corruption */
				btrfs_abort_transaction(trans, root, ret);
1893 1894 1895
		}
		goto out;
	}
1896

1897 1898
	lock_extent_bits(io_tree, ordered_extent->file_offset,
			 ordered_extent->file_offset + ordered_extent->len - 1,
1899
			 0, &cached_state);
1900

J
Josef Bacik 已提交
1901
	if (nolock)
1902
		trans = btrfs_join_transaction_nolock(root);
J
Josef Bacik 已提交
1903
	else
1904
		trans = btrfs_join_transaction(root);
1905 1906 1907 1908 1909
	if (IS_ERR(trans)) {
		ret = PTR_ERR(trans);
		trans = NULL;
		goto out_unlock;
	}
1910
	trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1911

C
Chris Mason 已提交
1912
	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1913
		compress_type = ordered_extent->compress_type;
Y
Yan Zheng 已提交
1914
	if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1915
		BUG_ON(compress_type);
Y
Yan, Zheng 已提交
1916
		ret = btrfs_mark_extent_written(trans, inode,
Y
Yan Zheng 已提交
1917 1918 1919 1920
						ordered_extent->file_offset,
						ordered_extent->file_offset +
						ordered_extent->len);
	} else {
1921
		BUG_ON(root == root->fs_info->tree_root);
Y
Yan Zheng 已提交
1922 1923 1924 1925 1926 1927
		ret = insert_reserved_file_extent(trans, inode,
						ordered_extent->file_offset,
						ordered_extent->start,
						ordered_extent->disk_len,
						ordered_extent->len,
						ordered_extent->len,
1928
						compress_type, 0, 0,
Y
Yan Zheng 已提交
1929 1930
						BTRFS_FILE_EXTENT_REG);
	}
J
Josef Bacik 已提交
1931 1932 1933
	unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
			   ordered_extent->file_offset, ordered_extent->len,
			   trans->transid);
1934 1935
	if (ret < 0) {
		btrfs_abort_transaction(trans, root, ret);
1936
		goto out_unlock;
1937
	}
1938

1939 1940 1941
	add_pending_csums(trans, inode, ordered_extent->file_offset,
			  &ordered_extent->list);

1942
	ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1943
	if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1944
		ret = btrfs_update_inode_fallback(trans, root, inode);
1945 1946
		if (ret) { /* -ENOMEM or corruption */
			btrfs_abort_transaction(trans, root, ret);
1947
			goto out_unlock;
1948
		}
1949 1950
	} else {
		btrfs_set_inode_last_trans(trans, inode);
1951 1952
	}
	ret = 0;
1953 1954 1955 1956
out_unlock:
	unlock_extent_cached(io_tree, ordered_extent->file_offset,
			     ordered_extent->file_offset +
			     ordered_extent->len - 1, &cached_state, GFP_NOFS);
1957
out:
1958
	if (root != root->fs_info->tree_root)
J
Josef Bacik 已提交
1959
		btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1960 1961
	if (trans) {
		if (nolock)
J
Josef Bacik 已提交
1962
			btrfs_end_transaction_nolock(trans, root);
1963
		else
J
Josef Bacik 已提交
1964 1965 1966
			btrfs_end_transaction(trans, root);
	}

1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
	if (ret)
		clear_extent_uptodate(io_tree, ordered_extent->file_offset,
				      ordered_extent->file_offset +
				      ordered_extent->len - 1, NULL, GFP_NOFS);

	/*
	 * This needs to be dont to make sure anybody waiting knows we are done
	 * upating everything for this ordered extent.
	 */
	btrfs_remove_ordered_extent(inode, ordered_extent);

1978 1979 1980 1981 1982
	/* once for us */
	btrfs_put_ordered_extent(ordered_extent);
	/* once for the tree */
	btrfs_put_ordered_extent(ordered_extent);

1983 1984 1985 1986 1987 1988 1989 1990
	return ret;
}

static void finish_ordered_fn(struct btrfs_work *work)
{
	struct btrfs_ordered_extent *ordered_extent;
	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
	btrfs_finish_ordered_io(ordered_extent);
1991 1992
}

1993
static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1994 1995
				struct extent_state *state, int uptodate)
{
1996 1997 1998 1999 2000
	struct inode *inode = page->mapping->host;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_ordered_extent *ordered_extent = NULL;
	struct btrfs_workers *workers;

2001 2002
	trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);

2003
	ClearPagePrivate2(page);
2004 2005 2006 2007 2008 2009 2010
	if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
					    end - start + 1, uptodate))
		return 0;

	ordered_extent->work.func = finish_ordered_fn;
	ordered_extent->work.flags = 0;

2011
	if (btrfs_is_free_space_inode(inode))
2012 2013 2014 2015 2016 2017
		workers = &root->fs_info->endio_freespace_worker;
	else
		workers = &root->fs_info->endio_write_workers;
	btrfs_queue_worker(workers, &ordered_extent->work);

	return 0;
2018 2019
}

C
Chris Mason 已提交
2020 2021
/*
 * when reads are done, we need to check csums to verify the data is correct
2022 2023
 * if there's a match, we allow the bio to finish.  If not, the code in
 * extent_io.c will try to find good copies for us.
C
Chris Mason 已提交
2024
 */
2025
static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2026
			       struct extent_state *state, int mirror)
2027
{
2028
	size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2029
	struct inode *inode = page->mapping->host;
2030
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2031
	char *kaddr;
2032
	u64 private = ~(u32)0;
2033
	int ret;
2034 2035
	struct btrfs_root *root = BTRFS_I(inode)->root;
	u32 csum = ~(u32)0;
2036

2037 2038 2039 2040
	if (PageChecked(page)) {
		ClearPageChecked(page);
		goto good;
	}
2041 2042

	if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2043
		goto good;
2044 2045

	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2046
	    test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2047 2048
		clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
				  GFP_NOFS);
2049
		return 0;
2050
	}
2051

Y
Yan 已提交
2052
	if (state && state->start == start) {
2053 2054 2055 2056 2057
		private = state->private;
		ret = 0;
	} else {
		ret = get_state_private(io_tree, start, &private);
	}
2058
	kaddr = kmap_atomic(page);
C
Chris Mason 已提交
2059
	if (ret)
2060
		goto zeroit;
C
Chris Mason 已提交
2061

2062 2063
	csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
	btrfs_csum_final(csum, (char *)&csum);
C
Chris Mason 已提交
2064
	if (csum != private)
2065
		goto zeroit;
C
Chris Mason 已提交
2066

2067
	kunmap_atomic(kaddr);
2068
good:
2069 2070 2071
	return 0;

zeroit:
2072
	printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
L
Li Zefan 已提交
2073 2074
		       "private %llu\n",
		       (unsigned long long)btrfs_ino(page->mapping->host),
C
Chris Mason 已提交
2075 2076
		       (unsigned long long)start, csum,
		       (unsigned long long)private);
2077 2078
	memset(kaddr + offset, 1, end - start + 1);
	flush_dcache_page(page);
2079
	kunmap_atomic(kaddr);
2080 2081
	if (private == 0)
		return 0;
2082
	return -EIO;
2083
}
2084

Y
Yan, Zheng 已提交
2085 2086 2087 2088 2089
struct delayed_iput {
	struct list_head list;
	struct inode *inode;
};

2090 2091
/* JDM: If this is fs-wide, why can't we add a pointer to
 * btrfs_inode instead and avoid the allocation? */
Y
Yan, Zheng 已提交
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 2124 2125 2126 2127 2128 2129 2130 2131 2132
void btrfs_add_delayed_iput(struct inode *inode)
{
	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
	struct delayed_iput *delayed;

	if (atomic_add_unless(&inode->i_count, -1, 1))
		return;

	delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
	delayed->inode = inode;

	spin_lock(&fs_info->delayed_iput_lock);
	list_add_tail(&delayed->list, &fs_info->delayed_iputs);
	spin_unlock(&fs_info->delayed_iput_lock);
}

void btrfs_run_delayed_iputs(struct btrfs_root *root)
{
	LIST_HEAD(list);
	struct btrfs_fs_info *fs_info = root->fs_info;
	struct delayed_iput *delayed;
	int empty;

	spin_lock(&fs_info->delayed_iput_lock);
	empty = list_empty(&fs_info->delayed_iputs);
	spin_unlock(&fs_info->delayed_iput_lock);
	if (empty)
		return;

	spin_lock(&fs_info->delayed_iput_lock);
	list_splice_init(&fs_info->delayed_iputs, &list);
	spin_unlock(&fs_info->delayed_iput_lock);

	while (!list_empty(&list)) {
		delayed = list_entry(list.next, struct delayed_iput, list);
		list_del(&delayed->list);
		iput(delayed->inode);
		kfree(delayed);
	}
}

2133 2134 2135 2136 2137 2138
enum btrfs_orphan_cleanup_state {
	ORPHAN_CLEANUP_STARTED	= 1,
	ORPHAN_CLEANUP_DONE	= 2,
};

/*
2139
 * This is called in transaction commit time. If there are no orphan
2140 2141 2142 2143 2144 2145
 * files in the subvolume, it removes orphan item and frees block_rsv
 * structure.
 */
void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
			      struct btrfs_root *root)
{
2146
	struct btrfs_block_rsv *block_rsv;
2147 2148
	int ret;

2149
	if (atomic_read(&root->orphan_inodes) ||
2150 2151 2152
	    root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
		return;

2153
	spin_lock(&root->orphan_lock);
2154
	if (atomic_read(&root->orphan_inodes)) {
2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167
		spin_unlock(&root->orphan_lock);
		return;
	}

	if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
		spin_unlock(&root->orphan_lock);
		return;
	}

	block_rsv = root->orphan_block_rsv;
	root->orphan_block_rsv = NULL;
	spin_unlock(&root->orphan_lock);

2168 2169 2170 2171 2172 2173 2174 2175
	if (root->orphan_item_inserted &&
	    btrfs_root_refs(&root->root_item) > 0) {
		ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
					    root->root_key.objectid);
		BUG_ON(ret);
		root->orphan_item_inserted = 0;
	}

2176 2177 2178
	if (block_rsv) {
		WARN_ON(block_rsv->size > 0);
		btrfs_free_block_rsv(root, block_rsv);
2179 2180 2181
	}
}

2182 2183 2184
/*
 * This creates an orphan entry for the given inode in case something goes
 * wrong in the middle of an unlink/truncate.
2185 2186 2187
 *
 * NOTE: caller of this function should reserve 5 units of metadata for
 *	 this function.
2188 2189 2190 2191
 */
int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
2192 2193 2194 2195
	struct btrfs_block_rsv *block_rsv = NULL;
	int reserve = 0;
	int insert = 0;
	int ret;
2196

2197
	if (!root->orphan_block_rsv) {
2198
		block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2199 2200
		if (!block_rsv)
			return -ENOMEM;
2201
	}
2202

2203 2204 2205 2206 2207 2208
	spin_lock(&root->orphan_lock);
	if (!root->orphan_block_rsv) {
		root->orphan_block_rsv = block_rsv;
	} else if (block_rsv) {
		btrfs_free_block_rsv(root, block_rsv);
		block_rsv = NULL;
2209 2210
	}

2211 2212
	if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
			      &BTRFS_I(inode)->runtime_flags)) {
2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224
#if 0
		/*
		 * For proper ENOSPC handling, we should do orphan
		 * cleanup when mounting. But this introduces backward
		 * compatibility issue.
		 */
		if (!xchg(&root->orphan_item_inserted, 1))
			insert = 2;
		else
			insert = 1;
#endif
		insert = 1;
2225
		atomic_inc(&root->orphan_inodes);
2226 2227
	}

2228 2229
	if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
			      &BTRFS_I(inode)->runtime_flags))
2230 2231
		reserve = 1;
	spin_unlock(&root->orphan_lock);
2232

2233 2234 2235
	/* grab metadata reservation from transaction handle */
	if (reserve) {
		ret = btrfs_orphan_reserve_metadata(trans, inode);
2236
		BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2237
	}
2238

2239 2240
	/* insert an orphan item to track this unlinked/truncated file */
	if (insert >= 1) {
L
Li Zefan 已提交
2241
		ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2242
		if (ret && ret != -EEXIST) {
2243 2244
			clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
				  &BTRFS_I(inode)->runtime_flags);
2245 2246 2247 2248
			btrfs_abort_transaction(trans, root, ret);
			return ret;
		}
		ret = 0;
2249 2250 2251 2252 2253 2254
	}

	/* insert an orphan item to track subvolume contains orphan files */
	if (insert >= 2) {
		ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
					       root->root_key.objectid);
2255 2256 2257 2258
		if (ret && ret != -EEXIST) {
			btrfs_abort_transaction(trans, root, ret);
			return ret;
		}
2259 2260
	}
	return 0;
2261 2262 2263 2264 2265 2266 2267 2268 2269
}

/*
 * We have done the truncate/delete so we can go ahead and remove the orphan
 * item for this particular inode.
 */
int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
2270 2271
	int delete_item = 0;
	int release_rsv = 0;
2272 2273
	int ret = 0;

2274
	spin_lock(&root->orphan_lock);
2275 2276
	if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
			       &BTRFS_I(inode)->runtime_flags))
2277
		delete_item = 1;
2278

2279 2280
	if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
			       &BTRFS_I(inode)->runtime_flags))
2281 2282
		release_rsv = 1;
	spin_unlock(&root->orphan_lock);
2283

2284
	if (trans && delete_item) {
L
Li Zefan 已提交
2285
		ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2286
		BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2287
	}
2288

2289
	if (release_rsv) {
2290
		btrfs_orphan_release_metadata(inode);
2291 2292
		atomic_dec(&root->orphan_inodes);
	}
2293

2294
	return 0;
2295 2296 2297 2298 2299 2300
}

/*
 * this cleans up any orphans that may be left on the list from the last use
 * of this root.
 */
2301
int btrfs_orphan_cleanup(struct btrfs_root *root)
2302 2303 2304 2305 2306 2307
{
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_key key, found_key;
	struct btrfs_trans_handle *trans;
	struct inode *inode;
2308
	u64 last_objectid = 0;
2309 2310
	int ret = 0, nr_unlink = 0, nr_truncate = 0;

2311
	if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2312
		return 0;
2313 2314

	path = btrfs_alloc_path();
2315 2316 2317 2318
	if (!path) {
		ret = -ENOMEM;
		goto out;
	}
2319 2320 2321 2322 2323 2324 2325 2326
	path->reada = -1;

	key.objectid = BTRFS_ORPHAN_OBJECTID;
	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
	key.offset = (u64)-1;

	while (1) {
		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2327 2328
		if (ret < 0)
			goto out;
2329 2330 2331

		/*
		 * if ret == 0 means we found what we were searching for, which
L
Lucas De Marchi 已提交
2332
		 * is weird, but possible, so only screw with path if we didn't
2333 2334 2335
		 * find the key and see if we have stuff that matches
		 */
		if (ret > 0) {
2336
			ret = 0;
2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352
			if (path->slots[0] == 0)
				break;
			path->slots[0]--;
		}

		/* pull out the item */
		leaf = path->nodes[0];
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);

		/* make sure the item matches what we want */
		if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
			break;
		if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
			break;

		/* release the path since we're done with it */
2353
		btrfs_release_path(path);
2354 2355 2356 2357 2358 2359

		/*
		 * this is where we are basically btrfs_lookup, without the
		 * crossing root thing.  we store the inode number in the
		 * offset of the orphan item.
		 */
2360 2361 2362 2363 2364 2365 2366 2367 2368 2369

		if (found_key.offset == last_objectid) {
			printk(KERN_ERR "btrfs: Error removing orphan entry, "
			       "stopping orphan cleanup\n");
			ret = -EINVAL;
			goto out;
		}

		last_objectid = found_key.offset;

2370 2371 2372
		found_key.objectid = found_key.offset;
		found_key.type = BTRFS_INODE_ITEM_KEY;
		found_key.offset = 0;
2373
		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
J
Josef Bacik 已提交
2374 2375
		ret = PTR_RET(inode);
		if (ret && ret != -ESTALE)
2376
			goto out;
2377

2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409
		if (ret == -ESTALE && root == root->fs_info->tree_root) {
			struct btrfs_root *dead_root;
			struct btrfs_fs_info *fs_info = root->fs_info;
			int is_dead_root = 0;

			/*
			 * this is an orphan in the tree root. Currently these
			 * could come from 2 sources:
			 *  a) a snapshot deletion in progress
			 *  b) a free space cache inode
			 * We need to distinguish those two, as the snapshot
			 * orphan must not get deleted.
			 * find_dead_roots already ran before us, so if this
			 * is a snapshot deletion, we should find the root
			 * in the dead_roots list
			 */
			spin_lock(&fs_info->trans_lock);
			list_for_each_entry(dead_root, &fs_info->dead_roots,
					    root_list) {
				if (dead_root->root_key.objectid ==
				    found_key.objectid) {
					is_dead_root = 1;
					break;
				}
			}
			spin_unlock(&fs_info->trans_lock);
			if (is_dead_root) {
				/* prevent this orphan from being found again */
				key.offset = found_key.objectid - 1;
				continue;
			}
		}
2410
		/*
J
Josef Bacik 已提交
2411 2412
		 * Inode is already gone but the orphan item is still there,
		 * kill the orphan item.
2413
		 */
J
Josef Bacik 已提交
2414 2415
		if (ret == -ESTALE) {
			trans = btrfs_start_transaction(root, 1);
2416 2417 2418 2419
			if (IS_ERR(trans)) {
				ret = PTR_ERR(trans);
				goto out;
			}
2420 2421
			printk(KERN_ERR "auto deleting %Lu\n",
			       found_key.objectid);
J
Josef Bacik 已提交
2422 2423
			ret = btrfs_del_orphan_item(trans, root,
						    found_key.objectid);
2424
			BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2425
			btrfs_end_transaction(trans, root);
2426 2427 2428
			continue;
		}

J
Josef Bacik 已提交
2429 2430 2431 2432
		/*
		 * add this inode to the orphan list so btrfs_orphan_del does
		 * the proper thing when we hit it
		 */
2433 2434
		set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
			&BTRFS_I(inode)->runtime_flags);
J
Josef Bacik 已提交
2435

2436 2437
		/* if we have links, this was a truncate, lets do that */
		if (inode->i_nlink) {
2438 2439 2440 2441 2442
			if (!S_ISREG(inode->i_mode)) {
				WARN_ON(1);
				iput(inode);
				continue;
			}
2443
			nr_truncate++;
2444
			ret = btrfs_truncate(inode);
2445 2446 2447 2448 2449 2450
		} else {
			nr_unlink++;
		}

		/* this will do delete_inode and everything for us */
		iput(inode);
2451 2452
		if (ret)
			goto out;
2453
	}
2454 2455 2456
	/* release the path since we're done with it */
	btrfs_release_path(path);

2457 2458 2459 2460 2461 2462 2463
	root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;

	if (root->orphan_block_rsv)
		btrfs_block_rsv_release(root, root->orphan_block_rsv,
					(u64)-1);

	if (root->orphan_block_rsv || root->orphan_item_inserted) {
2464
		trans = btrfs_join_transaction(root);
2465 2466
		if (!IS_ERR(trans))
			btrfs_end_transaction(trans, root);
2467
	}
2468 2469 2470 2471 2472

	if (nr_unlink)
		printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
	if (nr_truncate)
		printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2473 2474 2475 2476 2477 2478

out:
	if (ret)
		printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
	btrfs_free_path(path);
	return ret;
2479 2480
}

2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531
/*
 * very simple check to peek ahead in the leaf looking for xattrs.  If we
 * don't find any xattrs, we know there can't be any acls.
 *
 * slot is the slot the inode is in, objectid is the objectid of the inode
 */
static noinline int acls_after_inode_item(struct extent_buffer *leaf,
					  int slot, u64 objectid)
{
	u32 nritems = btrfs_header_nritems(leaf);
	struct btrfs_key found_key;
	int scanned = 0;

	slot++;
	while (slot < nritems) {
		btrfs_item_key_to_cpu(leaf, &found_key, slot);

		/* we found a different objectid, there must not be acls */
		if (found_key.objectid != objectid)
			return 0;

		/* we found an xattr, assume we've got an acl */
		if (found_key.type == BTRFS_XATTR_ITEM_KEY)
			return 1;

		/*
		 * we found a key greater than an xattr key, there can't
		 * be any acls later on
		 */
		if (found_key.type > BTRFS_XATTR_ITEM_KEY)
			return 0;

		slot++;
		scanned++;

		/*
		 * it goes inode, inode backrefs, xattrs, extents,
		 * so if there are a ton of hard links to an inode there can
		 * be a lot of backrefs.  Don't waste time searching too hard,
		 * this is just an optimization
		 */
		if (scanned >= 8)
			break;
	}
	/* we hit the end of the leaf before we found an xattr or
	 * something larger than an xattr.  We have to assume the inode
	 * has acls
	 */
	return 1;
}

C
Chris Mason 已提交
2532 2533 2534
/*
 * read an inode from the btree into the in-memory inode
 */
2535
static void btrfs_read_locked_inode(struct inode *inode)
C
Chris Mason 已提交
2536 2537
{
	struct btrfs_path *path;
2538
	struct extent_buffer *leaf;
C
Chris Mason 已提交
2539
	struct btrfs_inode_item *inode_item;
2540
	struct btrfs_timespec *tspec;
C
Chris Mason 已提交
2541 2542
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_key location;
2543
	int maybe_acls;
J
Josef Bacik 已提交
2544
	u32 rdev;
C
Chris Mason 已提交
2545
	int ret;
2546 2547 2548 2549 2550
	bool filled = false;

	ret = btrfs_fill_inode(inode, &rdev);
	if (!ret)
		filled = true;
C
Chris Mason 已提交
2551 2552

	path = btrfs_alloc_path();
2553 2554 2555
	if (!path)
		goto make_bad;

2556
	path->leave_spinning = 1;
C
Chris Mason 已提交
2557
	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
C
Chris Mason 已提交
2558

C
Chris Mason 已提交
2559
	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2560
	if (ret)
C
Chris Mason 已提交
2561 2562
		goto make_bad;

2563
	leaf = path->nodes[0];
2564 2565 2566 2567

	if (filled)
		goto cache_acl;

2568 2569 2570
	inode_item = btrfs_item_ptr(leaf, path->slots[0],
				    struct btrfs_inode_item);
	inode->i_mode = btrfs_inode_mode(leaf, inode_item);
M
Miklos Szeredi 已提交
2571
	set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2572 2573
	inode->i_uid = btrfs_inode_uid(leaf, inode_item);
	inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2574
	btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587

	tspec = btrfs_inode_atime(inode_item);
	inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
	inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);

	tspec = btrfs_inode_mtime(inode_item);
	inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
	inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);

	tspec = btrfs_inode_ctime(inode_item);
	inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
	inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);

2588
	inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2589
	BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
J
Josef Bacik 已提交
2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601
	BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);

	/*
	 * If we were modified in the current generation and evicted from memory
	 * and then re-read we need to do a full sync since we don't have any
	 * idea about which extents were modified before we were evicted from
	 * cache.
	 */
	if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
			&BTRFS_I(inode)->runtime_flags);

2602
	inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2603
	inode->i_generation = BTRFS_I(inode)->generation;
J
Josef Bacik 已提交
2604
	inode->i_rdev = 0;
2605 2606
	rdev = btrfs_inode_rdev(leaf, inode_item);

2607
	BTRFS_I(inode)->index_cnt = (u64)-1;
2608
	BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2609
cache_acl:
2610 2611 2612 2613
	/*
	 * try to precache a NULL acl entry for files that don't have
	 * any xattrs or acls
	 */
L
Li Zefan 已提交
2614 2615
	maybe_acls = acls_after_inode_item(leaf, path->slots[0],
					   btrfs_ino(inode));
2616 2617
	if (!maybe_acls)
		cache_no_acl(inode);
2618

C
Chris Mason 已提交
2619 2620 2621 2622 2623
	btrfs_free_path(path);

	switch (inode->i_mode & S_IFMT) {
	case S_IFREG:
		inode->i_mapping->a_ops = &btrfs_aops;
C
Chris Mason 已提交
2624
		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2625
		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
C
Chris Mason 已提交
2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638
		inode->i_fop = &btrfs_file_operations;
		inode->i_op = &btrfs_file_inode_operations;
		break;
	case S_IFDIR:
		inode->i_fop = &btrfs_dir_file_operations;
		if (root == root->fs_info->tree_root)
			inode->i_op = &btrfs_dir_ro_inode_operations;
		else
			inode->i_op = &btrfs_dir_inode_operations;
		break;
	case S_IFLNK:
		inode->i_op = &btrfs_symlink_inode_operations;
		inode->i_mapping->a_ops = &btrfs_symlink_aops;
C
Chris Mason 已提交
2639
		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
C
Chris Mason 已提交
2640
		break;
J
Josef Bacik 已提交
2641
	default:
J
Jim Owens 已提交
2642
		inode->i_op = &btrfs_special_inode_operations;
J
Josef Bacik 已提交
2643 2644
		init_special_inode(inode, inode->i_mode, rdev);
		break;
C
Chris Mason 已提交
2645
	}
2646 2647

	btrfs_update_iflags(inode);
C
Chris Mason 已提交
2648 2649 2650 2651 2652 2653 2654
	return;

make_bad:
	btrfs_free_path(path);
	make_bad_inode(inode);
}

C
Chris Mason 已提交
2655 2656 2657
/*
 * given a leaf and an inode, copy the inode fields into the leaf
 */
2658 2659
static void fill_inode_item(struct btrfs_trans_handle *trans,
			    struct extent_buffer *leaf,
2660
			    struct btrfs_inode_item *item,
C
Chris Mason 已提交
2661 2662
			    struct inode *inode)
{
2663 2664
	btrfs_set_inode_uid(leaf, item, inode->i_uid);
	btrfs_set_inode_gid(leaf, item, inode->i_gid);
2665
	btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683
	btrfs_set_inode_mode(leaf, item, inode->i_mode);
	btrfs_set_inode_nlink(leaf, item, inode->i_nlink);

	btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
			       inode->i_atime.tv_sec);
	btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
				inode->i_atime.tv_nsec);

	btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
			       inode->i_mtime.tv_sec);
	btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
				inode->i_mtime.tv_nsec);

	btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
			       inode->i_ctime.tv_sec);
	btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
				inode->i_ctime.tv_nsec);

2684
	btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2685
	btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2686
	btrfs_set_inode_sequence(leaf, item, inode->i_version);
2687
	btrfs_set_inode_transid(leaf, item, trans->transid);
2688
	btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
Y
Yan 已提交
2689
	btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2690
	btrfs_set_inode_block_group(leaf, item, 0);
C
Chris Mason 已提交
2691 2692
}

C
Chris Mason 已提交
2693 2694 2695
/*
 * copy everything in the in-memory inode into the btree.
 */
2696
static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
C
Chris Mason 已提交
2697
				struct btrfs_root *root, struct inode *inode)
C
Chris Mason 已提交
2698 2699 2700
{
	struct btrfs_inode_item *inode_item;
	struct btrfs_path *path;
2701
	struct extent_buffer *leaf;
C
Chris Mason 已提交
2702 2703 2704
	int ret;

	path = btrfs_alloc_path();
2705 2706 2707
	if (!path)
		return -ENOMEM;

2708
	path->leave_spinning = 1;
2709 2710
	ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
				 1);
C
Chris Mason 已提交
2711 2712 2713 2714 2715 2716
	if (ret) {
		if (ret > 0)
			ret = -ENOENT;
		goto failed;
	}

2717
	btrfs_unlock_up_safe(path, 1);
2718 2719
	leaf = path->nodes[0];
	inode_item = btrfs_item_ptr(leaf, path->slots[0],
2720
				    struct btrfs_inode_item);
C
Chris Mason 已提交
2721

2722
	fill_inode_item(trans, leaf, inode_item, inode);
2723
	btrfs_mark_buffer_dirty(leaf);
2724
	btrfs_set_inode_last_trans(trans, inode);
C
Chris Mason 已提交
2725 2726 2727 2728 2729 2730
	ret = 0;
failed:
	btrfs_free_path(path);
	return ret;
}

2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745
/*
 * copy everything in the in-memory inode into the btree.
 */
noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
				struct btrfs_root *root, struct inode *inode)
{
	int ret;

	/*
	 * If the inode is a free space inode, we can deadlock during commit
	 * if we put it into the delayed code.
	 *
	 * The data relocation inode should also be directly updated
	 * without delay
	 */
2746
	if (!btrfs_is_free_space_inode(inode)
2747
	    && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2748 2749
		btrfs_update_root_times(trans, root);

2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769
		ret = btrfs_delayed_update_inode(trans, root, inode);
		if (!ret)
			btrfs_set_inode_last_trans(trans, inode);
		return ret;
	}

	return btrfs_update_inode_item(trans, root, inode);
}

static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
				struct btrfs_root *root, struct inode *inode)
{
	int ret;

	ret = btrfs_update_inode(trans, root, inode);
	if (ret == -ENOSPC)
		return btrfs_update_inode_item(trans, root, inode);
	return ret;
}

C
Chris Mason 已提交
2770 2771 2772 2773 2774
/*
 * unlink helper that gets used here in inode.c and in the tree logging
 * recovery code.  It remove a link in a directory with a given name, and
 * also drops the back refs in the inode to the directory
 */
2775 2776 2777 2778
static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
				struct btrfs_root *root,
				struct inode *dir, struct inode *inode,
				const char *name, int name_len)
C
Chris Mason 已提交
2779 2780 2781
{
	struct btrfs_path *path;
	int ret = 0;
2782
	struct extent_buffer *leaf;
C
Chris Mason 已提交
2783
	struct btrfs_dir_item *di;
2784
	struct btrfs_key key;
2785
	u64 index;
L
Li Zefan 已提交
2786 2787
	u64 ino = btrfs_ino(inode);
	u64 dir_ino = btrfs_ino(dir);
C
Chris Mason 已提交
2788 2789

	path = btrfs_alloc_path();
2790 2791
	if (!path) {
		ret = -ENOMEM;
2792
		goto out;
2793 2794
	}

2795
	path->leave_spinning = 1;
L
Li Zefan 已提交
2796
	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
C
Chris Mason 已提交
2797 2798 2799 2800 2801 2802 2803 2804 2805
				    name, name_len, -1);
	if (IS_ERR(di)) {
		ret = PTR_ERR(di);
		goto err;
	}
	if (!di) {
		ret = -ENOENT;
		goto err;
	}
2806 2807
	leaf = path->nodes[0];
	btrfs_dir_item_key_to_cpu(leaf, di, &key);
C
Chris Mason 已提交
2808
	ret = btrfs_delete_one_dir_name(trans, root, path, di);
2809 2810
	if (ret)
		goto err;
2811
	btrfs_release_path(path);
C
Chris Mason 已提交
2812

L
Li Zefan 已提交
2813 2814
	ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
				  dir_ino, &index);
2815
	if (ret) {
C
Chris Mason 已提交
2816
		printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
L
Li Zefan 已提交
2817 2818
		       "inode %llu parent %llu\n", name_len, name,
		       (unsigned long long)ino, (unsigned long long)dir_ino);
2819
		btrfs_abort_transaction(trans, root, ret);
2820 2821 2822
		goto err;
	}

2823
	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2824 2825
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
C
Chris Mason 已提交
2826
		goto err;
2827
	}
C
Chris Mason 已提交
2828

2829
	ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
L
Li Zefan 已提交
2830
					 inode, dir_ino);
2831 2832 2833 2834
	if (ret != 0 && ret != -ENOENT) {
		btrfs_abort_transaction(trans, root, ret);
		goto err;
	}
2835 2836 2837

	ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
					   dir, index);
2838 2839
	if (ret == -ENOENT)
		ret = 0;
C
Chris Mason 已提交
2840 2841
err:
	btrfs_free_path(path);
2842 2843 2844 2845
	if (ret)
		goto out;

	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2846 2847
	inode_inc_iversion(inode);
	inode_inc_iversion(dir);
2848
	inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2849
	ret = btrfs_update_inode(trans, root, dir);
2850
out:
C
Chris Mason 已提交
2851 2852 2853
	return ret;
}

2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868
int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
		       struct btrfs_root *root,
		       struct inode *dir, struct inode *inode,
		       const char *name, int name_len)
{
	int ret;
	ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
	if (!ret) {
		btrfs_drop_nlink(inode);
		ret = btrfs_update_inode(trans, root, inode);
	}
	return ret;
}
		

2869 2870 2871
/* helper to check if there is any shared block in the path */
static int check_path_shared(struct btrfs_root *root,
			     struct btrfs_path *path)
C
Chris Mason 已提交
2872
{
2873 2874
	struct extent_buffer *eb;
	int level;
2875
	u64 refs = 1;
2876

2877
	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2878 2879
		int ret;

2880 2881 2882 2883 2884 2885 2886 2887 2888
		if (!path->nodes[level])
			break;
		eb = path->nodes[level];
		if (!btrfs_block_can_be_shared(root, eb))
			continue;
		ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
					       &refs, NULL);
		if (refs > 1)
			return 1;
2889
	}
2890
	return 0;
C
Chris Mason 已提交
2891 2892
}

2893 2894 2895 2896 2897 2898 2899 2900 2901
/*
 * helper to start transaction for unlink and rmdir.
 *
 * unlink and rmdir are special in btrfs, they do not always free space.
 * so in enospc case, we should make sure they will free space before
 * allowing them to use the global metadata reservation.
 */
static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
						       struct dentry *dentry)
2902
{
C
Chris Mason 已提交
2903
	struct btrfs_trans_handle *trans;
2904
	struct btrfs_root *root = BTRFS_I(dir)->root;
2905
	struct btrfs_path *path;
2906
	struct btrfs_inode_ref *ref;
2907
	struct btrfs_dir_item *di;
2908
	struct inode *inode = dentry->d_inode;
2909
	u64 index;
2910 2911
	int check_link = 1;
	int err = -ENOSPC;
2912
	int ret;
L
Li Zefan 已提交
2913 2914
	u64 ino = btrfs_ino(inode);
	u64 dir_ino = btrfs_ino(dir);
2915

2916 2917 2918 2919 2920 2921 2922 2923 2924 2925
	/*
	 * 1 for the possible orphan item
	 * 1 for the dir item
	 * 1 for the dir index
	 * 1 for the inode ref
	 * 1 for the inode ref in the tree log
	 * 2 for the dir entries in the log
	 * 1 for the inode
	 */
	trans = btrfs_start_transaction(root, 8);
2926 2927
	if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
		return trans;
2928

L
Li Zefan 已提交
2929
	if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2930
		return ERR_PTR(-ENOSPC);
2931

2932 2933 2934
	/* check if there is someone else holds reference */
	if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
		return ERR_PTR(-ENOSPC);
2935

2936 2937
	if (atomic_read(&inode->i_count) > 2)
		return ERR_PTR(-ENOSPC);
2938

2939 2940 2941 2942 2943 2944 2945
	if (xchg(&root->fs_info->enospc_unlink, 1))
		return ERR_PTR(-ENOSPC);

	path = btrfs_alloc_path();
	if (!path) {
		root->fs_info->enospc_unlink = 0;
		return ERR_PTR(-ENOMEM);
2946 2947
	}

2948 2949
	/* 1 for the orphan item */
	trans = btrfs_start_transaction(root, 1);
2950
	if (IS_ERR(trans)) {
2951 2952 2953 2954
		btrfs_free_path(path);
		root->fs_info->enospc_unlink = 0;
		return trans;
	}
2955

2956 2957
	path->skip_locking = 1;
	path->search_commit_root = 1;
2958

2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969
	ret = btrfs_lookup_inode(trans, root, path,
				&BTRFS_I(dir)->location, 0);
	if (ret < 0) {
		err = ret;
		goto out;
	}
	if (ret == 0) {
		if (check_path_shared(root, path))
			goto out;
	} else {
		check_link = 0;
2970
	}
2971
	btrfs_release_path(path);
2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984

	ret = btrfs_lookup_inode(trans, root, path,
				&BTRFS_I(inode)->location, 0);
	if (ret < 0) {
		err = ret;
		goto out;
	}
	if (ret == 0) {
		if (check_path_shared(root, path))
			goto out;
	} else {
		check_link = 0;
	}
2985
	btrfs_release_path(path);
2986 2987 2988

	if (ret == 0 && S_ISREG(inode->i_mode)) {
		ret = btrfs_lookup_file_extent(trans, root, path,
L
Li Zefan 已提交
2989
					       ino, (u64)-1, 0);
2990 2991 2992 2993
		if (ret < 0) {
			err = ret;
			goto out;
		}
2994
		BUG_ON(ret == 0); /* Corruption */
2995 2996
		if (check_path_shared(root, path))
			goto out;
2997
		btrfs_release_path(path);
2998 2999 3000 3001 3002 3003 3004
	}

	if (!check_link) {
		err = 0;
		goto out;
	}

L
Li Zefan 已提交
3005
	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017
				dentry->d_name.name, dentry->d_name.len, 0);
	if (IS_ERR(di)) {
		err = PTR_ERR(di);
		goto out;
	}
	if (di) {
		if (check_path_shared(root, path))
			goto out;
	} else {
		err = 0;
		goto out;
	}
3018
	btrfs_release_path(path);
3019 3020 3021

	ref = btrfs_lookup_inode_ref(trans, root, path,
				dentry->d_name.name, dentry->d_name.len,
L
Li Zefan 已提交
3022
				ino, dir_ino, 0);
3023 3024 3025 3026
	if (IS_ERR(ref)) {
		err = PTR_ERR(ref);
		goto out;
	}
3027
	BUG_ON(!ref); /* Logic error */
3028 3029 3030
	if (check_path_shared(root, path))
		goto out;
	index = btrfs_inode_ref_index(path->nodes[0], ref);
3031
	btrfs_release_path(path);
3032

3033 3034 3035 3036 3037 3038 3039 3040
	/*
	 * This is a commit root search, if we can lookup inode item and other
	 * relative items in the commit root, it means the transaction of
	 * dir/file creation has been committed, and the dir index item that we
	 * delay to insert has also been inserted into the commit root. So
	 * we needn't worry about the delayed insertion of the dir index item
	 * here.
	 */
L
Li Zefan 已提交
3041
	di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053
				dentry->d_name.name, dentry->d_name.len, 0);
	if (IS_ERR(di)) {
		err = PTR_ERR(di);
		goto out;
	}
	BUG_ON(ret == -ENOENT);
	if (check_path_shared(root, path))
		goto out;

	err = 0;
out:
	btrfs_free_path(path);
3054 3055 3056 3057
	/* Migrate the orphan reservation over */
	if (!err)
		err = btrfs_block_rsv_migrate(trans->block_rsv,
				&root->fs_info->global_block_rsv,
3058
				trans->bytes_reserved);
3059

3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072
	if (err) {
		btrfs_end_transaction(trans, root);
		root->fs_info->enospc_unlink = 0;
		return ERR_PTR(err);
	}

	trans->block_rsv = &root->fs_info->global_block_rsv;
	return trans;
}

static void __unlink_end_trans(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root)
{
3073
	if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3074 3075 3076
		btrfs_block_rsv_release(root, trans->block_rsv,
					trans->bytes_reserved);
		trans->block_rsv = &root->fs_info->trans_block_rsv;
3077 3078 3079
		BUG_ON(!root->fs_info->enospc_unlink);
		root->fs_info->enospc_unlink = 0;
	}
3080
	btrfs_end_transaction(trans, root);
3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093
}

static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
{
	struct btrfs_root *root = BTRFS_I(dir)->root;
	struct btrfs_trans_handle *trans;
	struct inode *inode = dentry->d_inode;
	int ret;
	unsigned long nr = 0;

	trans = __unlink_start_trans(dir, dentry);
	if (IS_ERR(trans))
		return PTR_ERR(trans);
3094

3095 3096
	btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);

3097 3098
	ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
				 dentry->d_name.name, dentry->d_name.len);
3099 3100
	if (ret)
		goto out;
3101

3102
	if (inode->i_nlink == 0) {
3103
		ret = btrfs_orphan_add(trans, inode);
3104 3105
		if (ret)
			goto out;
3106
	}
3107

3108
out:
3109
	nr = trans->blocks_used;
3110
	__unlink_end_trans(trans, root);
3111
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
3112 3113 3114
	return ret;
}

3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125
int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
			struct btrfs_root *root,
			struct inode *dir, u64 objectid,
			const char *name, int name_len)
{
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	struct btrfs_dir_item *di;
	struct btrfs_key key;
	u64 index;
	int ret;
L
Li Zefan 已提交
3126
	u64 dir_ino = btrfs_ino(dir);
3127 3128 3129 3130 3131

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

L
Li Zefan 已提交
3132
	di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3133
				   name, name_len, -1);
3134 3135 3136 3137 3138 3139 3140
	if (IS_ERR_OR_NULL(di)) {
		if (!di)
			ret = -ENOENT;
		else
			ret = PTR_ERR(di);
		goto out;
	}
3141 3142 3143 3144 3145

	leaf = path->nodes[0];
	btrfs_dir_item_key_to_cpu(leaf, di, &key);
	WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
	ret = btrfs_delete_one_dir_name(trans, root, path, di);
3146 3147 3148 3149
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto out;
	}
3150
	btrfs_release_path(path);
3151 3152 3153

	ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
				 objectid, root->root_key.objectid,
L
Li Zefan 已提交
3154
				 dir_ino, &index, name, name_len);
3155
	if (ret < 0) {
3156 3157 3158 3159
		if (ret != -ENOENT) {
			btrfs_abort_transaction(trans, root, ret);
			goto out;
		}
L
Li Zefan 已提交
3160
		di = btrfs_search_dir_index_item(root, path, dir_ino,
3161
						 name, name_len);
3162 3163 3164 3165 3166 3167 3168 3169
		if (IS_ERR_OR_NULL(di)) {
			if (!di)
				ret = -ENOENT;
			else
				ret = PTR_ERR(di);
			btrfs_abort_transaction(trans, root, ret);
			goto out;
		}
3170 3171 3172

		leaf = path->nodes[0];
		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3173
		btrfs_release_path(path);
3174 3175
		index = key.offset;
	}
3176
	btrfs_release_path(path);
3177

3178
	ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3179 3180 3181 3182
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto out;
	}
3183 3184

	btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3185
	inode_inc_iversion(dir);
3186
	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3187
	ret = btrfs_update_inode_fallback(trans, root, dir);
3188 3189 3190
	if (ret)
		btrfs_abort_transaction(trans, root, ret);
out:
3191
	btrfs_free_path(path);
3192
	return ret;
3193 3194
}

C
Chris Mason 已提交
3195 3196 3197
static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
{
	struct inode *inode = dentry->d_inode;
3198
	int err = 0;
C
Chris Mason 已提交
3199 3200
	struct btrfs_root *root = BTRFS_I(dir)->root;
	struct btrfs_trans_handle *trans;
3201
	unsigned long nr = 0;
C
Chris Mason 已提交
3202

3203
	if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
L
Li Zefan 已提交
3204
	    btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
Y
Yan 已提交
3205 3206
		return -ENOTEMPTY;

3207 3208
	trans = __unlink_start_trans(dir, dentry);
	if (IS_ERR(trans))
3209 3210
		return PTR_ERR(trans);

L
Li Zefan 已提交
3211
	if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3212 3213 3214 3215 3216 3217 3218
		err = btrfs_unlink_subvol(trans, root, dir,
					  BTRFS_I(inode)->location.objectid,
					  dentry->d_name.name,
					  dentry->d_name.len);
		goto out;
	}

3219 3220
	err = btrfs_orphan_add(trans, inode);
	if (err)
3221
		goto out;
3222

C
Chris Mason 已提交
3223
	/* now the directory is empty */
3224 3225
	err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
				 dentry->d_name.name, dentry->d_name.len);
C
Chris Mason 已提交
3226
	if (!err)
3227
		btrfs_i_size_write(inode, 0);
3228
out:
3229
	nr = trans->blocks_used;
3230
	__unlink_end_trans(trans, root);
3231
	btrfs_btree_balance_dirty(root, nr);
3232

C
Chris Mason 已提交
3233 3234 3235 3236 3237 3238
	return err;
}

/*
 * this can truncate away extent items, csum items and directory items.
 * It starts at a high offset and removes keys until it can't find
C
Chris Mason 已提交
3239
 * any higher than new_size
C
Chris Mason 已提交
3240 3241 3242
 *
 * csum items that cross the new i_size are truncated to the new size
 * as well.
3243 3244 3245
 *
 * min_type is the minimum key type to truncate down to.  If set to 0, this
 * will kill all the items on this inode, including the INODE_ITEM_KEY.
C
Chris Mason 已提交
3246
 */
3247 3248 3249 3250
int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
			       struct btrfs_root *root,
			       struct inode *inode,
			       u64 new_size, u32 min_type)
C
Chris Mason 已提交
3251 3252
{
	struct btrfs_path *path;
3253
	struct extent_buffer *leaf;
C
Chris Mason 已提交
3254
	struct btrfs_file_extent_item *fi;
3255 3256
	struct btrfs_key key;
	struct btrfs_key found_key;
C
Chris Mason 已提交
3257
	u64 extent_start = 0;
3258
	u64 extent_num_bytes = 0;
3259
	u64 extent_offset = 0;
C
Chris Mason 已提交
3260
	u64 item_end = 0;
3261 3262
	u64 mask = root->sectorsize - 1;
	u32 found_type = (u8)-1;
C
Chris Mason 已提交
3263 3264
	int found_extent;
	int del_item;
3265 3266
	int pending_del_nr = 0;
	int pending_del_slot = 0;
3267
	int extent_type = -1;
3268 3269
	int ret;
	int err = 0;
L
Li Zefan 已提交
3270
	u64 ino = btrfs_ino(inode);
3271 3272

	BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
C
Chris Mason 已提交
3273

3274 3275 3276 3277 3278
	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;
	path->reada = -1;

J
Josef Bacik 已提交
3279 3280 3281 3282 3283
	/*
	 * We want to drop from the next block forward in case this new size is
	 * not block aligned since we will be keeping the last block of the
	 * extent just the way it is.
	 */
3284
	if (root->ref_cows || root == root->fs_info->tree_root)
J
Josef Bacik 已提交
3285
		btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3286

3287 3288 3289 3290 3291 3292 3293 3294 3295
	/*
	 * This function is also used to drop the items in the log tree before
	 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
	 * it is used to drop the loged items. So we shouldn't kill the delayed
	 * items.
	 */
	if (min_type == 0 && root == BTRFS_I(inode)->root)
		btrfs_kill_delayed_inode_items(inode);

L
Li Zefan 已提交
3296
	key.objectid = ino;
C
Chris Mason 已提交
3297
	key.offset = (u64)-1;
3298 3299
	key.type = (u8)-1;

3300
search_again:
3301
	path->leave_spinning = 1;
3302
	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3303 3304 3305 3306
	if (ret < 0) {
		err = ret;
		goto out;
	}
C
Chris Mason 已提交
3307

3308
	if (ret > 0) {
3309 3310 3311
		/* there are no items in the tree for us to truncate, we're
		 * done
		 */
3312 3313
		if (path->slots[0] == 0)
			goto out;
3314 3315 3316
		path->slots[0]--;
	}

C
Chris Mason 已提交
3317
	while (1) {
C
Chris Mason 已提交
3318
		fi = NULL;
3319 3320 3321
		leaf = path->nodes[0];
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
		found_type = btrfs_key_type(&found_key);
C
Chris Mason 已提交
3322

L
Li Zefan 已提交
3323
		if (found_key.objectid != ino)
C
Chris Mason 已提交
3324
			break;
3325

3326
		if (found_type < min_type)
C
Chris Mason 已提交
3327 3328
			break;

3329
		item_end = found_key.offset;
C
Chris Mason 已提交
3330
		if (found_type == BTRFS_EXTENT_DATA_KEY) {
3331
			fi = btrfs_item_ptr(leaf, path->slots[0],
C
Chris Mason 已提交
3332
					    struct btrfs_file_extent_item);
3333 3334
			extent_type = btrfs_file_extent_type(leaf, fi);
			if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3335
				item_end +=
3336
				    btrfs_file_extent_num_bytes(leaf, fi);
3337 3338
			} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
				item_end += btrfs_file_extent_inline_len(leaf,
C
Chris Mason 已提交
3339
									 fi);
C
Chris Mason 已提交
3340
			}
3341
			item_end--;
C
Chris Mason 已提交
3342
		}
3343 3344 3345 3346
		if (found_type > min_type) {
			del_item = 1;
		} else {
			if (item_end < new_size)
3347
				break;
3348 3349 3350 3351
			if (found_key.offset >= new_size)
				del_item = 1;
			else
				del_item = 0;
C
Chris Mason 已提交
3352 3353 3354
		}
		found_extent = 0;
		/* FIXME, shrink the extent if the ref count is only 1 */
3355 3356 3357 3358
		if (found_type != BTRFS_EXTENT_DATA_KEY)
			goto delete;

		if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
C
Chris Mason 已提交
3359
			u64 num_dec;
3360
			extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3361
			if (!del_item) {
3362 3363
				u64 orig_num_bytes =
					btrfs_file_extent_num_bytes(leaf, fi);
3364
				extent_num_bytes = new_size -
3365
					found_key.offset + root->sectorsize - 1;
3366 3367
				extent_num_bytes = extent_num_bytes &
					~((u64)root->sectorsize - 1);
3368 3369 3370
				btrfs_set_file_extent_num_bytes(leaf, fi,
							 extent_num_bytes);
				num_dec = (orig_num_bytes -
C
Chris Mason 已提交
3371
					   extent_num_bytes);
3372
				if (root->ref_cows && extent_start != 0)
3373
					inode_sub_bytes(inode, num_dec);
3374
				btrfs_mark_buffer_dirty(leaf);
C
Chris Mason 已提交
3375
			} else {
3376 3377 3378
				extent_num_bytes =
					btrfs_file_extent_disk_num_bytes(leaf,
									 fi);
3379 3380 3381
				extent_offset = found_key.offset -
					btrfs_file_extent_offset(leaf, fi);

C
Chris Mason 已提交
3382
				/* FIXME blocksize != 4096 */
C
Chris Mason 已提交
3383
				num_dec = btrfs_file_extent_num_bytes(leaf, fi);
C
Chris Mason 已提交
3384 3385
				if (extent_start != 0) {
					found_extent = 1;
3386
					if (root->ref_cows)
3387
						inode_sub_bytes(inode, num_dec);
3388
				}
C
Chris Mason 已提交
3389
			}
C
Chris Mason 已提交
3390
		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
C
Chris Mason 已提交
3391 3392 3393 3394 3395 3396 3397 3398
			/*
			 * we can't truncate inline items that have had
			 * special encodings
			 */
			if (!del_item &&
			    btrfs_file_extent_compression(leaf, fi) == 0 &&
			    btrfs_file_extent_encryption(leaf, fi) == 0 &&
			    btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3399 3400 3401
				u32 size = new_size - found_key.offset;

				if (root->ref_cows) {
3402 3403
					inode_sub_bytes(inode, item_end + 1 -
							new_size);
3404 3405 3406
				}
				size =
				    btrfs_file_extent_calc_inline_size(size);
3407 3408
				btrfs_truncate_item(trans, root, path,
						    size, 1);
3409
			} else if (root->ref_cows) {
3410 3411
				inode_sub_bytes(inode, item_end + 1 -
						found_key.offset);
C
Chris Mason 已提交
3412
			}
C
Chris Mason 已提交
3413
		}
3414
delete:
C
Chris Mason 已提交
3415
		if (del_item) {
3416 3417 3418 3419 3420 3421 3422 3423 3424 3425
			if (!pending_del_nr) {
				/* no pending yet, add ourselves */
				pending_del_slot = path->slots[0];
				pending_del_nr = 1;
			} else if (pending_del_nr &&
				   path->slots[0] + 1 == pending_del_slot) {
				/* hop on the pending chunk */
				pending_del_nr++;
				pending_del_slot = path->slots[0];
			} else {
C
Chris Mason 已提交
3426
				BUG();
3427
			}
C
Chris Mason 已提交
3428 3429 3430
		} else {
			break;
		}
3431 3432
		if (found_extent && (root->ref_cows ||
				     root == root->fs_info->tree_root)) {
3433
			btrfs_set_path_blocking(path);
C
Chris Mason 已提交
3434
			ret = btrfs_free_extent(trans, root, extent_start,
3435 3436
						extent_num_bytes, 0,
						btrfs_header_owner(leaf),
A
Arne Jansen 已提交
3437
						ino, extent_offset, 0);
C
Chris Mason 已提交
3438 3439
			BUG_ON(ret);
		}
3440

3441 3442 3443 3444 3445 3446 3447 3448 3449
		if (found_type == BTRFS_INODE_ITEM_KEY)
			break;

		if (path->slots[0] == 0 ||
		    path->slots[0] != pending_del_slot) {
			if (pending_del_nr) {
				ret = btrfs_del_items(trans, root, path,
						pending_del_slot,
						pending_del_nr);
3450 3451 3452 3453 3454
				if (ret) {
					btrfs_abort_transaction(trans,
								root, ret);
					goto error;
				}
3455 3456
				pending_del_nr = 0;
			}
3457
			btrfs_release_path(path);
3458
			goto search_again;
3459 3460
		} else {
			path->slots[0]--;
3461
		}
C
Chris Mason 已提交
3462
	}
3463
out:
3464 3465 3466
	if (pending_del_nr) {
		ret = btrfs_del_items(trans, root, path, pending_del_slot,
				      pending_del_nr);
3467 3468
		if (ret)
			btrfs_abort_transaction(trans, root, ret);
3469
	}
3470
error:
C
Chris Mason 已提交
3471
	btrfs_free_path(path);
3472
	return err;
C
Chris Mason 已提交
3473 3474 3475
}

/*
J
Josef Bacik 已提交
3476 3477 3478 3479 3480 3481 3482 3483 3484
 * btrfs_truncate_page - read, zero a chunk and write a page
 * @inode - inode that we're zeroing
 * @from - the offset to start zeroing
 * @len - the length to zero, 0 to zero the entire range respective to the
 *	offset
 * @front - zero up to the offset instead of from the offset on
 *
 * This will find the page for the "from" offset and cow the page and zero the
 * part we want to zero.  This is used with truncate and hole punching.
C
Chris Mason 已提交
3485
 */
J
Josef Bacik 已提交
3486 3487
int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
			int front)
C
Chris Mason 已提交
3488
{
J
Josef Bacik 已提交
3489
	struct address_space *mapping = inode->i_mapping;
3490
	struct btrfs_root *root = BTRFS_I(inode)->root;
3491 3492
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
	struct btrfs_ordered_extent *ordered;
3493
	struct extent_state *cached_state = NULL;
3494
	char *kaddr;
3495
	u32 blocksize = root->sectorsize;
C
Chris Mason 已提交
3496 3497 3498
	pgoff_t index = from >> PAGE_CACHE_SHIFT;
	unsigned offset = from & (PAGE_CACHE_SIZE-1);
	struct page *page;
3499
	gfp_t mask = btrfs_alloc_write_mask(mapping);
C
Chris Mason 已提交
3500
	int ret = 0;
3501
	u64 page_start;
3502
	u64 page_end;
C
Chris Mason 已提交
3503

J
Josef Bacik 已提交
3504 3505
	if ((offset & (blocksize - 1)) == 0 &&
	    (!len || ((len & (blocksize - 1)) == 0)))
C
Chris Mason 已提交
3506
		goto out;
3507
	ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3508 3509
	if (ret)
		goto out;
C
Chris Mason 已提交
3510 3511

	ret = -ENOMEM;
3512
again:
3513
	page = find_or_create_page(mapping, index, mask);
3514
	if (!page) {
3515
		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
C
Chris Mason 已提交
3516
		goto out;
3517
	}
3518 3519 3520 3521

	page_start = page_offset(page);
	page_end = page_start + PAGE_CACHE_SIZE - 1;

C
Chris Mason 已提交
3522
	if (!PageUptodate(page)) {
C
Chris Mason 已提交
3523
		ret = btrfs_readpage(NULL, page);
C
Chris Mason 已提交
3524
		lock_page(page);
3525 3526 3527 3528 3529
		if (page->mapping != mapping) {
			unlock_page(page);
			page_cache_release(page);
			goto again;
		}
C
Chris Mason 已提交
3530 3531
		if (!PageUptodate(page)) {
			ret = -EIO;
3532
			goto out_unlock;
C
Chris Mason 已提交
3533 3534
		}
	}
3535
	wait_on_page_writeback(page);
3536

3537
	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3538 3539 3540 3541
	set_page_extent_mapped(page);

	ordered = btrfs_lookup_ordered_extent(inode, page_start);
	if (ordered) {
3542 3543
		unlock_extent_cached(io_tree, page_start, page_end,
				     &cached_state, GFP_NOFS);
3544 3545
		unlock_page(page);
		page_cache_release(page);
3546
		btrfs_start_ordered_extent(inode, ordered, 1);
3547 3548 3549 3550
		btrfs_put_ordered_extent(ordered);
		goto again;
	}

3551
	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3552 3553
			  EXTENT_DIRTY | EXTENT_DELALLOC |
			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3554
			  0, 0, &cached_state, GFP_NOFS);
3555

3556 3557
	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
					&cached_state);
J
Josef Bacik 已提交
3558
	if (ret) {
3559 3560
		unlock_extent_cached(io_tree, page_start, page_end,
				     &cached_state, GFP_NOFS);
J
Josef Bacik 已提交
3561 3562 3563
		goto out_unlock;
	}

3564 3565
	ret = 0;
	if (offset != PAGE_CACHE_SIZE) {
J
Josef Bacik 已提交
3566 3567
		if (!len)
			len = PAGE_CACHE_SIZE - offset;
3568
		kaddr = kmap(page);
J
Josef Bacik 已提交
3569 3570 3571 3572
		if (front)
			memset(kaddr, 0, offset);
		else
			memset(kaddr + offset, 0, len);
3573 3574 3575
		flush_dcache_page(page);
		kunmap(page);
	}
3576
	ClearPageChecked(page);
3577
	set_page_dirty(page);
3578 3579
	unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
			     GFP_NOFS);
C
Chris Mason 已提交
3580

3581
out_unlock:
3582
	if (ret)
3583
		btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
C
Chris Mason 已提交
3584 3585 3586 3587 3588 3589
	unlock_page(page);
	page_cache_release(page);
out:
	return ret;
}

3590 3591 3592 3593 3594 3595
/*
 * This function puts in dummy file extents for the area we're creating a hole
 * for.  So if we are truncating this file to a larger size we need to insert
 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
 * the range between oldsize and size
 */
3596
int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
C
Chris Mason 已提交
3597
{
Y
Yan Zheng 已提交
3598 3599 3600
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3601
	struct extent_map *em = NULL;
3602
	struct extent_state *cached_state = NULL;
J
Josef Bacik 已提交
3603
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
Y
Yan Zheng 已提交
3604
	u64 mask = root->sectorsize - 1;
3605
	u64 hole_start = (oldsize + mask) & ~mask;
Y
Yan Zheng 已提交
3606 3607 3608 3609
	u64 block_end = (size + mask) & ~mask;
	u64 last_byte;
	u64 cur_offset;
	u64 hole_size;
J
Josef Bacik 已提交
3610
	int err = 0;
C
Chris Mason 已提交
3611

Y
Yan Zheng 已提交
3612 3613 3614 3615 3616 3617 3618
	if (size <= hole_start)
		return 0;

	while (1) {
		struct btrfs_ordered_extent *ordered;
		btrfs_wait_ordered_range(inode, hole_start,
					 block_end - hole_start);
3619
		lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3620
				 &cached_state);
Y
Yan Zheng 已提交
3621 3622 3623
		ordered = btrfs_lookup_ordered_extent(inode, hole_start);
		if (!ordered)
			break;
3624 3625
		unlock_extent_cached(io_tree, hole_start, block_end - 1,
				     &cached_state, GFP_NOFS);
Y
Yan Zheng 已提交
3626 3627
		btrfs_put_ordered_extent(ordered);
	}
C
Chris Mason 已提交
3628

Y
Yan Zheng 已提交
3629 3630 3631 3632
	cur_offset = hole_start;
	while (1) {
		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
				block_end - cur_offset, 0);
3633 3634 3635 3636
		if (IS_ERR(em)) {
			err = PTR_ERR(em);
			break;
		}
Y
Yan Zheng 已提交
3637 3638
		last_byte = min(extent_map_end(em), block_end);
		last_byte = (last_byte + mask) & ~mask;
3639
		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
J
Josef Bacik 已提交
3640
			struct extent_map *hole_em;
Y
Yan Zheng 已提交
3641
			hole_size = last_byte - cur_offset;
J
Josef Bacik 已提交
3642

3643
			trans = btrfs_start_transaction(root, 3);
3644 3645
			if (IS_ERR(trans)) {
				err = PTR_ERR(trans);
J
Josef Bacik 已提交
3646
				break;
3647
			}
3648

J
Josef Bacik 已提交
3649 3650
			err = btrfs_drop_extents(trans, root, inode,
						 cur_offset,
3651
						 cur_offset + hole_size, 1);
3652
			if (err) {
3653
				btrfs_abort_transaction(trans, root, err);
3654
				btrfs_end_transaction(trans, root);
3655
				break;
3656
			}
3657

Y
Yan Zheng 已提交
3658
			err = btrfs_insert_file_extent(trans, root,
L
Li Zefan 已提交
3659
					btrfs_ino(inode), cur_offset, 0,
Y
Yan Zheng 已提交
3660 3661
					0, hole_size, 0, hole_size,
					0, 0, 0);
3662
			if (err) {
3663
				btrfs_abort_transaction(trans, root, err);
3664
				btrfs_end_transaction(trans, root);
3665
				break;
3666
			}
3667

J
Josef Bacik 已提交
3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684
			btrfs_drop_extent_cache(inode, cur_offset,
						cur_offset + hole_size - 1, 0);
			hole_em = alloc_extent_map();
			if (!hole_em) {
				set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
					&BTRFS_I(inode)->runtime_flags);
				goto next;
			}
			hole_em->start = cur_offset;
			hole_em->len = hole_size;
			hole_em->orig_start = cur_offset;

			hole_em->block_start = EXTENT_MAP_HOLE;
			hole_em->block_len = 0;
			hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
			hole_em->compress_type = BTRFS_COMPRESS_NONE;
			hole_em->generation = trans->transid;
3685

J
Josef Bacik 已提交
3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700
			while (1) {
				write_lock(&em_tree->lock);
				err = add_extent_mapping(em_tree, hole_em);
				if (!err)
					list_move(&hole_em->list,
						  &em_tree->modified_extents);
				write_unlock(&em_tree->lock);
				if (err != -EEXIST)
					break;
				btrfs_drop_extent_cache(inode, cur_offset,
							cur_offset +
							hole_size - 1, 0);
			}
			free_extent_map(hole_em);
next:
3701
			btrfs_update_inode(trans, root, inode);
3702
			btrfs_end_transaction(trans, root);
Y
Yan Zheng 已提交
3703 3704
		}
		free_extent_map(em);
3705
		em = NULL;
Y
Yan Zheng 已提交
3706
		cur_offset = last_byte;
3707
		if (cur_offset >= block_end)
Y
Yan Zheng 已提交
3708 3709
			break;
	}
3710

3711
	free_extent_map(em);
3712 3713
	unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
			     GFP_NOFS);
Y
Yan Zheng 已提交
3714 3715
	return err;
}
C
Chris Mason 已提交
3716

3717
static int btrfs_setsize(struct inode *inode, loff_t newsize)
3718
{
3719 3720
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
3721
	loff_t oldsize = i_size_read(inode);
3722 3723
	int ret;

3724
	if (newsize == oldsize)
3725 3726
		return 0;

3727 3728 3729
	if (newsize > oldsize) {
		truncate_pagecache(inode, oldsize, newsize);
		ret = btrfs_cont_expand(inode, oldsize, newsize);
3730
		if (ret)
3731 3732
			return ret;

3733 3734 3735 3736 3737 3738 3739
		trans = btrfs_start_transaction(root, 1);
		if (IS_ERR(trans))
			return PTR_ERR(trans);

		i_size_write(inode, newsize);
		btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
		ret = btrfs_update_inode(trans, root, inode);
3740
		btrfs_end_transaction(trans, root);
3741
	} else {
3742

3743 3744 3745 3746 3747 3748
		/*
		 * We're truncating a file that used to have good data down to
		 * zero. Make sure it gets into the ordered flush list so that
		 * any new writes get down to disk quickly.
		 */
		if (newsize == 0)
3749 3750
			set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
				&BTRFS_I(inode)->runtime_flags);
3751

3752 3753 3754
		/* we don't support swapfiles, so vmtruncate shouldn't fail */
		truncate_setsize(inode, newsize);
		ret = btrfs_truncate(inode);
3755 3756
	}

3757
	return ret;
3758 3759
}

Y
Yan Zheng 已提交
3760 3761 3762
static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
{
	struct inode *inode = dentry->d_inode;
L
Li Zefan 已提交
3763
	struct btrfs_root *root = BTRFS_I(inode)->root;
Y
Yan Zheng 已提交
3764
	int err;
C
Chris Mason 已提交
3765

L
Li Zefan 已提交
3766 3767 3768
	if (btrfs_root_readonly(root))
		return -EROFS;

Y
Yan Zheng 已提交
3769 3770 3771
	err = inode_change_ok(inode, attr);
	if (err)
		return err;
C
Chris Mason 已提交
3772

3773
	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3774
		err = btrfs_setsize(inode, attr->ia_size);
3775 3776
		if (err)
			return err;
C
Chris Mason 已提交
3777
	}
Y
Yan Zheng 已提交
3778

C
Christoph Hellwig 已提交
3779 3780
	if (attr->ia_valid) {
		setattr_copy(inode, attr);
3781
		inode_inc_iversion(inode);
3782
		err = btrfs_dirty_inode(inode);
C
Christoph Hellwig 已提交
3783

3784
		if (!err && attr->ia_valid & ATTR_MODE)
C
Christoph Hellwig 已提交
3785 3786
			err = btrfs_acl_chmod(inode);
	}
J
Josef Bacik 已提交
3787

C
Chris Mason 已提交
3788 3789
	return err;
}
3790

A
Al Viro 已提交
3791
void btrfs_evict_inode(struct inode *inode)
C
Chris Mason 已提交
3792 3793 3794
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(inode)->root;
3795
	struct btrfs_block_rsv *rsv, *global_rsv;
3796
	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3797
	unsigned long nr;
C
Chris Mason 已提交
3798 3799
	int ret;

3800 3801
	trace_btrfs_inode_evict(inode);

C
Chris Mason 已提交
3802
	truncate_inode_pages(&inode->i_data, 0);
3803
	if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3804
			       btrfs_is_free_space_inode(inode)))
A
Al Viro 已提交
3805 3806
		goto no_delete;

C
Chris Mason 已提交
3807
	if (is_bad_inode(inode)) {
3808
		btrfs_orphan_del(NULL, inode);
C
Chris Mason 已提交
3809 3810
		goto no_delete;
	}
A
Al Viro 已提交
3811
	/* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
C
Chris Mason 已提交
3812
	btrfs_wait_ordered_range(inode, 0, (u64)-1);
3813

3814
	if (root->fs_info->log_root_recovering) {
3815
		BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3816
				 &BTRFS_I(inode)->runtime_flags));
3817 3818 3819
		goto no_delete;
	}

3820 3821 3822 3823 3824
	if (inode->i_nlink > 0) {
		BUG_ON(btrfs_root_refs(&root->root_item) != 0);
		goto no_delete;
	}

3825
	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3826 3827 3828 3829
	if (!rsv) {
		btrfs_orphan_del(NULL, inode);
		goto no_delete;
	}
J
Josef Bacik 已提交
3830
	rsv->size = min_size;
3831
	rsv->failfast = 1;
3832
	global_rsv = &root->fs_info->global_block_rsv;
3833

3834
	btrfs_i_size_write(inode, 0);
3835

3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846
	/*
	 * This is a bit simpler than btrfs_truncate since
	 *
	 * 1) We've already reserved our space for our orphan item in the
	 *    unlink.
	 * 2) We're going to delete the inode item, so we don't need to update
	 *    it at all.
	 *
	 * So we just need to reserve some slack space in case we add bytes when
	 * doing the truncate.
	 */
3847
	while (1) {
3848
		ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3849 3850 3851 3852 3853 3854 3855 3856

		/*
		 * Try and steal from the global reserve since we will
		 * likely not use this space anyway, we want to try as
		 * hard as possible to get this to work.
		 */
		if (ret)
			ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3857 3858

		if (ret) {
3859
			printk(KERN_WARNING "Could not get space for a "
3860
			       "delete, will truncate on mount %d\n", ret);
3861 3862 3863
			btrfs_orphan_del(NULL, inode);
			btrfs_free_block_rsv(root, rsv);
			goto no_delete;
3864
		}
3865

3866 3867 3868 3869 3870
		trans = btrfs_start_transaction(root, 0);
		if (IS_ERR(trans)) {
			btrfs_orphan_del(NULL, inode);
			btrfs_free_block_rsv(root, rsv);
			goto no_delete;
3871
		}
3872

3873 3874
		trans->block_rsv = rsv;

3875
		ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3876
		if (ret != -ENOSPC)
3877
			break;
3878

3879 3880 3881 3882 3883
		nr = trans->blocks_used;
		btrfs_end_transaction(trans, root);
		trans = NULL;
		btrfs_btree_balance_dirty(root, nr);
	}
3884

3885 3886
	btrfs_free_block_rsv(root, rsv);

3887
	if (ret == 0) {
3888
		trans->block_rsv = root->orphan_block_rsv;
3889 3890 3891
		ret = btrfs_orphan_del(trans, inode);
		BUG_ON(ret);
	}
3892

3893
	trans->block_rsv = &root->fs_info->trans_block_rsv;
3894 3895
	if (!(root == root->fs_info->tree_root ||
	      root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
L
Li Zefan 已提交
3896
		btrfs_return_ino(root, btrfs_ino(inode));
3897

3898
	nr = trans->blocks_used;
3899
	btrfs_end_transaction(trans, root);
3900
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
3901
no_delete:
3902
	clear_inode(inode);
3903
	return;
C
Chris Mason 已提交
3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917
}

/*
 * this returns the key found in the dir entry in the location pointer.
 * If no dir entries were found, location->objectid is 0.
 */
static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
			       struct btrfs_key *location)
{
	const char *name = dentry->d_name.name;
	int namelen = dentry->d_name.len;
	struct btrfs_dir_item *di;
	struct btrfs_path *path;
	struct btrfs_root *root = BTRFS_I(dir)->root;
Y
Yan 已提交
3918
	int ret = 0;
C
Chris Mason 已提交
3919 3920

	path = btrfs_alloc_path();
3921 3922
	if (!path)
		return -ENOMEM;
3923

L
Li Zefan 已提交
3924
	di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
C
Chris Mason 已提交
3925
				    namelen, 0);
Y
Yan 已提交
3926 3927
	if (IS_ERR(di))
		ret = PTR_ERR(di);
C
Chris Mason 已提交
3928

3929
	if (IS_ERR_OR_NULL(di))
3930
		goto out_err;
C
Chris Mason 已提交
3931

3932
	btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
C
Chris Mason 已提交
3933 3934 3935
out:
	btrfs_free_path(path);
	return ret;
3936 3937 3938
out_err:
	location->objectid = 0;
	goto out;
C
Chris Mason 已提交
3939 3940 3941 3942 3943 3944 3945 3946
}

/*
 * when we hit a tree root in a directory, the btrfs part of the inode
 * needs to be changed to reflect the root directory of the tree root.  This
 * is kind of like crossing a mount point.
 */
static int fixup_tree_root_location(struct btrfs_root *root,
3947 3948 3949 3950
				    struct inode *dir,
				    struct dentry *dentry,
				    struct btrfs_key *location,
				    struct btrfs_root **sub_root)
C
Chris Mason 已提交
3951
{
3952 3953 3954 3955 3956 3957
	struct btrfs_path *path;
	struct btrfs_root *new_root;
	struct btrfs_root_ref *ref;
	struct extent_buffer *leaf;
	int ret;
	int err = 0;
C
Chris Mason 已提交
3958

3959 3960 3961 3962 3963
	path = btrfs_alloc_path();
	if (!path) {
		err = -ENOMEM;
		goto out;
	}
C
Chris Mason 已提交
3964

3965 3966 3967 3968 3969 3970 3971 3972 3973
	err = -ENOENT;
	ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
				  BTRFS_I(dir)->root->root_key.objectid,
				  location->objectid);
	if (ret) {
		if (ret < 0)
			err = ret;
		goto out;
	}
C
Chris Mason 已提交
3974

3975 3976
	leaf = path->nodes[0];
	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
L
Li Zefan 已提交
3977
	if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3978 3979
	    btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
		goto out;
C
Chris Mason 已提交
3980

3981 3982 3983 3984 3985 3986
	ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
				   (unsigned long)(ref + 1),
				   dentry->d_name.len);
	if (ret)
		goto out;

3987
	btrfs_release_path(path);
3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007

	new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
	if (IS_ERR(new_root)) {
		err = PTR_ERR(new_root);
		goto out;
	}

	if (btrfs_root_refs(&new_root->root_item) == 0) {
		err = -ENOENT;
		goto out;
	}

	*sub_root = new_root;
	location->objectid = btrfs_root_dirid(&new_root->root_item);
	location->type = BTRFS_INODE_ITEM_KEY;
	location->offset = 0;
	err = 0;
out:
	btrfs_free_path(path);
	return err;
C
Chris Mason 已提交
4008 4009
}

4010 4011 4012 4013
static void inode_tree_add(struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_inode *entry;
4014 4015
	struct rb_node **p;
	struct rb_node *parent;
L
Li Zefan 已提交
4016
	u64 ino = btrfs_ino(inode);
4017 4018 4019
again:
	p = &root->inode_tree.rb_node;
	parent = NULL;
4020

A
Al Viro 已提交
4021
	if (inode_unhashed(inode))
4022 4023
		return;

4024 4025 4026 4027 4028
	spin_lock(&root->inode_lock);
	while (*p) {
		parent = *p;
		entry = rb_entry(parent, struct btrfs_inode, rb_node);

L
Li Zefan 已提交
4029
		if (ino < btrfs_ino(&entry->vfs_inode))
4030
			p = &parent->rb_left;
L
Li Zefan 已提交
4031
		else if (ino > btrfs_ino(&entry->vfs_inode))
4032
			p = &parent->rb_right;
4033 4034
		else {
			WARN_ON(!(entry->vfs_inode.i_state &
A
Al Viro 已提交
4035
				  (I_WILL_FREE | I_FREEING)));
4036 4037 4038 4039
			rb_erase(parent, &root->inode_tree);
			RB_CLEAR_NODE(parent);
			spin_unlock(&root->inode_lock);
			goto again;
4040 4041 4042 4043 4044 4045 4046 4047 4048 4049
		}
	}
	rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
	rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
	spin_unlock(&root->inode_lock);
}

static void inode_tree_del(struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
4050
	int empty = 0;
4051

4052
	spin_lock(&root->inode_lock);
4053 4054 4055
	if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
		rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
		RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4056
		empty = RB_EMPTY_ROOT(&root->inode_tree);
4057
	}
4058
	spin_unlock(&root->inode_lock);
4059

4060 4061 4062 4063 4064 4065 4066 4067
	/*
	 * Free space cache has inodes in the tree root, but the tree root has a
	 * root_refs of 0, so this could end up dropping the tree root as a
	 * snapshot, so we need the extra !root->fs_info->tree_root check to
	 * make sure we don't drop it.
	 */
	if (empty && btrfs_root_refs(&root->root_item) == 0 &&
	    root != root->fs_info->tree_root) {
4068 4069 4070 4071 4072 4073 4074 4075 4076
		synchronize_srcu(&root->fs_info->subvol_srcu);
		spin_lock(&root->inode_lock);
		empty = RB_EMPTY_ROOT(&root->inode_tree);
		spin_unlock(&root->inode_lock);
		if (empty)
			btrfs_add_dead_root(root);
	}
}

4077
void btrfs_invalidate_inodes(struct btrfs_root *root)
4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094
{
	struct rb_node *node;
	struct rb_node *prev;
	struct btrfs_inode *entry;
	struct inode *inode;
	u64 objectid = 0;

	WARN_ON(btrfs_root_refs(&root->root_item) != 0);

	spin_lock(&root->inode_lock);
again:
	node = root->inode_tree.rb_node;
	prev = NULL;
	while (node) {
		prev = node;
		entry = rb_entry(node, struct btrfs_inode, rb_node);

L
Li Zefan 已提交
4095
		if (objectid < btrfs_ino(&entry->vfs_inode))
4096
			node = node->rb_left;
L
Li Zefan 已提交
4097
		else if (objectid > btrfs_ino(&entry->vfs_inode))
4098 4099 4100 4101 4102 4103 4104
			node = node->rb_right;
		else
			break;
	}
	if (!node) {
		while (prev) {
			entry = rb_entry(prev, struct btrfs_inode, rb_node);
L
Li Zefan 已提交
4105
			if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4106 4107 4108 4109 4110 4111 4112 4113
				node = prev;
				break;
			}
			prev = rb_next(prev);
		}
	}
	while (node) {
		entry = rb_entry(node, struct btrfs_inode, rb_node);
L
Li Zefan 已提交
4114
		objectid = btrfs_ino(&entry->vfs_inode) + 1;
4115 4116 4117 4118 4119 4120
		inode = igrab(&entry->vfs_inode);
		if (inode) {
			spin_unlock(&root->inode_lock);
			if (atomic_read(&inode->i_count) > 1)
				d_prune_aliases(inode);
			/*
4121
			 * btrfs_drop_inode will have it removed from
4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136
			 * the inode cache when its usage count
			 * hits zero.
			 */
			iput(inode);
			cond_resched();
			spin_lock(&root->inode_lock);
			goto again;
		}

		if (cond_resched_lock(&root->inode_lock))
			goto again;

		node = rb_next(node);
	}
	spin_unlock(&root->inode_lock);
4137 4138
}

4139 4140 4141 4142 4143
static int btrfs_init_locked_inode(struct inode *inode, void *p)
{
	struct btrfs_iget_args *args = p;
	inode->i_ino = args->ino;
	BTRFS_I(inode)->root = args->root;
C
Chris Mason 已提交
4144 4145 4146 4147 4148 4149
	return 0;
}

static int btrfs_find_actor(struct inode *inode, void *opaque)
{
	struct btrfs_iget_args *args = opaque;
L
Li Zefan 已提交
4150
	return args->ino == btrfs_ino(inode) &&
C
Chris Mason 已提交
4151
		args->root == BTRFS_I(inode)->root;
C
Chris Mason 已提交
4152 4153
}

4154 4155 4156
static struct inode *btrfs_iget_locked(struct super_block *s,
				       u64 objectid,
				       struct btrfs_root *root)
C
Chris Mason 已提交
4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168
{
	struct inode *inode;
	struct btrfs_iget_args args;
	args.ino = objectid;
	args.root = root;

	inode = iget5_locked(s, objectid, btrfs_find_actor,
			     btrfs_init_locked_inode,
			     (void *)&args);
	return inode;
}

B
Balaji Rao 已提交
4169 4170 4171 4172
/* Get an inode object given its location and corresponding root.
 * Returns in *is_new if the inode was read from disk
 */
struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4173
			 struct btrfs_root *root, int *new)
B
Balaji Rao 已提交
4174 4175 4176 4177 4178
{
	struct inode *inode;

	inode = btrfs_iget_locked(s, location->objectid, root);
	if (!inode)
4179
		return ERR_PTR(-ENOMEM);
B
Balaji Rao 已提交
4180 4181 4182 4183 4184

	if (inode->i_state & I_NEW) {
		BTRFS_I(inode)->root = root;
		memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
		btrfs_read_locked_inode(inode);
4185 4186 4187 4188 4189 4190
		if (!is_bad_inode(inode)) {
			inode_tree_add(inode);
			unlock_new_inode(inode);
			if (new)
				*new = 1;
		} else {
4191 4192 4193
			unlock_new_inode(inode);
			iput(inode);
			inode = ERR_PTR(-ESTALE);
4194 4195 4196
		}
	}

B
Balaji Rao 已提交
4197 4198 4199
	return inode;
}

4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210
static struct inode *new_simple_dir(struct super_block *s,
				    struct btrfs_key *key,
				    struct btrfs_root *root)
{
	struct inode *inode = new_inode(s);

	if (!inode)
		return ERR_PTR(-ENOMEM);

	BTRFS_I(inode)->root = root;
	memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4211
	set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4212 4213

	inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
L
Li Zefan 已提交
4214
	inode->i_op = &btrfs_dir_ro_inode_operations;
4215 4216 4217 4218 4219 4220 4221
	inode->i_fop = &simple_dir_operations;
	inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;

	return inode;
}

4222
struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
C
Chris Mason 已提交
4223
{
C
Chris Mason 已提交
4224
	struct inode *inode;
4225
	struct btrfs_root *root = BTRFS_I(dir)->root;
C
Chris Mason 已提交
4226 4227
	struct btrfs_root *sub_root = root;
	struct btrfs_key location;
4228
	int index;
4229
	int ret = 0;
C
Chris Mason 已提交
4230 4231 4232

	if (dentry->d_name.len > BTRFS_NAME_LEN)
		return ERR_PTR(-ENAMETOOLONG);
4233

4234 4235 4236 4237
	if (unlikely(d_need_lookup(dentry))) {
		memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
		kfree(dentry->d_fsdata);
		dentry->d_fsdata = NULL;
4238 4239
		/* This thing is hashed, drop it for now */
		d_drop(dentry);
4240 4241 4242
	} else {
		ret = btrfs_inode_by_name(dir, dentry, &location);
	}
4243

C
Chris Mason 已提交
4244 4245
	if (ret < 0)
		return ERR_PTR(ret);
4246

4247 4248 4249 4250
	if (location.objectid == 0)
		return NULL;

	if (location.type == BTRFS_INODE_ITEM_KEY) {
4251
		inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4252 4253 4254 4255 4256
		return inode;
	}

	BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);

4257
	index = srcu_read_lock(&root->fs_info->subvol_srcu);
4258 4259 4260 4261 4262 4263 4264 4265
	ret = fixup_tree_root_location(root, dir, dentry,
				       &location, &sub_root);
	if (ret < 0) {
		if (ret != -ENOENT)
			inode = ERR_PTR(ret);
		else
			inode = new_simple_dir(dir->i_sb, &location, sub_root);
	} else {
4266
		inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
C
Chris Mason 已提交
4267
	}
4268 4269
	srcu_read_unlock(&root->fs_info->subvol_srcu, index);

4270
	if (!IS_ERR(inode) && root != sub_root) {
4271 4272
		down_read(&root->fs_info->cleanup_work_sem);
		if (!(inode->i_sb->s_flags & MS_RDONLY))
4273
			ret = btrfs_orphan_cleanup(sub_root);
4274
		up_read(&root->fs_info->cleanup_work_sem);
4275 4276
		if (ret)
			inode = ERR_PTR(ret);
4277 4278
	}

4279 4280 4281
	return inode;
}

N
Nick Piggin 已提交
4282
static int btrfs_dentry_delete(const struct dentry *dentry)
4283 4284
{
	struct btrfs_root *root;
L
Li Zefan 已提交
4285
	struct inode *inode = dentry->d_inode;
4286

L
Li Zefan 已提交
4287 4288
	if (!inode && !IS_ROOT(dentry))
		inode = dentry->d_parent->d_inode;
4289

L
Li Zefan 已提交
4290 4291
	if (inode) {
		root = BTRFS_I(inode)->root;
4292 4293
		if (btrfs_root_refs(&root->root_item) == 0)
			return 1;
L
Li Zefan 已提交
4294 4295 4296

		if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
			return 1;
4297
	}
4298 4299 4300
	return 0;
}

4301 4302 4303 4304 4305 4306
static void btrfs_dentry_release(struct dentry *dentry)
{
	if (dentry->d_fsdata)
		kfree(dentry->d_fsdata);
}

4307
static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
A
Al Viro 已提交
4308
				   unsigned int flags)
4309
{
4310 4311 4312 4313 4314 4315 4316 4317 4318
	struct dentry *ret;

	ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
	if (unlikely(d_need_lookup(dentry))) {
		spin_lock(&dentry->d_lock);
		dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
		spin_unlock(&dentry->d_lock);
	}
	return ret;
C
Chris Mason 已提交
4319 4320
}

4321
unsigned char btrfs_filetype_table[] = {
C
Chris Mason 已提交
4322 4323 4324
	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
};

4325 4326
static int btrfs_real_readdir(struct file *filp, void *dirent,
			      filldir_t filldir)
C
Chris Mason 已提交
4327
{
4328
	struct inode *inode = filp->f_dentry->d_inode;
C
Chris Mason 已提交
4329 4330 4331 4332
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_item *item;
	struct btrfs_dir_item *di;
	struct btrfs_key key;
4333
	struct btrfs_key found_key;
C
Chris Mason 已提交
4334
	struct btrfs_path *path;
4335 4336
	struct list_head ins_list;
	struct list_head del_list;
C
Chris Mason 已提交
4337
	int ret;
4338
	struct extent_buffer *leaf;
C
Chris Mason 已提交
4339 4340 4341 4342 4343 4344 4345
	int slot;
	unsigned char d_type;
	int over = 0;
	u32 di_cur;
	u32 di_total;
	u32 di_len;
	int key_type = BTRFS_DIR_INDEX_KEY;
4346 4347 4348
	char tmp_name[32];
	char *name_ptr;
	int name_len;
4349
	int is_curr = 0;	/* filp->f_pos points to the current index? */
C
Chris Mason 已提交
4350 4351 4352 4353

	/* FIXME, use a real flag for deciding about the key type */
	if (root->fs_info->tree_root == root)
		key_type = BTRFS_DIR_ITEM_KEY;
4354

4355 4356
	/* special case for "." */
	if (filp->f_pos == 0) {
4357 4358
		over = filldir(dirent, ".", 1,
			       filp->f_pos, btrfs_ino(inode), DT_DIR);
4359 4360 4361 4362 4363 4364
		if (over)
			return 0;
		filp->f_pos = 1;
	}
	/* special case for .., just use the back ref */
	if (filp->f_pos == 1) {
4365
		u64 pino = parent_ino(filp->f_path.dentry);
4366
		over = filldir(dirent, "..", 2,
4367
			       filp->f_pos, pino, DT_DIR);
4368
		if (over)
4369
			return 0;
4370 4371
		filp->f_pos = 2;
	}
4372
	path = btrfs_alloc_path();
4373 4374
	if (!path)
		return -ENOMEM;
C
Chris Mason 已提交
4375

J
Josef Bacik 已提交
4376
	path->reada = 1;
4377

4378 4379 4380 4381 4382 4383
	if (key_type == BTRFS_DIR_INDEX_KEY) {
		INIT_LIST_HEAD(&ins_list);
		INIT_LIST_HEAD(&del_list);
		btrfs_get_delayed_items(inode, &ins_list, &del_list);
	}

C
Chris Mason 已提交
4384 4385
	btrfs_set_key_type(&key, key_type);
	key.offset = filp->f_pos;
L
Li Zefan 已提交
4386
	key.objectid = btrfs_ino(inode);
4387

C
Chris Mason 已提交
4388 4389 4390
	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		goto err;
4391 4392

	while (1) {
4393
		leaf = path->nodes[0];
C
Chris Mason 已提交
4394
		slot = path->slots[0];
4395 4396 4397 4398 4399 4400 4401
		if (slot >= btrfs_header_nritems(leaf)) {
			ret = btrfs_next_leaf(root, path);
			if (ret < 0)
				goto err;
			else if (ret > 0)
				break;
			continue;
C
Chris Mason 已提交
4402
		}
4403

4404 4405 4406 4407
		item = btrfs_item_nr(leaf, slot);
		btrfs_item_key_to_cpu(leaf, &found_key, slot);

		if (found_key.objectid != key.objectid)
C
Chris Mason 已提交
4408
			break;
4409
		if (btrfs_key_type(&found_key) != key_type)
C
Chris Mason 已提交
4410
			break;
4411
		if (found_key.offset < filp->f_pos)
4412
			goto next;
4413 4414 4415 4416
		if (key_type == BTRFS_DIR_INDEX_KEY &&
		    btrfs_should_delete_dir_index(&del_list,
						  found_key.offset))
			goto next;
4417 4418

		filp->f_pos = found_key.offset;
4419
		is_curr = 1;
4420

C
Chris Mason 已提交
4421 4422
		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
		di_cur = 0;
4423
		di_total = btrfs_item_size(leaf, item);
4424 4425

		while (di_cur < di_total) {
4426 4427
			struct btrfs_key location;

4428 4429 4430
			if (verify_dir_item(root, leaf, di))
				break;

4431
			name_len = btrfs_dir_name_len(leaf, di);
4432
			if (name_len <= sizeof(tmp_name)) {
4433 4434 4435
				name_ptr = tmp_name;
			} else {
				name_ptr = kmalloc(name_len, GFP_NOFS);
4436 4437 4438 4439
				if (!name_ptr) {
					ret = -ENOMEM;
					goto err;
				}
4440 4441 4442 4443 4444 4445
			}
			read_extent_buffer(leaf, name_ptr,
					   (unsigned long)(di + 1), name_len);

			d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
			btrfs_dir_item_key_to_cpu(leaf, di, &location);
4446

4447

4448
			/* is this a reference to our own snapshot? If so
A
Arne Jansen 已提交
4449 4450 4451 4452 4453 4454 4455
			 * skip it.
			 *
			 * In contrast to old kernels, we insert the snapshot's
			 * dir item and dir index after it has been created, so
			 * we won't find a reference to our own snapshot. We
			 * still keep the following code for backward
			 * compatibility.
4456 4457 4458 4459 4460 4461
			 */
			if (location.type == BTRFS_ROOT_ITEM_KEY &&
			    location.objectid == root->root_key.objectid) {
				over = 0;
				goto skip;
			}
4462
			over = filldir(dirent, name_ptr, name_len,
4463
				       found_key.offset, location.objectid,
C
Chris Mason 已提交
4464
				       d_type);
4465

4466
skip:
4467 4468 4469
			if (name_ptr != tmp_name)
				kfree(name_ptr);

C
Chris Mason 已提交
4470 4471
			if (over)
				goto nopos;
J
Josef Bacik 已提交
4472
			di_len = btrfs_dir_name_len(leaf, di) +
4473
				 btrfs_dir_data_len(leaf, di) + sizeof(*di);
C
Chris Mason 已提交
4474 4475 4476
			di_cur += di_len;
			di = (struct btrfs_dir_item *)((char *)di + di_len);
		}
4477 4478
next:
		path->slots[0]++;
C
Chris Mason 已提交
4479
	}
4480

4481 4482 4483 4484 4485 4486 4487 4488 4489
	if (key_type == BTRFS_DIR_INDEX_KEY) {
		if (is_curr)
			filp->f_pos++;
		ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
						      &ins_list);
		if (ret)
			goto nopos;
	}

4490
	/* Reached end of directory/root. Bump pos past the last item. */
4491
	if (key_type == BTRFS_DIR_INDEX_KEY)
4492 4493 4494 4495 4496
		/*
		 * 32-bit glibc will use getdents64, but then strtol -
		 * so the last number we can serve is this.
		 */
		filp->f_pos = 0x7fffffff;
4497 4498
	else
		filp->f_pos++;
C
Chris Mason 已提交
4499 4500 4501
nopos:
	ret = 0;
err:
4502 4503
	if (key_type == BTRFS_DIR_INDEX_KEY)
		btrfs_put_delayed_items(&ins_list, &del_list);
C
Chris Mason 已提交
4504 4505 4506 4507
	btrfs_free_path(path);
	return ret;
}

4508
int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
C
Chris Mason 已提交
4509 4510 4511 4512
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
	int ret = 0;
4513
	bool nolock = false;
C
Chris Mason 已提交
4514

4515
	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4516 4517
		return 0;

4518
	if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4519
		nolock = true;
4520

4521
	if (wbc->sync_mode == WB_SYNC_ALL) {
4522
		if (nolock)
4523
			trans = btrfs_join_transaction_nolock(root);
4524
		else
4525
			trans = btrfs_join_transaction(root);
4526 4527
		if (IS_ERR(trans))
			return PTR_ERR(trans);
4528 4529 4530 4531
		if (nolock)
			ret = btrfs_end_transaction_nolock(trans, root);
		else
			ret = btrfs_commit_transaction(trans, root);
C
Chris Mason 已提交
4532 4533 4534 4535 4536
	}
	return ret;
}

/*
4537
 * This is somewhat expensive, updating the tree every time the
C
Chris Mason 已提交
4538 4539 4540 4541
 * inode changes.  But, it is most likely to find the inode in cache.
 * FIXME, needs more benchmarking...there are no reasons other than performance
 * to keep or drop this code.
 */
4542
int btrfs_dirty_inode(struct inode *inode)
C
Chris Mason 已提交
4543 4544 4545
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
4546 4547
	int ret;

4548
	if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4549
		return 0;
C
Chris Mason 已提交
4550

4551
	trans = btrfs_join_transaction(root);
4552 4553
	if (IS_ERR(trans))
		return PTR_ERR(trans);
4554 4555

	ret = btrfs_update_inode(trans, root, inode);
4556 4557 4558 4559
	if (ret && ret == -ENOSPC) {
		/* whoops, lets try again with the full transaction */
		btrfs_end_transaction(trans, root);
		trans = btrfs_start_transaction(root, 1);
4560 4561
		if (IS_ERR(trans))
			return PTR_ERR(trans);
4562

4563 4564
		ret = btrfs_update_inode(trans, root, inode);
	}
C
Chris Mason 已提交
4565
	btrfs_end_transaction(trans, root);
4566 4567
	if (BTRFS_I(inode)->delayed_node)
		btrfs_balance_delayed_items(root);
4568 4569 4570 4571 4572 4573 4574 4575

	return ret;
}

/*
 * This is a copy of file_update_time.  We need this so we can return error on
 * ENOSPC for updating the inode in the case of file write and mmap writes.
 */
4576 4577
static int btrfs_update_time(struct inode *inode, struct timespec *now,
			     int flags)
4578
{
4579 4580 4581 4582 4583
	struct btrfs_root *root = BTRFS_I(inode)->root;

	if (btrfs_root_readonly(root))
		return -EROFS;

4584
	if (flags & S_VERSION)
4585
		inode_inc_iversion(inode);
4586 4587 4588 4589 4590 4591 4592
	if (flags & S_CTIME)
		inode->i_ctime = *now;
	if (flags & S_MTIME)
		inode->i_mtime = *now;
	if (flags & S_ATIME)
		inode->i_atime = *now;
	return btrfs_dirty_inode(inode);
C
Chris Mason 已提交
4593 4594
}

C
Chris Mason 已提交
4595 4596 4597 4598 4599
/*
 * find the highest existing sequence number in a directory
 * and then set the in-memory index_cnt variable to reflect
 * free sequence numbers
 */
4600 4601 4602 4603 4604 4605 4606 4607
static int btrfs_set_inode_index_count(struct inode *inode)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_key key, found_key;
	struct btrfs_path *path;
	struct extent_buffer *leaf;
	int ret;

L
Li Zefan 已提交
4608
	key.objectid = btrfs_ino(inode);
4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639
	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
	key.offset = (u64)-1;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
	if (ret < 0)
		goto out;
	/* FIXME: we should be able to handle this */
	if (ret == 0)
		goto out;
	ret = 0;

	/*
	 * MAGIC NUMBER EXPLANATION:
	 * since we search a directory based on f_pos we have to start at 2
	 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
	 * else has to start at 2
	 */
	if (path->slots[0] == 0) {
		BTRFS_I(inode)->index_cnt = 2;
		goto out;
	}

	path->slots[0]--;

	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);

L
Li Zefan 已提交
4640
	if (found_key.objectid != btrfs_ino(inode) ||
4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651
	    btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
		BTRFS_I(inode)->index_cnt = 2;
		goto out;
	}

	BTRFS_I(inode)->index_cnt = found_key.offset + 1;
out:
	btrfs_free_path(path);
	return ret;
}

C
Chris Mason 已提交
4652 4653 4654 4655
/*
 * helper to find a free sequence number in a given directory.  This current
 * code is very simple, later versions will do smarter things in the btree
 */
4656
int btrfs_set_inode_index(struct inode *dir, u64 *index)
4657 4658 4659 4660
{
	int ret = 0;

	if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4661 4662 4663 4664 4665 4666
		ret = btrfs_inode_delayed_dir_index_count(dir);
		if (ret) {
			ret = btrfs_set_inode_index_count(dir);
			if (ret)
				return ret;
		}
4667 4668
	}

4669
	*index = BTRFS_I(dir)->index_cnt;
4670 4671 4672 4673 4674
	BTRFS_I(dir)->index_cnt++;

	return ret;
}

C
Chris Mason 已提交
4675 4676
static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
				     struct btrfs_root *root,
4677
				     struct inode *dir,
4678
				     const char *name, int name_len,
A
Al Viro 已提交
4679 4680
				     u64 ref_objectid, u64 objectid,
				     umode_t mode, u64 *index)
C
Chris Mason 已提交
4681 4682
{
	struct inode *inode;
4683
	struct btrfs_inode_item *inode_item;
C
Chris Mason 已提交
4684
	struct btrfs_key *location;
4685
	struct btrfs_path *path;
4686 4687 4688 4689
	struct btrfs_inode_ref *ref;
	struct btrfs_key key[2];
	u32 sizes[2];
	unsigned long ptr;
C
Chris Mason 已提交
4690 4691 4692
	int ret;
	int owner;

4693
	path = btrfs_alloc_path();
4694 4695
	if (!path)
		return ERR_PTR(-ENOMEM);
4696

C
Chris Mason 已提交
4697
	inode = new_inode(root->fs_info->sb);
4698 4699
	if (!inode) {
		btrfs_free_path(path);
C
Chris Mason 已提交
4700
		return ERR_PTR(-ENOMEM);
4701
	}
C
Chris Mason 已提交
4702

4703 4704 4705 4706 4707 4708
	/*
	 * we have to initialize this early, so we can reclaim the inode
	 * number if we fail afterwards in this function.
	 */
	inode->i_ino = objectid;

4709
	if (dir) {
4710 4711
		trace_btrfs_inode_request(dir);

4712
		ret = btrfs_set_inode_index(dir, index);
4713
		if (ret) {
4714
			btrfs_free_path(path);
4715
			iput(inode);
4716
			return ERR_PTR(ret);
4717
		}
4718 4719 4720 4721 4722 4723 4724
	}
	/*
	 * index_cnt is ignored for everything but a dir,
	 * btrfs_get_inode_index_count has an explanation for the magic
	 * number
	 */
	BTRFS_I(inode)->index_cnt = 2;
C
Chris Mason 已提交
4725
	BTRFS_I(inode)->root = root;
4726
	BTRFS_I(inode)->generation = trans->transid;
4727
	inode->i_generation = BTRFS_I(inode)->generation;
4728

J
Josef Bacik 已提交
4729 4730 4731 4732 4733 4734 4735 4736
	/*
	 * We could have gotten an inode number from somebody who was fsynced
	 * and then removed in this same transaction, so let's just set full
	 * sync since it will be a full sync anyway and this will blow away the
	 * old info in the log.
	 */
	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);

4737
	if (S_ISDIR(mode))
C
Chris Mason 已提交
4738 4739 4740
		owner = 0;
	else
		owner = 1;
4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752

	key[0].objectid = objectid;
	btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
	key[0].offset = 0;

	key[1].objectid = objectid;
	btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
	key[1].offset = ref_objectid;

	sizes[0] = sizeof(struct btrfs_inode_item);
	sizes[1] = name_len + sizeof(*ref);

4753
	path->leave_spinning = 1;
4754 4755
	ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
	if (ret != 0)
4756 4757
		goto fail;

4758
	inode_init_owner(inode, dir, mode);
4759
	inode_set_bytes(inode, 0);
C
Chris Mason 已提交
4760
	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4761 4762
	inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
				  struct btrfs_inode_item);
4763 4764
	memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
			     sizeof(*inode_item));
4765
	fill_inode_item(trans, path->nodes[0], inode_item, inode);
4766 4767 4768 4769

	ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
			     struct btrfs_inode_ref);
	btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4770
	btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4771 4772 4773
	ptr = (unsigned long)(ref + 1);
	write_extent_buffer(path->nodes[0], name, ptr, name_len);

4774 4775 4776
	btrfs_mark_buffer_dirty(path->nodes[0]);
	btrfs_free_path(path);

C
Chris Mason 已提交
4777 4778 4779 4780 4781
	location = &BTRFS_I(inode)->location;
	location->objectid = objectid;
	location->offset = 0;
	btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);

4782 4783
	btrfs_inherit_iflags(inode, dir);

4784
	if (S_ISREG(mode)) {
4785 4786
		if (btrfs_test_opt(root, NODATASUM))
			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4787 4788
		if (btrfs_test_opt(root, NODATACOW) ||
		    (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4789 4790 4791
			BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
	}

C
Chris Mason 已提交
4792
	insert_inode_hash(inode);
4793
	inode_tree_add(inode);
4794 4795

	trace_btrfs_inode_new(inode);
4796
	btrfs_set_inode_last_trans(trans, inode);
4797

4798 4799
	btrfs_update_root_times(trans, root);

C
Chris Mason 已提交
4800
	return inode;
4801
fail:
4802 4803
	if (dir)
		BTRFS_I(dir)->index_cnt--;
4804
	btrfs_free_path(path);
4805
	iput(inode);
4806
	return ERR_PTR(ret);
C
Chris Mason 已提交
4807 4808 4809 4810 4811 4812 4813
}

static inline u8 btrfs_inode_type(struct inode *inode)
{
	return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
}

C
Chris Mason 已提交
4814 4815 4816 4817 4818 4819
/*
 * utility function to add 'inode' into 'parent_inode' with
 * a give name and a given sequence number.
 * if 'add_backref' is true, also insert a backref from the
 * inode to the parent directory.
 */
4820 4821 4822
int btrfs_add_link(struct btrfs_trans_handle *trans,
		   struct inode *parent_inode, struct inode *inode,
		   const char *name, int name_len, int add_backref, u64 index)
C
Chris Mason 已提交
4823
{
4824
	int ret = 0;
C
Chris Mason 已提交
4825
	struct btrfs_key key;
4826
	struct btrfs_root *root = BTRFS_I(parent_inode)->root;
L
Li Zefan 已提交
4827 4828
	u64 ino = btrfs_ino(inode);
	u64 parent_ino = btrfs_ino(parent_inode);
4829

L
Li Zefan 已提交
4830
	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4831 4832
		memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
	} else {
L
Li Zefan 已提交
4833
		key.objectid = ino;
4834 4835 4836 4837
		btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
		key.offset = 0;
	}

L
Li Zefan 已提交
4838
	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4839 4840
		ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
					 key.objectid, root->root_key.objectid,
L
Li Zefan 已提交
4841
					 parent_ino, index, name, name_len);
4842
	} else if (add_backref) {
L
Li Zefan 已提交
4843 4844
		ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
					     parent_ino, index);
4845
	}
C
Chris Mason 已提交
4846

4847 4848 4849
	/* Nothing to clean up yet */
	if (ret)
		return ret;
4850

4851 4852 4853 4854 4855 4856 4857 4858
	ret = btrfs_insert_dir_item(trans, root, name, name_len,
				    parent_inode, &key,
				    btrfs_inode_type(inode), index);
	if (ret == -EEXIST)
		goto fail_dir_item;
	else if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		return ret;
C
Chris Mason 已提交
4859
	}
4860 4861 4862

	btrfs_i_size_write(parent_inode, parent_inode->i_size +
			   name_len * 2);
4863
	inode_inc_iversion(parent_inode);
4864 4865 4866 4867
	parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
	ret = btrfs_update_inode(trans, root, parent_inode);
	if (ret)
		btrfs_abort_transaction(trans, root, ret);
C
Chris Mason 已提交
4868
	return ret;
4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885

fail_dir_item:
	if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
		u64 local_index;
		int err;
		err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
				 key.objectid, root->root_key.objectid,
				 parent_ino, &local_index, name, name_len);

	} else if (add_backref) {
		u64 local_index;
		int err;

		err = btrfs_del_inode_ref(trans, root, name, name_len,
					  ino, parent_ino, &local_index);
	}
	return ret;
C
Chris Mason 已提交
4886 4887 4888
}

static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4889 4890
			    struct inode *dir, struct dentry *dentry,
			    struct inode *inode, int backref, u64 index)
C
Chris Mason 已提交
4891
{
4892 4893 4894
	int err = btrfs_add_link(trans, dir, inode,
				 dentry->d_name.name, dentry->d_name.len,
				 backref, index);
C
Chris Mason 已提交
4895 4896 4897 4898 4899
	if (err > 0)
		err = -EEXIST;
	return err;
}

J
Josef Bacik 已提交
4900
static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
A
Al Viro 已提交
4901
			umode_t mode, dev_t rdev)
J
Josef Bacik 已提交
4902 4903 4904
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(dir)->root;
4905
	struct inode *inode = NULL;
J
Josef Bacik 已提交
4906 4907 4908
	int err;
	int drop_inode = 0;
	u64 objectid;
4909
	unsigned long nr = 0;
4910
	u64 index = 0;
J
Josef Bacik 已提交
4911 4912 4913 4914

	if (!new_valid_dev(rdev))
		return -EINVAL;

J
Josef Bacik 已提交
4915 4916 4917 4918 4919
	/*
	 * 2 for inode item and ref
	 * 2 for dir items
	 * 1 for xattr if selinux is on
	 */
4920 4921 4922
	trans = btrfs_start_transaction(root, 5);
	if (IS_ERR(trans))
		return PTR_ERR(trans);
4923

4924 4925 4926 4927
	err = btrfs_find_free_ino(root, &objectid);
	if (err)
		goto out_unlock;

4928
	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
L
Li Zefan 已提交
4929
				dentry->d_name.len, btrfs_ino(dir), objectid,
4930
				mode, &index);
4931 4932
	if (IS_ERR(inode)) {
		err = PTR_ERR(inode);
J
Josef Bacik 已提交
4933
		goto out_unlock;
4934
	}
J
Josef Bacik 已提交
4935

4936
	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
J
Josef Bacik 已提交
4937 4938 4939 4940 4941
	if (err) {
		drop_inode = 1;
		goto out_unlock;
	}

4942 4943 4944 4945 4946 4947 4948 4949
	/*
	* If the active LSM wants to access the inode during
	* d_instantiate it needs these. Smack checks to see
	* if the filesystem supports xattrs by looking at the
	* ops vector.
	*/

	inode->i_op = &btrfs_special_inode_operations;
4950
	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
J
Josef Bacik 已提交
4951 4952 4953 4954
	if (err)
		drop_inode = 1;
	else {
		init_special_inode(inode, inode->i_mode, rdev);
4955
		btrfs_update_inode(trans, root, inode);
4956
		d_instantiate(dentry, inode);
J
Josef Bacik 已提交
4957 4958
	}
out_unlock:
4959
	nr = trans->blocks_used;
4960
	btrfs_end_transaction(trans, root);
4961
	btrfs_btree_balance_dirty(root, nr);
J
Josef Bacik 已提交
4962 4963 4964 4965 4966 4967 4968
	if (drop_inode) {
		inode_dec_link_count(inode);
		iput(inode);
	}
	return err;
}

C
Chris Mason 已提交
4969
static int btrfs_create(struct inode *dir, struct dentry *dentry,
A
Al Viro 已提交
4970
			umode_t mode, bool excl)
C
Chris Mason 已提交
4971 4972 4973
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(dir)->root;
4974
	struct inode *inode = NULL;
C
Chris Mason 已提交
4975
	int drop_inode = 0;
4976
	int err;
4977
	unsigned long nr = 0;
C
Chris Mason 已提交
4978
	u64 objectid;
4979
	u64 index = 0;
C
Chris Mason 已提交
4980

J
Josef Bacik 已提交
4981 4982 4983 4984 4985
	/*
	 * 2 for inode item and ref
	 * 2 for dir items
	 * 1 for xattr if selinux is on
	 */
4986 4987 4988
	trans = btrfs_start_transaction(root, 5);
	if (IS_ERR(trans))
		return PTR_ERR(trans);
J
Josef Bacik 已提交
4989

4990 4991 4992 4993
	err = btrfs_find_free_ino(root, &objectid);
	if (err)
		goto out_unlock;

4994
	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
L
Li Zefan 已提交
4995
				dentry->d_name.len, btrfs_ino(dir), objectid,
4996
				mode, &index);
4997 4998
	if (IS_ERR(inode)) {
		err = PTR_ERR(inode);
C
Chris Mason 已提交
4999
		goto out_unlock;
5000
	}
C
Chris Mason 已提交
5001

5002
	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
J
Josef Bacik 已提交
5003 5004 5005 5006 5007
	if (err) {
		drop_inode = 1;
		goto out_unlock;
	}

5008 5009 5010 5011 5012 5013 5014 5015 5016
	/*
	* If the active LSM wants to access the inode during
	* d_instantiate it needs these. Smack checks to see
	* if the filesystem supports xattrs by looking at the
	* ops vector.
	*/
	inode->i_fop = &btrfs_file_operations;
	inode->i_op = &btrfs_file_inode_operations;

5017
	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
C
Chris Mason 已提交
5018 5019 5020 5021
	if (err)
		drop_inode = 1;
	else {
		inode->i_mapping->a_ops = &btrfs_aops;
C
Chris Mason 已提交
5022
		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5023
		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5024
		d_instantiate(dentry, inode);
C
Chris Mason 已提交
5025 5026
	}
out_unlock:
5027
	nr = trans->blocks_used;
5028
	btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
5029 5030 5031 5032
	if (drop_inode) {
		inode_dec_link_count(inode);
		iput(inode);
	}
5033
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
5034 5035 5036 5037 5038 5039 5040 5041 5042
	return err;
}

static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
		      struct dentry *dentry)
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(dir)->root;
	struct inode *inode = old_dentry->d_inode;
5043
	u64 index;
5044
	unsigned long nr = 0;
C
Chris Mason 已提交
5045 5046 5047
	int err;
	int drop_inode = 0;

5048 5049
	/* do not allow sys_link's with other subvols of the same device */
	if (root->objectid != BTRFS_I(inode)->root->objectid)
5050
		return -EXDEV;
5051

5052 5053
	if (inode->i_nlink == ~0U)
		return -EMLINK;
5054

5055
	err = btrfs_set_inode_index(dir, &index);
5056 5057 5058
	if (err)
		goto fail;

5059
	/*
M
Miao Xie 已提交
5060
	 * 2 items for inode and inode ref
5061
	 * 2 items for dir items
M
Miao Xie 已提交
5062
	 * 1 item for parent inode
5063
	 */
M
Miao Xie 已提交
5064
	trans = btrfs_start_transaction(root, 5);
5065 5066 5067 5068
	if (IS_ERR(trans)) {
		err = PTR_ERR(trans);
		goto fail;
	}
5069

5070
	btrfs_inc_nlink(inode);
5071
	inode_inc_iversion(inode);
5072
	inode->i_ctime = CURRENT_TIME;
A
Al Viro 已提交
5073
	ihold(inode);
5074

5075
	err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5076

5077
	if (err) {
5078
		drop_inode = 1;
5079
	} else {
5080
		struct dentry *parent = dentry->d_parent;
5081
		err = btrfs_update_inode(trans, root, inode);
5082 5083
		if (err)
			goto fail;
5084
		d_instantiate(dentry, inode);
5085
		btrfs_log_new_name(trans, inode, NULL, parent);
5086
	}
C
Chris Mason 已提交
5087

5088
	nr = trans->blocks_used;
5089
	btrfs_end_transaction(trans, root);
5090
fail:
C
Chris Mason 已提交
5091 5092 5093 5094
	if (drop_inode) {
		inode_dec_link_count(inode);
		iput(inode);
	}
5095
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
5096 5097 5098
	return err;
}

5099
static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
C
Chris Mason 已提交
5100
{
5101
	struct inode *inode = NULL;
C
Chris Mason 已提交
5102 5103 5104 5105
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(dir)->root;
	int err = 0;
	int drop_on_err = 0;
5106
	u64 objectid = 0;
5107
	u64 index = 0;
5108
	unsigned long nr = 1;
C
Chris Mason 已提交
5109

J
Josef Bacik 已提交
5110 5111 5112 5113 5114
	/*
	 * 2 items for inode and ref
	 * 2 items for dir items
	 * 1 for xattr if selinux is on
	 */
5115 5116 5117
	trans = btrfs_start_transaction(root, 5);
	if (IS_ERR(trans))
		return PTR_ERR(trans);
C
Chris Mason 已提交
5118

5119 5120 5121 5122
	err = btrfs_find_free_ino(root, &objectid);
	if (err)
		goto out_fail;

5123
	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
L
Li Zefan 已提交
5124
				dentry->d_name.len, btrfs_ino(dir), objectid,
5125
				S_IFDIR | mode, &index);
C
Chris Mason 已提交
5126 5127 5128 5129
	if (IS_ERR(inode)) {
		err = PTR_ERR(inode);
		goto out_fail;
	}
5130

C
Chris Mason 已提交
5131
	drop_on_err = 1;
J
Josef Bacik 已提交
5132

5133
	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
J
Josef Bacik 已提交
5134 5135 5136
	if (err)
		goto out_fail;

C
Chris Mason 已提交
5137 5138 5139
	inode->i_op = &btrfs_dir_inode_operations;
	inode->i_fop = &btrfs_dir_file_operations;

5140
	btrfs_i_size_write(inode, 0);
C
Chris Mason 已提交
5141 5142 5143
	err = btrfs_update_inode(trans, root, inode);
	if (err)
		goto out_fail;
5144

5145 5146
	err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
			     dentry->d_name.len, 0, index);
C
Chris Mason 已提交
5147 5148
	if (err)
		goto out_fail;
5149

C
Chris Mason 已提交
5150 5151 5152 5153
	d_instantiate(dentry, inode);
	drop_on_err = 0;

out_fail:
5154
	nr = trans->blocks_used;
5155
	btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
5156 5157
	if (drop_on_err)
		iput(inode);
5158
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
5159 5160 5161
	return err;
}

C
Chris Mason 已提交
5162 5163 5164 5165
/* helper for btfs_get_extent.  Given an existing extent in the tree,
 * and an extent that you want to insert, deal with overlap and insert
 * the new extent into the tree.
 */
5166 5167
static int merge_extent_mapping(struct extent_map_tree *em_tree,
				struct extent_map *existing,
5168 5169
				struct extent_map *em,
				u64 map_start, u64 map_len)
5170 5171 5172
{
	u64 start_diff;

5173 5174 5175 5176
	BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
	start_diff = map_start - em->start;
	em->start = map_start;
	em->len = map_len;
C
Chris Mason 已提交
5177 5178
	if (em->block_start < EXTENT_MAP_LAST_BYTE &&
	    !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5179
		em->block_start += start_diff;
C
Chris Mason 已提交
5180 5181
		em->block_len -= start_diff;
	}
5182
	return add_extent_mapping(em_tree, em);
5183 5184
}

C
Chris Mason 已提交
5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195
static noinline int uncompress_inline(struct btrfs_path *path,
				      struct inode *inode, struct page *page,
				      size_t pg_offset, u64 extent_offset,
				      struct btrfs_file_extent_item *item)
{
	int ret;
	struct extent_buffer *leaf = path->nodes[0];
	char *tmp;
	size_t max_size;
	unsigned long inline_size;
	unsigned long ptr;
5196
	int compress_type;
C
Chris Mason 已提交
5197 5198

	WARN_ON(pg_offset != 0);
5199
	compress_type = btrfs_file_extent_compression(leaf, item);
C
Chris Mason 已提交
5200 5201 5202 5203
	max_size = btrfs_file_extent_ram_bytes(leaf, item);
	inline_size = btrfs_file_extent_inline_item_len(leaf,
					btrfs_item_nr(leaf, path->slots[0]));
	tmp = kmalloc(inline_size, GFP_NOFS);
5204 5205
	if (!tmp)
		return -ENOMEM;
C
Chris Mason 已提交
5206 5207 5208 5209
	ptr = btrfs_file_extent_inline_start(item);

	read_extent_buffer(leaf, tmp, ptr, inline_size);

5210
	max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5211 5212
	ret = btrfs_decompress(compress_type, tmp, page,
			       extent_offset, inline_size, max_size);
C
Chris Mason 已提交
5213
	if (ret) {
5214
		char *kaddr = kmap_atomic(page);
C
Chris Mason 已提交
5215 5216 5217 5218
		unsigned long copy_size = min_t(u64,
				  PAGE_CACHE_SIZE - pg_offset,
				  max_size - extent_offset);
		memset(kaddr + pg_offset, 0, copy_size);
5219
		kunmap_atomic(kaddr);
C
Chris Mason 已提交
5220 5221 5222 5223 5224
	}
	kfree(tmp);
	return 0;
}

C
Chris Mason 已提交
5225 5226
/*
 * a bit scary, this does extent mapping from logical file offset to the disk.
C
Chris Mason 已提交
5227 5228
 * the ugly parts come from merging extents from the disk with the in-ram
 * representation.  This gets more complex because of the data=ordered code,
C
Chris Mason 已提交
5229 5230 5231 5232
 * where the in-ram extents might be locked pending data=ordered completion.
 *
 * This also copies inline extents directly into the page.
 */
C
Chris Mason 已提交
5233

5234
struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5235
				    size_t pg_offset, u64 start, u64 len,
5236 5237 5238 5239
				    int create)
{
	int ret;
	int err = 0;
5240
	u64 bytenr;
5241 5242
	u64 extent_start = 0;
	u64 extent_end = 0;
L
Li Zefan 已提交
5243
	u64 objectid = btrfs_ino(inode);
5244
	u32 found_type;
5245
	struct btrfs_path *path = NULL;
5246 5247
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_file_extent_item *item;
5248 5249
	struct extent_buffer *leaf;
	struct btrfs_key found_key;
5250 5251
	struct extent_map *em = NULL;
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5252
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5253
	struct btrfs_trans_handle *trans = NULL;
5254
	int compress_type;
5255 5256

again:
5257
	read_lock(&em_tree->lock);
5258
	em = lookup_extent_mapping(em_tree, start, len);
5259 5260
	if (em)
		em->bdev = root->fs_info->fs_devices->latest_bdev;
5261
	read_unlock(&em_tree->lock);
5262

5263
	if (em) {
5264 5265 5266
		if (em->start > start || em->start + em->len <= start)
			free_extent_map(em);
		else if (em->block_start == EXTENT_MAP_INLINE && page)
5267 5268 5269
			free_extent_map(em);
		else
			goto out;
5270
	}
5271
	em = alloc_extent_map();
5272
	if (!em) {
5273 5274
		err = -ENOMEM;
		goto out;
5275
	}
5276
	em->bdev = root->fs_info->fs_devices->latest_bdev;
5277
	em->start = EXTENT_MAP_HOLE;
5278
	em->orig_start = EXTENT_MAP_HOLE;
5279
	em->len = (u64)-1;
C
Chris Mason 已提交
5280
	em->block_len = (u64)-1;
5281 5282 5283

	if (!path) {
		path = btrfs_alloc_path();
J
Josef Bacik 已提交
5284 5285 5286 5287 5288 5289 5290 5291 5292
		if (!path) {
			err = -ENOMEM;
			goto out;
		}
		/*
		 * Chances are we'll be called again, so go ahead and do
		 * readahead
		 */
		path->reada = 1;
5293 5294
	}

5295 5296
	ret = btrfs_lookup_file_extent(trans, root, path,
				       objectid, start, trans != NULL);
5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307
	if (ret < 0) {
		err = ret;
		goto out;
	}

	if (ret != 0) {
		if (path->slots[0] == 0)
			goto not_found;
		path->slots[0]--;
	}

5308 5309
	leaf = path->nodes[0];
	item = btrfs_item_ptr(leaf, path->slots[0],
5310 5311
			      struct btrfs_file_extent_item);
	/* are we inside the extent that was found? */
5312 5313 5314
	btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
	found_type = btrfs_key_type(&found_key);
	if (found_key.objectid != objectid ||
5315 5316 5317 5318
	    found_type != BTRFS_EXTENT_DATA_KEY) {
		goto not_found;
	}

5319 5320
	found_type = btrfs_file_extent_type(leaf, item);
	extent_start = found_key.offset;
5321
	compress_type = btrfs_file_extent_compression(leaf, item);
Y
Yan Zheng 已提交
5322 5323
	if (found_type == BTRFS_FILE_EXTENT_REG ||
	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5324
		extent_end = extent_start +
5325
		       btrfs_file_extent_num_bytes(leaf, item);
Y
Yan Zheng 已提交
5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339
	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
		size_t size;
		size = btrfs_file_extent_inline_len(leaf, item);
		extent_end = (extent_start + size + root->sectorsize - 1) &
			~((u64)root->sectorsize - 1);
	}

	if (start >= extent_end) {
		path->slots[0]++;
		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
			ret = btrfs_next_leaf(root, path);
			if (ret < 0) {
				err = ret;
				goto out;
5340
			}
Y
Yan Zheng 已提交
5341 5342 5343
			if (ret > 0)
				goto not_found;
			leaf = path->nodes[0];
5344
		}
Y
Yan Zheng 已提交
5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355
		btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
		if (found_key.objectid != objectid ||
		    found_key.type != BTRFS_EXTENT_DATA_KEY)
			goto not_found;
		if (start + len <= found_key.offset)
			goto not_found;
		em->start = start;
		em->len = found_key.offset - start;
		goto not_found_em;
	}

Y
Yan Zheng 已提交
5356 5357
	if (found_type == BTRFS_FILE_EXTENT_REG ||
	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
Y
Yan Zheng 已提交
5358 5359
		em->start = extent_start;
		em->len = extent_end - extent_start;
5360 5361
		em->orig_start = extent_start -
				 btrfs_file_extent_offset(leaf, item);
5362 5363
		bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
		if (bytenr == 0) {
5364
			em->block_start = EXTENT_MAP_HOLE;
5365 5366
			goto insert;
		}
5367
		if (compress_type != BTRFS_COMPRESS_NONE) {
C
Chris Mason 已提交
5368
			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5369
			em->compress_type = compress_type;
C
Chris Mason 已提交
5370 5371 5372 5373 5374 5375 5376
			em->block_start = bytenr;
			em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
									 item);
		} else {
			bytenr += btrfs_file_extent_offset(leaf, item);
			em->block_start = bytenr;
			em->block_len = em->len;
Y
Yan Zheng 已提交
5377 5378
			if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
				set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
C
Chris Mason 已提交
5379
		}
5380 5381
		goto insert;
	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5382
		unsigned long ptr;
5383
		char *map;
5384 5385 5386
		size_t size;
		size_t extent_offset;
		size_t copy_size;
5387

5388
		em->block_start = EXTENT_MAP_INLINE;
C
Chris Mason 已提交
5389
		if (!page || create) {
5390
			em->start = extent_start;
Y
Yan Zheng 已提交
5391
			em->len = extent_end - extent_start;
5392 5393
			goto out;
		}
5394

Y
Yan Zheng 已提交
5395 5396
		size = btrfs_file_extent_inline_len(leaf, item);
		extent_offset = page_offset(page) + pg_offset - extent_start;
5397
		copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5398 5399
				size - extent_offset);
		em->start = extent_start + extent_offset;
5400 5401
		em->len = (copy_size + root->sectorsize - 1) &
			~((u64)root->sectorsize - 1);
5402
		em->orig_start = EXTENT_MAP_INLINE;
5403
		if (compress_type) {
C
Chris Mason 已提交
5404
			set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5405 5406
			em->compress_type = compress_type;
		}
5407
		ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5408
		if (create == 0 && !PageUptodate(page)) {
5409 5410
			if (btrfs_file_extent_compression(leaf, item) !=
			    BTRFS_COMPRESS_NONE) {
C
Chris Mason 已提交
5411 5412 5413
				ret = uncompress_inline(path, inode, page,
							pg_offset,
							extent_offset, item);
5414
				BUG_ON(ret); /* -ENOMEM */
C
Chris Mason 已提交
5415 5416 5417 5418
			} else {
				map = kmap(page);
				read_extent_buffer(leaf, map + pg_offset, ptr,
						   copy_size);
5419 5420 5421 5422 5423
				if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
					memset(map + pg_offset + copy_size, 0,
					       PAGE_CACHE_SIZE - pg_offset -
					       copy_size);
				}
C
Chris Mason 已提交
5424 5425
				kunmap(page);
			}
5426 5427
			flush_dcache_page(page);
		} else if (create && PageUptodate(page)) {
5428
			BUG();
5429 5430 5431 5432
			if (!trans) {
				kunmap(page);
				free_extent_map(em);
				em = NULL;
C
Chris Mason 已提交
5433

5434
				btrfs_release_path(path);
5435
				trans = btrfs_join_transaction(root);
C
Chris Mason 已提交
5436

5437 5438
				if (IS_ERR(trans))
					return ERR_CAST(trans);
5439 5440
				goto again;
			}
C
Chris Mason 已提交
5441
			map = kmap(page);
5442
			write_extent_buffer(leaf, map + pg_offset, ptr,
5443
					    copy_size);
C
Chris Mason 已提交
5444
			kunmap(page);
5445
			btrfs_mark_buffer_dirty(leaf);
5446
		}
5447
		set_extent_uptodate(io_tree, em->start,
5448
				    extent_map_end(em) - 1, NULL, GFP_NOFS);
5449 5450
		goto insert;
	} else {
C
Chris Mason 已提交
5451
		printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5452 5453 5454 5455
		WARN_ON(1);
	}
not_found:
	em->start = start;
5456
	em->len = len;
5457
not_found_em:
5458
	em->block_start = EXTENT_MAP_HOLE;
Y
Yan Zheng 已提交
5459
	set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5460
insert:
5461
	btrfs_release_path(path);
5462
	if (em->start > start || extent_map_end(em) <= start) {
C
Chris Mason 已提交
5463 5464 5465 5466 5467
		printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
		       "[%llu %llu]\n", (unsigned long long)em->start,
		       (unsigned long long)em->len,
		       (unsigned long long)start,
		       (unsigned long long)len);
5468 5469 5470
		err = -EIO;
		goto out;
	}
5471 5472

	err = 0;
5473
	write_lock(&em_tree->lock);
5474
	ret = add_extent_mapping(em_tree, em);
5475 5476 5477 5478
	/* it is possible that someone inserted the extent into the tree
	 * while we had the lock dropped.  It is also possible that
	 * an overlapping map exists in the tree
	 */
5479
	if (ret == -EEXIST) {
5480
		struct extent_map *existing;
5481 5482 5483

		ret = 0;

5484
		existing = lookup_extent_mapping(em_tree, start, len);
5485 5486 5487 5488 5489
		if (existing && (existing->start > start ||
		    existing->start + existing->len <= start)) {
			free_extent_map(existing);
			existing = NULL;
		}
5490 5491 5492 5493 5494
		if (!existing) {
			existing = lookup_extent_mapping(em_tree, em->start,
							 em->len);
			if (existing) {
				err = merge_extent_mapping(em_tree, existing,
5495 5496
							   em, start,
							   root->sectorsize);
5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509
				free_extent_map(existing);
				if (err) {
					free_extent_map(em);
					em = NULL;
				}
			} else {
				err = -EIO;
				free_extent_map(em);
				em = NULL;
			}
		} else {
			free_extent_map(em);
			em = existing;
5510
			err = 0;
5511 5512
		}
	}
5513
	write_unlock(&em_tree->lock);
5514
out:
5515 5516 5517

	trace_btrfs_get_extent(root, em);

5518 5519
	if (path)
		btrfs_free_path(path);
5520 5521
	if (trans) {
		ret = btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
5522
		if (!err)
5523 5524 5525 5526 5527 5528
			err = ret;
	}
	if (err) {
		free_extent_map(em);
		return ERR_PTR(err);
	}
5529
	BUG_ON(!em); /* Error is always set */
5530 5531 5532
	return em;
}

5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594
struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
					   size_t pg_offset, u64 start, u64 len,
					   int create)
{
	struct extent_map *em;
	struct extent_map *hole_em = NULL;
	u64 range_start = start;
	u64 end;
	u64 found;
	u64 found_end;
	int err = 0;

	em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
	if (IS_ERR(em))
		return em;
	if (em) {
		/*
		 * if our em maps to a hole, there might
		 * actually be delalloc bytes behind it
		 */
		if (em->block_start != EXTENT_MAP_HOLE)
			return em;
		else
			hole_em = em;
	}

	/* check to see if we've wrapped (len == -1 or similar) */
	end = start + len;
	if (end < start)
		end = (u64)-1;
	else
		end -= 1;

	em = NULL;

	/* ok, we didn't find anything, lets look for delalloc */
	found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
				 end, len, EXTENT_DELALLOC, 1);
	found_end = range_start + found;
	if (found_end < range_start)
		found_end = (u64)-1;

	/*
	 * we didn't find anything useful, return
	 * the original results from get_extent()
	 */
	if (range_start > end || found_end <= start) {
		em = hole_em;
		hole_em = NULL;
		goto out;
	}

	/* adjust the range_start to make sure it doesn't
	 * go backwards from the start they passed in
	 */
	range_start = max(start,range_start);
	found = found_end - range_start;

	if (found > 0) {
		u64 hole_start = start;
		u64 hole_len = len;

5595
		em = alloc_extent_map();
5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654
		if (!em) {
			err = -ENOMEM;
			goto out;
		}
		/*
		 * when btrfs_get_extent can't find anything it
		 * returns one huge hole
		 *
		 * make sure what it found really fits our range, and
		 * adjust to make sure it is based on the start from
		 * the caller
		 */
		if (hole_em) {
			u64 calc_end = extent_map_end(hole_em);

			if (calc_end <= start || (hole_em->start > end)) {
				free_extent_map(hole_em);
				hole_em = NULL;
			} else {
				hole_start = max(hole_em->start, start);
				hole_len = calc_end - hole_start;
			}
		}
		em->bdev = NULL;
		if (hole_em && range_start > hole_start) {
			/* our hole starts before our delalloc, so we
			 * have to return just the parts of the hole
			 * that go until  the delalloc starts
			 */
			em->len = min(hole_len,
				      range_start - hole_start);
			em->start = hole_start;
			em->orig_start = hole_start;
			/*
			 * don't adjust block start at all,
			 * it is fixed at EXTENT_MAP_HOLE
			 */
			em->block_start = hole_em->block_start;
			em->block_len = hole_len;
		} else {
			em->start = range_start;
			em->len = found;
			em->orig_start = range_start;
			em->block_start = EXTENT_MAP_DELALLOC;
			em->block_len = found;
		}
	} else if (hole_em) {
		return hole_em;
	}
out:

	free_extent_map(hole_em);
	if (err) {
		free_extent_map(em);
		return ERR_PTR(err);
	}
	return em;
}

5655
static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5656
						  struct extent_map *em,
5657 5658 5659 5660 5661 5662 5663 5664
						  u64 start, u64 len)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_trans_handle *trans;
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	struct btrfs_key ins;
	u64 alloc_hint;
	int ret;
5665
	bool insert = false;
5666

5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679
	/*
	 * Ok if the extent map we looked up is a hole and is for the exact
	 * range we want, there is no reason to allocate a new one, however if
	 * it is not right then we need to free this one and drop the cache for
	 * our range.
	 */
	if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
	    em->len != len) {
		free_extent_map(em);
		em = NULL;
		insert = true;
		btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
	}
5680

5681
	trans = btrfs_join_transaction(root);
5682 5683
	if (IS_ERR(trans))
		return ERR_CAST(trans);
5684

C
Chris Mason 已提交
5685 5686 5687
	if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
		btrfs_add_inode_defrag(trans, inode);

5688 5689 5690 5691
	trans->block_rsv = &root->fs_info->delalloc_block_rsv;

	alloc_hint = get_extent_allocation_hint(inode, start, len);
	ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5692
				   alloc_hint, &ins, 1);
5693 5694 5695 5696 5697 5698
	if (ret) {
		em = ERR_PTR(ret);
		goto out;
	}

	if (!em) {
5699
		em = alloc_extent_map();
5700 5701 5702 5703
		if (!em) {
			em = ERR_PTR(-ENOMEM);
			goto out;
		}
5704 5705 5706 5707 5708 5709 5710 5711 5712
	}

	em->start = start;
	em->orig_start = em->start;
	em->len = ins.offset;

	em->block_start = ins.objectid;
	em->block_len = ins.offset;
	em->bdev = root->fs_info->fs_devices->latest_bdev;
5713 5714 5715 5716 5717 5718

	/*
	 * We need to do this because if we're using the original em we searched
	 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
	 */
	em->flags = 0;
5719 5720
	set_bit(EXTENT_FLAG_PINNED, &em->flags);

5721
	while (insert) {
5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740
		write_lock(&em_tree->lock);
		ret = add_extent_mapping(em_tree, em);
		write_unlock(&em_tree->lock);
		if (ret != -EEXIST)
			break;
		btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
	}

	ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
					   ins.offset, ins.offset, 0);
	if (ret) {
		btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
		em = ERR_PTR(ret);
	}
out:
	btrfs_end_transaction(trans, root);
	return em;
}

5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764
/*
 * returns 1 when the nocow is safe, < 1 on error, 0 if the
 * block must be cow'd
 */
static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
				      struct inode *inode, u64 offset, u64 len)
{
	struct btrfs_path *path;
	int ret;
	struct extent_buffer *leaf;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_file_extent_item *fi;
	struct btrfs_key key;
	u64 disk_bytenr;
	u64 backref_offset;
	u64 extent_end;
	u64 num_bytes;
	int slot;
	int found_type;

	path = btrfs_alloc_path();
	if (!path)
		return -ENOMEM;

L
Li Zefan 已提交
5765
	ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781
				       offset, 0);
	if (ret < 0)
		goto out;

	slot = path->slots[0];
	if (ret == 1) {
		if (slot == 0) {
			/* can't find the item, must cow */
			ret = 0;
			goto out;
		}
		slot--;
	}
	ret = 0;
	leaf = path->nodes[0];
	btrfs_item_key_to_cpu(leaf, &key, slot);
L
Li Zefan 已提交
5782
	if (key.objectid != btrfs_ino(inode) ||
5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815
	    key.type != BTRFS_EXTENT_DATA_KEY) {
		/* not our file or wrong item type, must cow */
		goto out;
	}

	if (key.offset > offset) {
		/* Wrong offset, must cow */
		goto out;
	}

	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
	found_type = btrfs_file_extent_type(leaf, fi);
	if (found_type != BTRFS_FILE_EXTENT_REG &&
	    found_type != BTRFS_FILE_EXTENT_PREALLOC) {
		/* not a regular extent, must cow */
		goto out;
	}
	disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
	backref_offset = btrfs_file_extent_offset(leaf, fi);

	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
	if (extent_end < offset + len) {
		/* extent doesn't include our full range, must cow */
		goto out;
	}

	if (btrfs_extent_readonly(root, disk_bytenr))
		goto out;

	/*
	 * look for other files referencing this extent, if we
	 * find any we must cow
	 */
L
Li Zefan 已提交
5816
	if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840
				  key.offset - backref_offset, disk_bytenr))
		goto out;

	/*
	 * adjust disk_bytenr and num_bytes to cover just the bytes
	 * in this extent we are about to write.  If there
	 * are any csums in that range we have to cow in order
	 * to keep the csums correct
	 */
	disk_bytenr += backref_offset;
	disk_bytenr += offset - key.offset;
	num_bytes = min(offset + len, extent_end) - offset;
	if (csum_exist_in_range(root, disk_bytenr, num_bytes))
				goto out;
	/*
	 * all of the above have passed, it is safe to overwrite this extent
	 * without cow
	 */
	ret = 1;
out:
	btrfs_free_path(path);
	return ret;
}

5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901
static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
			      struct extent_state **cached_state, int writing)
{
	struct btrfs_ordered_extent *ordered;
	int ret = 0;

	while (1) {
		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
				 0, cached_state);
		/*
		 * We're concerned with the entire range that we're going to be
		 * doing DIO to, so we need to make sure theres no ordered
		 * extents in this range.
		 */
		ordered = btrfs_lookup_ordered_range(inode, lockstart,
						     lockend - lockstart + 1);

		/*
		 * We need to make sure there are no buffered pages in this
		 * range either, we could have raced between the invalidate in
		 * generic_file_direct_write and locking the extent.  The
		 * invalidate needs to happen so that reads after a write do not
		 * get stale data.
		 */
		if (!ordered && (!writing ||
		    !test_range_bit(&BTRFS_I(inode)->io_tree,
				    lockstart, lockend, EXTENT_UPTODATE, 0,
				    *cached_state)))
			break;

		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
				     cached_state, GFP_NOFS);

		if (ordered) {
			btrfs_start_ordered_extent(inode, ordered, 1);
			btrfs_put_ordered_extent(ordered);
		} else {
			/* Screw you mmap */
			ret = filemap_write_and_wait_range(inode->i_mapping,
							   lockstart,
							   lockend);
			if (ret)
				break;

			/*
			 * If we found a page that couldn't be invalidated just
			 * fall back to buffered.
			 */
			ret = invalidate_inode_pages2_range(inode->i_mapping,
					lockstart >> PAGE_CACHE_SHIFT,
					lockend >> PAGE_CACHE_SHIFT);
			if (ret)
				break;
		}

		cond_resched();
	}

	return ret;
}

5902 5903 5904 5905 5906
static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
				   struct buffer_head *bh_result, int create)
{
	struct extent_map *em;
	struct btrfs_root *root = BTRFS_I(inode)->root;
5907
	struct extent_state *cached_state = NULL;
5908
	u64 start = iblock << inode->i_blkbits;
5909
	u64 lockstart, lockend;
5910
	u64 len = bh_result->b_size;
5911
	struct btrfs_trans_handle *trans;
5912 5913 5914 5915 5916 5917 5918 5919
	int unlock_bits = EXTENT_LOCKED;
	int ret;

	if (create) {
		ret = btrfs_delalloc_reserve_space(inode, len);
		if (ret)
			return ret;
		unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5920 5921
	} else {
		len = min_t(u64, len, root->sectorsize);
5922 5923
	}

5924 5925 5926
	lockstart = start;
	lockend = start + len - 1;

5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940
	/*
	 * If this errors out it's because we couldn't invalidate pagecache for
	 * this range and we need to fallback to buffered.
	 */
	if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
		return -ENOTBLK;

	if (create) {
		ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
				     lockend, EXTENT_DELALLOC, NULL,
				     &cached_state, GFP_NOFS);
		if (ret)
			goto unlock_err;
	}
5941 5942

	em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5943 5944 5945 5946
	if (IS_ERR(em)) {
		ret = PTR_ERR(em);
		goto unlock_err;
	}
5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964

	/*
	 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
	 * io.  INLINE is special, and we could probably kludge it in here, but
	 * it's still buffered so for safety lets just fall back to the generic
	 * buffered path.
	 *
	 * For COMPRESSED we _have_ to read the entire extent in so we can
	 * decompress it, so there will be buffering required no matter what we
	 * do, so go ahead and fallback to buffered.
	 *
	 * We return -ENOTBLK because thats what makes DIO go ahead and go back
	 * to buffered IO.  Don't blame me, this is the price we pay for using
	 * the generic code.
	 */
	if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
	    em->block_start == EXTENT_MAP_INLINE) {
		free_extent_map(em);
5965 5966
		ret = -ENOTBLK;
		goto unlock_err;
5967 5968 5969 5970 5971 5972
	}

	/* Just a good old fashioned hole, return */
	if (!create && (em->block_start == EXTENT_MAP_HOLE ||
			test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
		free_extent_map(em);
5973 5974
		ret = 0;
		goto unlock_err;
5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985
	}

	/*
	 * We don't allocate a new extent in the following cases
	 *
	 * 1) The inode is marked as NODATACOW.  In this case we'll just use the
	 * existing extent.
	 * 2) The extent is marked as PREALLOC.  We're good to go here and can
	 * just use the extent.
	 *
	 */
5986
	if (!create) {
5987 5988 5989
		len = min(len, em->len - (start - em->start));
		lockstart = start + len;
		goto unlock;
5990
	}
5991 5992 5993 5994 5995 5996

	if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
	    ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
	     em->block_start != EXTENT_MAP_HOLE)) {
		int type;
		int ret;
5997
		u64 block_start;
5998 5999 6000 6001 6002

		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
			type = BTRFS_ORDERED_PREALLOC;
		else
			type = BTRFS_ORDERED_NOCOW;
6003
		len = min(len, em->len - (start - em->start));
6004
		block_start = em->block_start + (start - em->start);
6005 6006 6007 6008 6009 6010

		/*
		 * we're not going to log anything, but we do need
		 * to make sure the current transaction stays open
		 * while we look for nocow cross refs
		 */
6011
		trans = btrfs_join_transaction(root);
6012
		if (IS_ERR(trans))
6013 6014 6015 6016 6017 6018 6019 6020
			goto must_cow;

		if (can_nocow_odirect(trans, inode, start, len) == 1) {
			ret = btrfs_add_ordered_extent_dio(inode, start,
					   block_start, len, len, type);
			btrfs_end_transaction(trans, root);
			if (ret) {
				free_extent_map(em);
6021
				goto unlock_err;
6022 6023
			}
			goto unlock;
6024
		}
6025
		btrfs_end_transaction(trans, root);
6026
	}
6027 6028 6029 6030 6031 6032
must_cow:
	/*
	 * this will cow the extent, reset the len in case we changed
	 * it above
	 */
	len = bh_result->b_size;
6033
	em = btrfs_new_extent_direct(inode, em, start, len);
6034 6035 6036 6037
	if (IS_ERR(em)) {
		ret = PTR_ERR(em);
		goto unlock_err;
	}
6038 6039
	len = min(len, em->len - (start - em->start));
unlock:
6040 6041
	bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
		inode->i_blkbits;
6042
	bh_result->b_size = len;
6043 6044
	bh_result->b_bdev = em->bdev;
	set_buffer_mapped(bh_result);
6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055
	if (create) {
		if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
			set_buffer_new(bh_result);

		/*
		 * Need to update the i_size under the extent lock so buffered
		 * readers will get the updated i_size when we unlock.
		 */
		if (start + len > i_size_read(inode))
			i_size_write(inode, start + len);
	}
6056

6057 6058 6059 6060 6061
	/*
	 * In the case of write we need to clear and unlock the entire range,
	 * in the case of read we need to unlock only the end area that we
	 * aren't using if there is any left over space.
	 */
L
Liu Bo 已提交
6062 6063 6064
	if (lockstart < lockend) {
		if (create && len < lockend - lockstart) {
			clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6065 6066
					 lockstart + len - 1,
					 unlock_bits | EXTENT_DEFRAG, 1, 0,
L
Liu Bo 已提交
6067 6068 6069 6070 6071 6072 6073
					 &cached_state, GFP_NOFS);
			/*
			 * Beside unlock, we also need to cleanup reserved space
			 * for the left range by attaching EXTENT_DO_ACCOUNTING.
			 */
			clear_extent_bit(&BTRFS_I(inode)->io_tree,
					 lockstart + len, lockend,
6074 6075
					 unlock_bits | EXTENT_DO_ACCOUNTING |
					 EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
L
Liu Bo 已提交
6076 6077 6078 6079 6080 6081
		} else {
			clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
					 lockend, unlock_bits, 1, 0,
					 &cached_state, GFP_NOFS);
		}
	} else {
6082
		free_extent_state(cached_state);
L
Liu Bo 已提交
6083
	}
6084

6085 6086 6087
	free_extent_map(em);

	return 0;
6088 6089 6090 6091 6092 6093 6094 6095

unlock_err:
	if (create)
		unlock_bits |= EXTENT_DO_ACCOUNTING;

	clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
			 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
	return ret;
6096 6097 6098 6099 6100 6101 6102 6103
}

struct btrfs_dio_private {
	struct inode *inode;
	u64 logical_offset;
	u64 disk_bytenr;
	u64 bytes;
	void *private;
M
Miao Xie 已提交
6104 6105 6106 6107 6108 6109 6110 6111

	/* number of bios pending for this dio */
	atomic_t pending_bios;

	/* IO errors */
	int errors;

	struct bio *orig_bio;
6112 6113 6114 6115
};

static void btrfs_endio_direct_read(struct bio *bio, int err)
{
M
Miao Xie 已提交
6116
	struct btrfs_dio_private *dip = bio->bi_private;
6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128
	struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
	struct bio_vec *bvec = bio->bi_io_vec;
	struct inode *inode = dip->inode;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	u64 start;

	start = dip->logical_offset;
	do {
		if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
			struct page *page = bvec->bv_page;
			char *kaddr;
			u32 csum = ~(u32)0;
6129
			u64 private = ~(u32)0;
6130 6131
			unsigned long flags;

6132 6133 6134
			if (get_state_private(&BTRFS_I(inode)->io_tree,
					      start, &private))
				goto failed;
6135
			local_irq_save(flags);
6136
			kaddr = kmap_atomic(page);
6137 6138 6139
			csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
					       csum, bvec->bv_len);
			btrfs_csum_final(csum, (char *)&csum);
6140
			kunmap_atomic(kaddr);
6141 6142 6143
			local_irq_restore(flags);

			flush_dcache_page(bvec->bv_page);
6144 6145
			if (csum != private) {
failed:
L
Li Zefan 已提交
6146
				printk(KERN_ERR "btrfs csum failed ino %llu off"
6147
				      " %llu csum %u private %u\n",
L
Li Zefan 已提交
6148 6149
				      (unsigned long long)btrfs_ino(inode),
				      (unsigned long long)start,
6150
				      csum, (unsigned)private);
6151 6152 6153 6154 6155 6156 6157 6158 6159
				err = -EIO;
			}
		}

		start += bvec->bv_len;
		bvec++;
	} while (bvec <= bvec_end);

	unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6160
		      dip->logical_offset + dip->bytes - 1);
6161 6162 6163
	bio->bi_private = dip->private;

	kfree(dip);
6164 6165 6166 6167

	/* If we had a csum failure make sure to clear the uptodate flag */
	if (err)
		clear_bit(BIO_UPTODATE, &bio->bi_flags);
6168 6169 6170 6171 6172 6173 6174 6175 6176
	dio_end_io(bio, err);
}

static void btrfs_endio_direct_write(struct bio *bio, int err)
{
	struct btrfs_dio_private *dip = bio->bi_private;
	struct inode *inode = dip->inode;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_ordered_extent *ordered = NULL;
6177 6178
	u64 ordered_offset = dip->logical_offset;
	u64 ordered_bytes = dip->bytes;
6179 6180 6181 6182
	int ret;

	if (err)
		goto out_done;
6183 6184 6185
again:
	ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
						   &ordered_offset,
6186
						   ordered_bytes, !err);
6187
	if (!ret)
6188
		goto out_test;
6189

6190 6191 6192 6193
	ordered->work.func = finish_ordered_fn;
	ordered->work.flags = 0;
	btrfs_queue_worker(&root->fs_info->endio_write_workers,
			   &ordered->work);
6194 6195 6196 6197 6198 6199 6200 6201
out_test:
	/*
	 * our bio might span multiple ordered extents.  If we haven't
	 * completed the accounting for the whole dio, go back and try again
	 */
	if (ordered_offset < dip->logical_offset + dip->bytes) {
		ordered_bytes = dip->logical_offset + dip->bytes -
			ordered_offset;
6202
		ordered = NULL;
6203 6204
		goto again;
	}
6205 6206 6207 6208
out_done:
	bio->bi_private = dip->private;

	kfree(dip);
6209 6210 6211 6212

	/* If we had an error make sure to clear the uptodate flag */
	if (err)
		clear_bit(BIO_UPTODATE, &bio->bi_flags);
6213 6214 6215
	dio_end_io(bio, err);
}

6216 6217 6218 6219 6220 6221 6222
static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
				    struct bio *bio, int mirror_num,
				    unsigned long bio_flags, u64 offset)
{
	int ret;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6223
	BUG_ON(ret); /* -ENOMEM */
6224 6225 6226
	return 0;
}

M
Miao Xie 已提交
6227 6228 6229 6230 6231
static void btrfs_end_dio_bio(struct bio *bio, int err)
{
	struct btrfs_dio_private *dip = bio->bi_private;

	if (err) {
L
Li Zefan 已提交
6232
		printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
J
Jan Beulich 已提交
6233
		      "sector %#Lx len %u err no %d\n",
L
Li Zefan 已提交
6234
		      (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
J
Jan Beulich 已提交
6235
		      (unsigned long long)bio->bi_sector, bio->bi_size, err);
M
Miao Xie 已提交
6236 6237 6238 6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267
		dip->errors = 1;

		/*
		 * before atomic variable goto zero, we must make sure
		 * dip->errors is perceived to be set.
		 */
		smp_mb__before_atomic_dec();
	}

	/* if there are more bios still pending for this dio, just exit */
	if (!atomic_dec_and_test(&dip->pending_bios))
		goto out;

	if (dip->errors)
		bio_io_error(dip->orig_bio);
	else {
		set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
		bio_endio(dip->orig_bio, 0);
	}
out:
	bio_put(bio);
}

static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
				       u64 first_sector, gfp_t gfp_flags)
{
	int nr_vecs = bio_get_nr_vecs(bdev);
	return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
}

static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
					 int rw, u64 file_offset, int skip_sum,
6268
					 int async_submit)
M
Miao Xie 已提交
6269 6270 6271 6272 6273 6274
{
	int write = rw & REQ_WRITE;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	int ret;

	bio_get(bio);
6275 6276 6277 6278 6279 6280

	if (!write) {
		ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
		if (ret)
			goto err;
	}
M
Miao Xie 已提交
6281

6282 6283 6284 6285
	if (skip_sum)
		goto map;

	if (write && async_submit) {
M
Miao Xie 已提交
6286 6287 6288 6289 6290 6291
		ret = btrfs_wq_submit_bio(root->fs_info,
				   inode, rw, bio, 0, 0,
				   file_offset,
				   __btrfs_submit_bio_start_direct_io,
				   __btrfs_submit_bio_done);
		goto err;
6292 6293 6294 6295 6296 6297 6298 6299
	} else if (write) {
		/*
		 * If we aren't doing async submit, calculate the csum of the
		 * bio now.
		 */
		ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
		if (ret)
			goto err;
6300
	} else if (!skip_sum) {
6301
		ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6302 6303 6304
		if (ret)
			goto err;
	}
M
Miao Xie 已提交
6305

6306 6307
map:
	ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
M
Miao Xie 已提交
6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327
err:
	bio_put(bio);
	return ret;
}

static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
				    int skip_sum)
{
	struct inode *inode = dip->inode;
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
	struct bio *bio;
	struct bio *orig_bio = dip->orig_bio;
	struct bio_vec *bvec = orig_bio->bi_io_vec;
	u64 start_sector = orig_bio->bi_sector;
	u64 file_offset = dip->logical_offset;
	u64 submit_len = 0;
	u64 map_length;
	int nr_pages = 0;
	int ret = 0;
6328
	int async_submit = 0;
M
Miao Xie 已提交
6329 6330 6331 6332 6333

	map_length = orig_bio->bi_size;
	ret = btrfs_map_block(map_tree, READ, start_sector << 9,
			      &map_length, NULL, 0);
	if (ret) {
6334
		bio_put(orig_bio);
M
Miao Xie 已提交
6335 6336 6337
		return -EIO;
	}

6338 6339 6340 6341 6342
	if (map_length >= orig_bio->bi_size) {
		bio = orig_bio;
		goto submit;
	}

6343
	async_submit = 1;
6344 6345 6346 6347 6348 6349 6350
	bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
	if (!bio)
		return -ENOMEM;
	bio->bi_private = dip;
	bio->bi_end_io = btrfs_end_dio_bio;
	atomic_inc(&dip->pending_bios);

M
Miao Xie 已提交
6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363
	while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
		if (unlikely(map_length < submit_len + bvec->bv_len ||
		    bio_add_page(bio, bvec->bv_page, bvec->bv_len,
				 bvec->bv_offset) < bvec->bv_len)) {
			/*
			 * inc the count before we submit the bio so
			 * we know the end IO handler won't happen before
			 * we inc the count. Otherwise, the dip might get freed
			 * before we're done setting it up
			 */
			atomic_inc(&dip->pending_bios);
			ret = __btrfs_submit_dio_bio(bio, inode, rw,
						     file_offset, skip_sum,
6364
						     async_submit);
M
Miao Xie 已提交
6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397
			if (ret) {
				bio_put(bio);
				atomic_dec(&dip->pending_bios);
				goto out_err;
			}

			start_sector += submit_len >> 9;
			file_offset += submit_len;

			submit_len = 0;
			nr_pages = 0;

			bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
						  start_sector, GFP_NOFS);
			if (!bio)
				goto out_err;
			bio->bi_private = dip;
			bio->bi_end_io = btrfs_end_dio_bio;

			map_length = orig_bio->bi_size;
			ret = btrfs_map_block(map_tree, READ, start_sector << 9,
					      &map_length, NULL, 0);
			if (ret) {
				bio_put(bio);
				goto out_err;
			}
		} else {
			submit_len += bvec->bv_len;
			nr_pages ++;
			bvec++;
		}
	}

6398
submit:
M
Miao Xie 已提交
6399
	ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6400
				     async_submit);
M
Miao Xie 已提交
6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418
	if (!ret)
		return 0;

	bio_put(bio);
out_err:
	dip->errors = 1;
	/*
	 * before atomic variable goto zero, we must
	 * make sure dip->errors is perceived to be set.
	 */
	smp_mb__before_atomic_dec();
	if (atomic_dec_and_test(&dip->pending_bios))
		bio_io_error(dip->orig_bio);

	/* bio_end_io() will handle error, so we needn't return it */
	return 0;
}

6419 6420 6421 6422 6423 6424 6425
static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
				loff_t file_offset)
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_dio_private *dip;
	struct bio_vec *bvec = bio->bi_io_vec;
	int skip_sum;
6426
	int write = rw & REQ_WRITE;
6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446
	int ret = 0;

	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;

	dip = kmalloc(sizeof(*dip), GFP_NOFS);
	if (!dip) {
		ret = -ENOMEM;
		goto free_ordered;
	}

	dip->private = bio->bi_private;
	dip->inode = inode;
	dip->logical_offset = file_offset;

	dip->bytes = 0;
	do {
		dip->bytes += bvec->bv_len;
		bvec++;
	} while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));

6447
	dip->disk_bytenr = (u64)bio->bi_sector << 9;
6448
	bio->bi_private = dip;
M
Miao Xie 已提交
6449 6450 6451
	dip->errors = 0;
	dip->orig_bio = bio;
	atomic_set(&dip->pending_bios, 0);
6452 6453 6454 6455 6456 6457

	if (write)
		bio->bi_end_io = btrfs_endio_direct_write;
	else
		bio->bi_end_io = btrfs_endio_direct_read;

M
Miao Xie 已提交
6458 6459
	ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
	if (!ret)
6460
		return;
6461 6462 6463 6464 6465 6466 6467
free_ordered:
	/*
	 * If this is a write, we need to clean up the reserved space and kill
	 * the ordered extent.
	 */
	if (write) {
		struct btrfs_ordered_extent *ordered;
6468
		ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6469 6470 6471 6472 6473 6474 6475 6476 6477 6478
		if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
		    !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
			btrfs_free_reserved_extent(root, ordered->start,
						   ordered->disk_len);
		btrfs_put_ordered_extent(ordered);
		btrfs_put_ordered_extent(ordered);
	}
	bio_endio(bio, ret);
}

C
Chris Mason 已提交
6479 6480 6481 6482 6483
static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
			const struct iovec *iov, loff_t offset,
			unsigned long nr_segs)
{
	int seg;
6484
	int i;
C
Chris Mason 已提交
6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498
	size_t size;
	unsigned long addr;
	unsigned blocksize_mask = root->sectorsize - 1;
	ssize_t retval = -EINVAL;
	loff_t end = offset;

	if (offset & blocksize_mask)
		goto out;

	/* Check the memory alignment.  Blocks cannot straddle pages */
	for (seg = 0; seg < nr_segs; seg++) {
		addr = (unsigned long)iov[seg].iov_base;
		size = iov[seg].iov_len;
		end += size;
6499
		if ((addr & blocksize_mask) || (size & blocksize_mask))
C
Chris Mason 已提交
6500
			goto out;
6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514

		/* If this is a write we don't need to check anymore */
		if (rw & WRITE)
			continue;

		/*
		 * Check to make sure we don't have duplicate iov_base's in this
		 * iovec, if so return EINVAL, otherwise we'll get csum errors
		 * when reading back.
		 */
		for (i = seg + 1; i < nr_segs; i++) {
			if (iov[seg].iov_base == iov[i].iov_base)
				goto out;
		}
C
Chris Mason 已提交
6515 6516 6517 6518 6519
	}
	retval = 0;
out:
	return retval;
}
6520

6521 6522 6523 6524
static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
			const struct iovec *iov, loff_t offset,
			unsigned long nr_segs)
{
6525 6526 6527
	struct file *file = iocb->ki_filp;
	struct inode *inode = file->f_mapping->host;

C
Chris Mason 已提交
6528
	if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6529
			    offset, nr_segs))
C
Chris Mason 已提交
6530
		return 0;
6531

6532
	return __blockdev_direct_IO(rw, iocb, inode,
C
Chris Mason 已提交
6533 6534 6535
		   BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
		   iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
		   btrfs_submit_direct, 0);
6536 6537
}

Y
Yehuda Sadeh 已提交
6538 6539 6540
static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
		__u64 start, __u64 len)
{
6541
	return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
Y
Yehuda Sadeh 已提交
6542 6543
}

6544
int btrfs_readpage(struct file *file, struct page *page)
C
Chris Mason 已提交
6545
{
6546 6547
	struct extent_io_tree *tree;
	tree = &BTRFS_I(page->mapping->host)->io_tree;
6548
	return extent_read_full_page(tree, page, btrfs_get_extent, 0);
C
Chris Mason 已提交
6549
}
6550

6551
static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
C
Chris Mason 已提交
6552
{
6553
	struct extent_io_tree *tree;
6554 6555 6556 6557 6558 6559 6560


	if (current->flags & PF_MEMALLOC) {
		redirty_page_for_writepage(wbc, page);
		unlock_page(page);
		return 0;
	}
6561
	tree = &BTRFS_I(page->mapping->host)->io_tree;
6562
	return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
C
Chris Mason 已提交
6563 6564
}

6565 6566
int btrfs_writepages(struct address_space *mapping,
		     struct writeback_control *wbc)
C
Chris Mason 已提交
6567
{
6568
	struct extent_io_tree *tree;
6569

6570
	tree = &BTRFS_I(mapping->host)->io_tree;
C
Chris Mason 已提交
6571 6572 6573
	return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
}

C
Chris Mason 已提交
6574 6575 6576 6577
static int
btrfs_readpages(struct file *file, struct address_space *mapping,
		struct list_head *pages, unsigned nr_pages)
{
6578 6579
	struct extent_io_tree *tree;
	tree = &BTRFS_I(mapping->host)->io_tree;
C
Chris Mason 已提交
6580 6581 6582
	return extent_readpages(tree, mapping, pages, nr_pages,
				btrfs_get_extent);
}
6583
static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
C
Chris Mason 已提交
6584
{
6585 6586
	struct extent_io_tree *tree;
	struct extent_map_tree *map;
6587
	int ret;
6588

6589 6590
	tree = &BTRFS_I(page->mapping->host)->io_tree;
	map = &BTRFS_I(page->mapping->host)->extent_tree;
6591
	ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6592 6593 6594 6595
	if (ret == 1) {
		ClearPagePrivate(page);
		set_page_private(page, 0);
		page_cache_release(page);
C
Chris Mason 已提交
6596
	}
6597
	return ret;
C
Chris Mason 已提交
6598 6599
}

6600 6601
static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
{
6602 6603
	if (PageWriteback(page) || PageDirty(page))
		return 0;
6604
	return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6605 6606
}

6607
static void btrfs_invalidatepage(struct page *page, unsigned long offset)
C
Chris Mason 已提交
6608
{
6609
	struct inode *inode = page->mapping->host;
6610
	struct extent_io_tree *tree;
6611
	struct btrfs_ordered_extent *ordered;
6612
	struct extent_state *cached_state = NULL;
6613 6614
	u64 page_start = page_offset(page);
	u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
C
Chris Mason 已提交
6615

6616 6617 6618 6619 6620 6621 6622
	/*
	 * we have the page locked, so new writeback can't start,
	 * and the dirty bit won't be cleared while we are here.
	 *
	 * Wait for IO on this page so that we can safely clear
	 * the PagePrivate2 bit and do ordered accounting
	 */
6623
	wait_on_page_writeback(page);
6624

6625
	tree = &BTRFS_I(inode)->io_tree;
6626 6627 6628 6629
	if (offset) {
		btrfs_releasepage(page, GFP_NOFS);
		return;
	}
6630
	lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6631
	ordered = btrfs_lookup_ordered_extent(inode,
6632 6633
					   page_offset(page));
	if (ordered) {
6634 6635 6636 6637
		/*
		 * IO on this page will never be started, so we need
		 * to account for any ordered extents now
		 */
6638 6639
		clear_extent_bit(tree, page_start, page_end,
				 EXTENT_DIRTY | EXTENT_DELALLOC |
6640 6641
				 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
				 EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6642 6643 6644 6645
		/*
		 * whoever cleared the private bit is responsible
		 * for the finish_ordered_io
		 */
6646 6647 6648 6649
		if (TestClearPagePrivate2(page) &&
		    btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
						   PAGE_CACHE_SIZE, 1)) {
			btrfs_finish_ordered_io(ordered);
6650
		}
6651
		btrfs_put_ordered_extent(ordered);
6652
		cached_state = NULL;
6653
		lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6654 6655
	}
	clear_extent_bit(tree, page_start, page_end,
6656
		 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6657 6658
		 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
		 &cached_state, GFP_NOFS);
6659 6660
	__btrfs_releasepage(page, GFP_NOFS);

C
Chris Mason 已提交
6661
	ClearPageChecked(page);
6662 6663 6664 6665 6666
	if (PagePrivate(page)) {
		ClearPagePrivate(page);
		set_page_private(page, 0);
		page_cache_release(page);
	}
C
Chris Mason 已提交
6667 6668
}

C
Chris Mason 已提交
6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682 6683
/*
 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
 * called from a page fault handler when a page is first dirtied. Hence we must
 * be careful to check for EOF conditions here. We set the page up correctly
 * for a written page which means we get ENOSPC checking when writing into
 * holes and correct delalloc and unwritten extent mapping on filesystems that
 * support these features.
 *
 * We are not allowed to take the i_mutex here so we have to play games to
 * protect against truncate races as the page could now be beyond EOF.  Because
 * vmtruncate() writes the inode size before removing pages, once we have the
 * page lock we can determine safely if the page is beyond EOF. If it is not
 * beyond EOF, then the page is guaranteed safe against truncation until we
 * unlock the page.
 */
6684
int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
C
Chris Mason 已提交
6685
{
6686
	struct page *page = vmf->page;
6687
	struct inode *inode = fdentry(vma->vm_file)->d_inode;
6688
	struct btrfs_root *root = BTRFS_I(inode)->root;
6689 6690
	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
	struct btrfs_ordered_extent *ordered;
6691
	struct extent_state *cached_state = NULL;
6692 6693
	char *kaddr;
	unsigned long zero_start;
C
Chris Mason 已提交
6694
	loff_t size;
6695
	int ret;
6696
	int reserved = 0;
6697
	u64 page_start;
6698
	u64 page_end;
C
Chris Mason 已提交
6699

6700
	sb_start_pagefault(inode->i_sb);
6701
	ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6702
	if (!ret) {
6703
		ret = file_update_time(vma->vm_file);
6704 6705
		reserved = 1;
	}
6706 6707 6708 6709 6710
	if (ret) {
		if (ret == -ENOMEM)
			ret = VM_FAULT_OOM;
		else /* -ENOSPC, -EIO, etc */
			ret = VM_FAULT_SIGBUS;
6711 6712 6713
		if (reserved)
			goto out;
		goto out_noreserve;
6714
	}
6715

6716
	ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6717
again:
C
Chris Mason 已提交
6718 6719
	lock_page(page);
	size = i_size_read(inode);
6720 6721
	page_start = page_offset(page);
	page_end = page_start + PAGE_CACHE_SIZE - 1;
6722

C
Chris Mason 已提交
6723
	if ((page->mapping != inode->i_mapping) ||
6724
	    (page_start >= size)) {
C
Chris Mason 已提交
6725 6726 6727
		/* page got truncated out from underneath us */
		goto out_unlock;
	}
6728 6729
	wait_on_page_writeback(page);

6730
	lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6731 6732
	set_page_extent_mapped(page);

6733 6734 6735 6736
	/*
	 * we can't set the delalloc bits if there are pending ordered
	 * extents.  Drop our locks and wait for them to finish
	 */
6737 6738
	ordered = btrfs_lookup_ordered_extent(inode, page_start);
	if (ordered) {
6739 6740
		unlock_extent_cached(io_tree, page_start, page_end,
				     &cached_state, GFP_NOFS);
6741
		unlock_page(page);
6742
		btrfs_start_ordered_extent(inode, ordered, 1);
6743 6744 6745 6746
		btrfs_put_ordered_extent(ordered);
		goto again;
	}

J
Josef Bacik 已提交
6747 6748 6749 6750 6751 6752 6753
	/*
	 * XXX - page_mkwrite gets called every time the page is dirtied, even
	 * if it was already dirty, so for space accounting reasons we need to
	 * clear any delalloc bits for the range we are fixing to save.  There
	 * is probably a better way to do this, but for now keep consistent with
	 * prepare_pages in the normal write path.
	 */
6754
	clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6755 6756
			  EXTENT_DIRTY | EXTENT_DELALLOC |
			  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6757
			  0, 0, &cached_state, GFP_NOFS);
J
Josef Bacik 已提交
6758

6759 6760
	ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
					&cached_state);
J
Josef Bacik 已提交
6761
	if (ret) {
6762 6763
		unlock_extent_cached(io_tree, page_start, page_end,
				     &cached_state, GFP_NOFS);
J
Josef Bacik 已提交
6764 6765 6766
		ret = VM_FAULT_SIGBUS;
		goto out_unlock;
	}
6767
	ret = 0;
C
Chris Mason 已提交
6768 6769

	/* page is wholly or partially inside EOF */
6770
	if (page_start + PAGE_CACHE_SIZE > size)
6771
		zero_start = size & ~PAGE_CACHE_MASK;
C
Chris Mason 已提交
6772
	else
6773
		zero_start = PAGE_CACHE_SIZE;
C
Chris Mason 已提交
6774

6775 6776 6777 6778 6779 6780
	if (zero_start != PAGE_CACHE_SIZE) {
		kaddr = kmap(page);
		memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
		flush_dcache_page(page);
		kunmap(page);
	}
6781
	ClearPageChecked(page);
6782
	set_page_dirty(page);
6783
	SetPageUptodate(page);
6784

6785 6786
	BTRFS_I(inode)->last_trans = root->fs_info->generation;
	BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6787
	BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6788

6789
	unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
C
Chris Mason 已提交
6790 6791

out_unlock:
6792 6793
	if (!ret) {
		sb_end_pagefault(inode->i_sb);
6794
		return VM_FAULT_LOCKED;
6795
	}
C
Chris Mason 已提交
6796
	unlock_page(page);
6797
out:
6798
	btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6799
out_noreserve:
6800
	sb_end_pagefault(inode->i_sb);
C
Chris Mason 已提交
6801 6802 6803
	return ret;
}

6804
static int btrfs_truncate(struct inode *inode)
C
Chris Mason 已提交
6805 6806
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
6807
	struct btrfs_block_rsv *rsv;
C
Chris Mason 已提交
6808
	int ret;
6809
	int err = 0;
C
Chris Mason 已提交
6810
	struct btrfs_trans_handle *trans;
6811
	unsigned long nr;
6812
	u64 mask = root->sectorsize - 1;
6813
	u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
C
Chris Mason 已提交
6814

J
Josef Bacik 已提交
6815
	ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6816
	if (ret)
6817
		return ret;
6818

C
Chris Mason 已提交
6819
	btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6820
	btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
C
Chris Mason 已提交
6821

6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845 6846 6847 6848 6849 6850 6851 6852 6853 6854 6855 6856 6857
	/*
	 * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
	 * 3 things going on here
	 *
	 * 1) We need to reserve space for our orphan item and the space to
	 * delete our orphan item.  Lord knows we don't want to have a dangling
	 * orphan item because we didn't reserve space to remove it.
	 *
	 * 2) We need to reserve space to update our inode.
	 *
	 * 3) We need to have something to cache all the space that is going to
	 * be free'd up by the truncate operation, but also have some slack
	 * space reserved in case it uses space during the truncate (thank you
	 * very much snapshotting).
	 *
	 * And we need these to all be seperate.  The fact is we can use alot of
	 * space doing the truncate, and we have no earthly idea how much space
	 * we will use, so we need the truncate reservation to be seperate so it
	 * doesn't end up using space reserved for updating the inode or
	 * removing the orphan item.  We also need to be able to stop the
	 * transaction and start a new one, which means we need to be able to
	 * update the inode several times, and we have no idea of knowing how
	 * many times that will be, so we can't just reserve 1 item for the
	 * entirety of the opration, so that has to be done seperately as well.
	 * Then there is the orphan item, which does indeed need to be held on
	 * to for the whole operation, and we need nobody to touch this reserved
	 * space except the orphan code.
	 *
	 * So that leaves us with
	 *
	 * 1) root->orphan_block_rsv - for the orphan deletion.
	 * 2) rsv - for the truncate reservation, which we will steal from the
	 * transaction reservation.
	 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
	 * updating the inode.
	 */
6858
	rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6859 6860
	if (!rsv)
		return -ENOMEM;
J
Josef Bacik 已提交
6861
	rsv->size = min_size;
6862
	rsv->failfast = 1;
6863

6864
	/*
6865
	 * 1 for the truncate slack space
6866 6867 6868 6869
	 * 1 for the orphan item we're going to add
	 * 1 for the orphan item deletion
	 * 1 for updating the inode.
	 */
6870 6871 6872 6873 6874
	trans = btrfs_start_transaction(root, 4);
	if (IS_ERR(trans)) {
		err = PTR_ERR(trans);
		goto out;
	}
6875

6876 6877 6878
	/* Migrate the slack space for the truncate to our reserve */
	ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
				      min_size);
6879
	BUG_ON(ret);
6880 6881 6882 6883

	ret = btrfs_orphan_add(trans, inode);
	if (ret) {
		btrfs_end_transaction(trans, root);
6884
		goto out;
6885 6886
	}

6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899 6900 6901 6902 6903
	/*
	 * setattr is responsible for setting the ordered_data_close flag,
	 * but that is only tested during the last file release.  That
	 * could happen well after the next commit, leaving a great big
	 * window where new writes may get lost if someone chooses to write
	 * to this file after truncating to zero
	 *
	 * The inode doesn't have any dirty data here, and so if we commit
	 * this is a noop.  If someone immediately starts writing to the inode
	 * it is very likely we'll catch some of their writes in this
	 * transaction, and the commit will find this file on the ordered
	 * data list with good things to send down.
	 *
	 * This is a best effort solution, there is still a window where
	 * using truncate to replace the contents of the file will
	 * end up with a zero length file after a crash.
	 */
6904 6905
	if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
					   &BTRFS_I(inode)->runtime_flags))
6906 6907
		btrfs_add_ordered_operation(trans, root, inode);

J
Josef Bacik 已提交
6908 6909 6910 6911 6912 6913 6914 6915
	/*
	 * So if we truncate and then write and fsync we normally would just
	 * write the extents that changed, which is a problem if we need to
	 * first truncate that entire inode.  So set this flag so we write out
	 * all of the extents in the inode to the sync log so we're completely
	 * safe.
	 */
	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6916
	trans->block_rsv = rsv;
J
Josef Bacik 已提交
6917

6918 6919 6920 6921
	while (1) {
		ret = btrfs_truncate_inode_items(trans, root, inode,
						 inode->i_size,
						 BTRFS_EXTENT_DATA_KEY);
6922
		if (ret != -ENOSPC) {
6923
			err = ret;
6924
			break;
6925
		}
C
Chris Mason 已提交
6926

6927
		trans->block_rsv = &root->fs_info->trans_block_rsv;
6928
		ret = btrfs_update_inode(trans, root, inode);
6929 6930 6931 6932
		if (ret) {
			err = ret;
			break;
		}
6933

6934 6935 6936
		nr = trans->blocks_used;
		btrfs_end_transaction(trans, root);
		btrfs_btree_balance_dirty(root, nr);
6937 6938 6939 6940 6941 6942 6943 6944 6945 6946 6947 6948

		trans = btrfs_start_transaction(root, 2);
		if (IS_ERR(trans)) {
			ret = err = PTR_ERR(trans);
			trans = NULL;
			break;
		}

		ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
					      rsv, min_size);
		BUG_ON(ret);	/* shouldn't happen */
		trans->block_rsv = rsv;
6949 6950 6951
	}

	if (ret == 0 && inode->i_nlink > 0) {
6952
		trans->block_rsv = root->orphan_block_rsv;
6953
		ret = btrfs_orphan_del(trans, inode);
6954 6955
		if (ret)
			err = ret;
6956 6957 6958 6959 6960 6961
	} else if (ret && inode->i_nlink > 0) {
		/*
		 * Failed to do the truncate, remove us from the in memory
		 * orphan list.
		 */
		ret = btrfs_orphan_del(NULL, inode);
6962 6963
	}

6964 6965 6966 6967 6968
	if (trans) {
		trans->block_rsv = &root->fs_info->trans_block_rsv;
		ret = btrfs_update_inode(trans, root, inode);
		if (ret && !err)
			err = ret;
6969

6970
		nr = trans->blocks_used;
6971
		ret = btrfs_end_transaction(trans, root);
6972 6973
		btrfs_btree_balance_dirty(root, nr);
	}
6974 6975 6976 6977

out:
	btrfs_free_block_rsv(root, rsv);

6978 6979
	if (ret && !err)
		err = ret;
6980

6981
	return err;
C
Chris Mason 已提交
6982 6983
}

C
Chris Mason 已提交
6984 6985 6986
/*
 * create a new subvolume directory/inode (helper for the ioctl).
 */
6987
int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6988
			     struct btrfs_root *new_root, u64 new_dirid)
C
Chris Mason 已提交
6989 6990
{
	struct inode *inode;
6991
	int err;
6992
	u64 index = 0;
C
Chris Mason 已提交
6993

6994 6995 6996 6997
	inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
				new_dirid, new_dirid,
				S_IFDIR | (~current_umask() & S_IRWXUGO),
				&index);
6998
	if (IS_ERR(inode))
C
Christoph Hellwig 已提交
6999
		return PTR_ERR(inode);
C
Chris Mason 已提交
7000 7001 7002
	inode->i_op = &btrfs_dir_inode_operations;
	inode->i_fop = &btrfs_dir_file_operations;

M
Miklos Szeredi 已提交
7003
	set_nlink(inode, 1);
7004
	btrfs_i_size_write(inode, 0);
7005

7006
	err = btrfs_update_inode(trans, new_root, inode);
7007

7008
	iput(inode);
7009
	return err;
C
Chris Mason 已提交
7010 7011 7012 7013 7014
}

struct inode *btrfs_alloc_inode(struct super_block *sb)
{
	struct btrfs_inode *ei;
Y
Yan, Zheng 已提交
7015
	struct inode *inode;
C
Chris Mason 已提交
7016 7017 7018 7019

	ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
	if (!ei)
		return NULL;
Y
Yan, Zheng 已提交
7020 7021 7022

	ei->root = NULL;
	ei->generation = 0;
7023
	ei->last_trans = 0;
7024
	ei->last_sub_trans = 0;
7025
	ei->logged_trans = 0;
Y
Yan, Zheng 已提交
7026 7027 7028
	ei->delalloc_bytes = 0;
	ei->disk_i_size = 0;
	ei->flags = 0;
7029
	ei->csum_bytes = 0;
Y
Yan, Zheng 已提交
7030 7031
	ei->index_cnt = (u64)-1;
	ei->last_unlink_trans = 0;
7032
	ei->last_log_commit = 0;
Y
Yan, Zheng 已提交
7033

7034 7035 7036
	spin_lock_init(&ei->lock);
	ei->outstanding_extents = 0;
	ei->reserved_extents = 0;
Y
Yan, Zheng 已提交
7037

7038
	ei->runtime_flags = 0;
7039
	ei->force_compress = BTRFS_COMPRESS_NONE;
Y
Yan, Zheng 已提交
7040

7041 7042
	ei->delayed_node = NULL;

Y
Yan, Zheng 已提交
7043
	inode = &ei->vfs_inode;
7044
	extent_map_tree_init(&ei->extent_tree);
7045 7046
	extent_io_tree_init(&ei->io_tree, &inode->i_data);
	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7047 7048
	ei->io_tree.track_uptodate = 1;
	ei->io_failure_tree.track_uptodate = 1;
Y
Yan, Zheng 已提交
7049
	mutex_init(&ei->log_mutex);
7050
	mutex_init(&ei->delalloc_mutex);
7051
	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
Y
Yan, Zheng 已提交
7052
	INIT_LIST_HEAD(&ei->delalloc_inodes);
7053
	INIT_LIST_HEAD(&ei->ordered_operations);
Y
Yan, Zheng 已提交
7054 7055 7056
	RB_CLEAR_NODE(&ei->rb_node);

	return inode;
C
Chris Mason 已提交
7057 7058
}

N
Nick Piggin 已提交
7059 7060 7061 7062 7063 7064
static void btrfs_i_callback(struct rcu_head *head)
{
	struct inode *inode = container_of(head, struct inode, i_rcu);
	kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
}

C
Chris Mason 已提交
7065 7066
void btrfs_destroy_inode(struct inode *inode)
{
7067
	struct btrfs_ordered_extent *ordered;
7068 7069
	struct btrfs_root *root = BTRFS_I(inode)->root;

A
Al Viro 已提交
7070
	WARN_ON(!hlist_empty(&inode->i_dentry));
C
Chris Mason 已提交
7071
	WARN_ON(inode->i_data.nrpages);
7072 7073
	WARN_ON(BTRFS_I(inode)->outstanding_extents);
	WARN_ON(BTRFS_I(inode)->reserved_extents);
7074 7075
	WARN_ON(BTRFS_I(inode)->delalloc_bytes);
	WARN_ON(BTRFS_I(inode)->csum_bytes);
C
Chris Mason 已提交
7076

7077 7078 7079 7080 7081 7082 7083 7084
	/*
	 * This can happen where we create an inode, but somebody else also
	 * created the same inode and we need to destroy the one we already
	 * created.
	 */
	if (!root)
		goto free;

7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095
	/*
	 * Make sure we're properly removed from the ordered operation
	 * lists.
	 */
	smp_mb();
	if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
		spin_lock(&root->fs_info->ordered_extent_lock);
		list_del_init(&BTRFS_I(inode)->ordered_operations);
		spin_unlock(&root->fs_info->ordered_extent_lock);
	}

7096 7097
	if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
		     &BTRFS_I(inode)->runtime_flags)) {
L
Li Zefan 已提交
7098 7099
		printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
		       (unsigned long long)btrfs_ino(inode));
7100
		atomic_dec(&root->orphan_inodes);
7101 7102
	}

C
Chris Mason 已提交
7103
	while (1) {
7104 7105 7106 7107
		ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
		if (!ordered)
			break;
		else {
C
Chris Mason 已提交
7108 7109 7110 7111
			printk(KERN_ERR "btrfs found ordered "
			       "extent %llu %llu on inode cleanup\n",
			       (unsigned long long)ordered->file_offset,
			       (unsigned long long)ordered->len);
7112 7113 7114 7115 7116
			btrfs_remove_ordered_extent(inode, ordered);
			btrfs_put_ordered_extent(ordered);
			btrfs_put_ordered_extent(ordered);
		}
	}
7117
	inode_tree_del(inode);
7118
	btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7119
free:
7120
	btrfs_remove_delayed_node(inode);
N
Nick Piggin 已提交
7121
	call_rcu(&inode->i_rcu, btrfs_i_callback);
C
Chris Mason 已提交
7122 7123
}

7124
int btrfs_drop_inode(struct inode *inode)
7125 7126
{
	struct btrfs_root *root = BTRFS_I(inode)->root;
7127

7128
	if (btrfs_root_refs(&root->root_item) == 0 &&
7129
	    !btrfs_is_free_space_inode(inode))
7130
		return 1;
7131
	else
7132
		return generic_drop_inode(inode);
7133 7134
}

7135
static void init_once(void *foo)
C
Chris Mason 已提交
7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147 7148 7149 7150 7151
{
	struct btrfs_inode *ei = (struct btrfs_inode *) foo;

	inode_init_once(&ei->vfs_inode);
}

void btrfs_destroy_cachep(void)
{
	if (btrfs_inode_cachep)
		kmem_cache_destroy(btrfs_inode_cachep);
	if (btrfs_trans_handle_cachep)
		kmem_cache_destroy(btrfs_trans_handle_cachep);
	if (btrfs_transaction_cachep)
		kmem_cache_destroy(btrfs_transaction_cachep);
	if (btrfs_path_cachep)
		kmem_cache_destroy(btrfs_path_cachep);
7152 7153
	if (btrfs_free_space_cachep)
		kmem_cache_destroy(btrfs_free_space_cachep);
C
Chris Mason 已提交
7154 7155 7156 7157
}

int btrfs_init_cachep(void)
{
D
David Sterba 已提交
7158
	btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7159 7160
			sizeof(struct btrfs_inode), 0,
			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
C
Chris Mason 已提交
7161 7162
	if (!btrfs_inode_cachep)
		goto fail;
7163

D
David Sterba 已提交
7164
	btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7165 7166
			sizeof(struct btrfs_trans_handle), 0,
			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
C
Chris Mason 已提交
7167 7168
	if (!btrfs_trans_handle_cachep)
		goto fail;
7169

D
David Sterba 已提交
7170
	btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7171 7172
			sizeof(struct btrfs_transaction), 0,
			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
C
Chris Mason 已提交
7173 7174
	if (!btrfs_transaction_cachep)
		goto fail;
7175

D
David Sterba 已提交
7176
	btrfs_path_cachep = kmem_cache_create("btrfs_path",
7177 7178
			sizeof(struct btrfs_path), 0,
			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
C
Chris Mason 已提交
7179 7180
	if (!btrfs_path_cachep)
		goto fail;
7181

D
David Sterba 已提交
7182
	btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7183 7184 7185 7186 7187
			sizeof(struct btrfs_free_space), 0,
			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
	if (!btrfs_free_space_cachep)
		goto fail;

C
Chris Mason 已提交
7188 7189 7190 7191 7192 7193 7194 7195 7196 7197
	return 0;
fail:
	btrfs_destroy_cachep();
	return -ENOMEM;
}

static int btrfs_getattr(struct vfsmount *mnt,
			 struct dentry *dentry, struct kstat *stat)
{
	struct inode *inode = dentry->d_inode;
D
David Sterba 已提交
7198 7199
	u32 blocksize = inode->i_sb->s_blocksize;

C
Chris Mason 已提交
7200
	generic_fillattr(inode, stat);
7201
	stat->dev = BTRFS_I(inode)->root->anon_dev;
C
Chris Mason 已提交
7202
	stat->blksize = PAGE_CACHE_SIZE;
D
David Sterba 已提交
7203 7204
	stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
		ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
C
Chris Mason 已提交
7205 7206 7207
	return 0;
}

7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221
/*
 * If a file is moved, it will inherit the cow and compression flags of the new
 * directory.
 */
static void fixup_inode_flags(struct inode *dir, struct inode *inode)
{
	struct btrfs_inode *b_dir = BTRFS_I(dir);
	struct btrfs_inode *b_inode = BTRFS_I(inode);

	if (b_dir->flags & BTRFS_INODE_NODATACOW)
		b_inode->flags |= BTRFS_INODE_NODATACOW;
	else
		b_inode->flags &= ~BTRFS_INODE_NODATACOW;

7222
	if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7223
		b_inode->flags |= BTRFS_INODE_COMPRESS;
7224 7225 7226 7227 7228
		b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
	} else {
		b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
				    BTRFS_INODE_NOCOMPRESS);
	}
7229 7230
}

C
Chris Mason 已提交
7231 7232
static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
			   struct inode *new_dir, struct dentry *new_dentry)
C
Chris Mason 已提交
7233 7234 7235
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(old_dir)->root;
7236
	struct btrfs_root *dest = BTRFS_I(new_dir)->root;
C
Chris Mason 已提交
7237 7238 7239
	struct inode *new_inode = new_dentry->d_inode;
	struct inode *old_inode = old_dentry->d_inode;
	struct timespec ctime = CURRENT_TIME;
7240
	u64 index = 0;
7241
	u64 root_objectid;
C
Chris Mason 已提交
7242
	int ret;
L
Li Zefan 已提交
7243
	u64 old_ino = btrfs_ino(old_inode);
C
Chris Mason 已提交
7244

L
Li Zefan 已提交
7245
	if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7246 7247
		return -EPERM;

7248
	/* we only allow rename subvolume link between subvolumes */
L
Li Zefan 已提交
7249
	if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7250 7251
		return -EXDEV;

L
Li Zefan 已提交
7252 7253
	if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
	    (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
C
Chris Mason 已提交
7254
		return -ENOTEMPTY;
7255

7256 7257 7258
	if (S_ISDIR(old_inode->i_mode) && new_inode &&
	    new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
		return -ENOTEMPTY;
7259 7260 7261 7262 7263
	/*
	 * we're using rename to replace one file with another.
	 * and the replacement file is large.  Start IO on it now so
	 * we don't add too much work to the end of the transaction
	 */
7264
	if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7265 7266 7267
	    old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
		filemap_flush(old_inode->i_mapping);

7268
	/* close the racy window with snapshot create/destroy ioctl */
L
Li Zefan 已提交
7269
	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7270
		down_read(&root->fs_info->subvol_sem);
7271 7272 7273 7274 7275 7276 7277 7278 7279
	/*
	 * We want to reserve the absolute worst case amount of items.  So if
	 * both inodes are subvols and we need to unlink them then that would
	 * require 4 item modifications, but if they are both normal inodes it
	 * would require 5 item modifications, so we'll assume their normal
	 * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
	 * should cover the worst case number of items we'll modify.
	 */
	trans = btrfs_start_transaction(root, 20);
7280 7281 7282 7283
	if (IS_ERR(trans)) {
                ret = PTR_ERR(trans);
                goto out_notrans;
        }
7284

7285 7286
	if (dest != root)
		btrfs_record_root_in_trans(trans, dest);
7287

7288 7289 7290
	ret = btrfs_set_inode_index(new_dir, &index);
	if (ret)
		goto out_fail;
7291

L
Li Zefan 已提交
7292
	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7293 7294 7295
		/* force full log commit if subvolume involved. */
		root->fs_info->last_trans_log_full_commit = trans->transid;
	} else {
7296 7297 7298
		ret = btrfs_insert_inode_ref(trans, dest,
					     new_dentry->d_name.name,
					     new_dentry->d_name.len,
L
Li Zefan 已提交
7299 7300
					     old_ino,
					     btrfs_ino(new_dir), index);
7301 7302
		if (ret)
			goto out_fail;
7303 7304 7305 7306 7307 7308 7309 7310 7311
		/*
		 * this is an ugly little race, but the rename is required
		 * to make sure that if we crash, the inode is either at the
		 * old name or the new one.  pinning the log transaction lets
		 * us make sure we don't allow a log commit to come in after
		 * we unlink the name but before we add the new name back in.
		 */
		btrfs_pin_log_trans(root);
	}
7312 7313 7314 7315
	/*
	 * make sure the inode gets flushed if it is replacing
	 * something.
	 */
L
Li Zefan 已提交
7316
	if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7317 7318
		btrfs_add_ordered_operation(trans, root, old_inode);

7319 7320 7321
	inode_inc_iversion(old_dir);
	inode_inc_iversion(new_dir);
	inode_inc_iversion(old_inode);
C
Chris Mason 已提交
7322 7323 7324
	old_dir->i_ctime = old_dir->i_mtime = ctime;
	new_dir->i_ctime = new_dir->i_mtime = ctime;
	old_inode->i_ctime = ctime;
7325

7326 7327 7328
	if (old_dentry->d_parent != new_dentry->d_parent)
		btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);

L
Li Zefan 已提交
7329
	if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7330 7331 7332 7333 7334
		root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
		ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
					old_dentry->d_name.name,
					old_dentry->d_name.len);
	} else {
7335 7336 7337 7338 7339 7340
		ret = __btrfs_unlink_inode(trans, root, old_dir,
					old_dentry->d_inode,
					old_dentry->d_name.name,
					old_dentry->d_name.len);
		if (!ret)
			ret = btrfs_update_inode(trans, root, old_inode);
7341
	}
7342 7343 7344 7345
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto out_fail;
	}
C
Chris Mason 已提交
7346 7347

	if (new_inode) {
7348
		inode_inc_iversion(new_inode);
C
Chris Mason 已提交
7349
		new_inode->i_ctime = CURRENT_TIME;
L
Li Zefan 已提交
7350
		if (unlikely(btrfs_ino(new_inode) ==
7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363
			     BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
			root_objectid = BTRFS_I(new_inode)->location.objectid;
			ret = btrfs_unlink_subvol(trans, dest, new_dir,
						root_objectid,
						new_dentry->d_name.name,
						new_dentry->d_name.len);
			BUG_ON(new_inode->i_nlink == 0);
		} else {
			ret = btrfs_unlink_inode(trans, dest, new_dir,
						 new_dentry->d_inode,
						 new_dentry->d_name.name,
						 new_dentry->d_name.len);
		}
7364
		if (!ret && new_inode->i_nlink == 0) {
7365
			ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7366
			BUG_ON(ret);
7367
		}
7368 7369 7370 7371
		if (ret) {
			btrfs_abort_transaction(trans, root, ret);
			goto out_fail;
		}
C
Chris Mason 已提交
7372
	}
7373

7374 7375
	fixup_inode_flags(new_dir, old_inode);

7376 7377
	ret = btrfs_add_link(trans, new_dir, old_inode,
			     new_dentry->d_name.name,
7378
			     new_dentry->d_name.len, 0, index);
7379 7380 7381 7382
	if (ret) {
		btrfs_abort_transaction(trans, root, ret);
		goto out_fail;
	}
C
Chris Mason 已提交
7383

L
Li Zefan 已提交
7384
	if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7385
		struct dentry *parent = new_dentry->d_parent;
7386
		btrfs_log_new_name(trans, old_inode, old_dir, parent);
7387 7388
		btrfs_end_log_trans(root);
	}
C
Chris Mason 已提交
7389
out_fail:
7390
	btrfs_end_transaction(trans, root);
7391
out_notrans:
L
Li Zefan 已提交
7392
	if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7393
		up_read(&root->fs_info->subvol_sem);
J
Josef Bacik 已提交
7394

C
Chris Mason 已提交
7395 7396 7397
	return ret;
}

C
Chris Mason 已提交
7398 7399 7400 7401
/*
 * some fairly slow code that needs optimization. This walks the list
 * of all the inodes with pending delalloc and forces them to disk.
 */
Y
Yan, Zheng 已提交
7402
int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7403 7404 7405
{
	struct list_head *head = &root->fs_info->delalloc_inodes;
	struct btrfs_inode *binode;
7406
	struct inode *inode;
7407

Y
Yan Zheng 已提交
7408 7409 7410
	if (root->fs_info->sb->s_flags & MS_RDONLY)
		return -EROFS;

7411
	spin_lock(&root->fs_info->delalloc_lock);
C
Chris Mason 已提交
7412
	while (!list_empty(head)) {
7413 7414
		binode = list_entry(head->next, struct btrfs_inode,
				    delalloc_inodes);
7415 7416 7417
		inode = igrab(&binode->vfs_inode);
		if (!inode)
			list_del_init(&binode->delalloc_inodes);
7418
		spin_unlock(&root->fs_info->delalloc_lock);
7419
		if (inode) {
7420
			filemap_flush(inode->i_mapping);
Y
Yan, Zheng 已提交
7421 7422 7423 7424
			if (delay_iput)
				btrfs_add_delayed_iput(inode);
			else
				iput(inode);
7425 7426
		}
		cond_resched();
7427
		spin_lock(&root->fs_info->delalloc_lock);
7428
	}
7429
	spin_unlock(&root->fs_info->delalloc_lock);
7430 7431 7432 7433 7434 7435

	/* the filemap_flush will queue IO into the worker threads, but
	 * we have to make sure the IO is actually started and that
	 * ordered extents get created before we return
	 */
	atomic_inc(&root->fs_info->async_submit_draining);
C
Chris Mason 已提交
7436
	while (atomic_read(&root->fs_info->nr_async_submits) ||
7437
	      atomic_read(&root->fs_info->async_delalloc_pages)) {
7438
		wait_event(root->fs_info->async_submit_wait,
7439 7440
		   (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7441 7442
	}
	atomic_dec(&root->fs_info->async_submit_draining);
7443 7444 7445
	return 0;
}

C
Chris Mason 已提交
7446 7447 7448 7449 7450 7451 7452
static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
			 const char *symname)
{
	struct btrfs_trans_handle *trans;
	struct btrfs_root *root = BTRFS_I(dir)->root;
	struct btrfs_path *path;
	struct btrfs_key key;
7453
	struct inode *inode = NULL;
C
Chris Mason 已提交
7454 7455 7456
	int err;
	int drop_inode = 0;
	u64 objectid;
7457
	u64 index = 0 ;
C
Chris Mason 已提交
7458 7459
	int name_len;
	int datasize;
7460
	unsigned long ptr;
C
Chris Mason 已提交
7461
	struct btrfs_file_extent_item *ei;
7462
	struct extent_buffer *leaf;
7463
	unsigned long nr = 0;
C
Chris Mason 已提交
7464 7465 7466 7467

	name_len = strlen(symname) + 1;
	if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
		return -ENAMETOOLONG;
7468

J
Josef Bacik 已提交
7469 7470 7471 7472 7473
	/*
	 * 2 items for inode item and ref
	 * 2 items for dir items
	 * 1 item for xattr if selinux is on
	 */
7474 7475 7476
	trans = btrfs_start_transaction(root, 5);
	if (IS_ERR(trans))
		return PTR_ERR(trans);
7477

7478 7479 7480 7481
	err = btrfs_find_free_ino(root, &objectid);
	if (err)
		goto out_unlock;

7482
	inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
L
Li Zefan 已提交
7483
				dentry->d_name.len, btrfs_ino(dir), objectid,
7484
				S_IFLNK|S_IRWXUGO, &index);
7485 7486
	if (IS_ERR(inode)) {
		err = PTR_ERR(inode);
C
Chris Mason 已提交
7487
		goto out_unlock;
7488
	}
C
Chris Mason 已提交
7489

7490
	err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
J
Josef Bacik 已提交
7491 7492 7493 7494 7495
	if (err) {
		drop_inode = 1;
		goto out_unlock;
	}

7496 7497 7498 7499 7500 7501 7502 7503 7504
	/*
	* If the active LSM wants to access the inode during
	* d_instantiate it needs these. Smack checks to see
	* if the filesystem supports xattrs by looking at the
	* ops vector.
	*/
	inode->i_fop = &btrfs_file_operations;
	inode->i_op = &btrfs_file_inode_operations;

7505
	err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
C
Chris Mason 已提交
7506 7507 7508 7509
	if (err)
		drop_inode = 1;
	else {
		inode->i_mapping->a_ops = &btrfs_aops;
C
Chris Mason 已提交
7510
		inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7511
		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
C
Chris Mason 已提交
7512 7513 7514 7515 7516
	}
	if (drop_inode)
		goto out_unlock;

	path = btrfs_alloc_path();
7517 7518 7519 7520 7521
	if (!path) {
		err = -ENOMEM;
		drop_inode = 1;
		goto out_unlock;
	}
L
Li Zefan 已提交
7522
	key.objectid = btrfs_ino(inode);
C
Chris Mason 已提交
7523 7524 7525 7526 7527
	key.offset = 0;
	btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
	datasize = btrfs_file_extent_calc_inline_size(name_len);
	err = btrfs_insert_empty_item(trans, root, path, &key,
				      datasize);
7528 7529
	if (err) {
		drop_inode = 1;
7530
		btrfs_free_path(path);
7531 7532
		goto out_unlock;
	}
7533 7534 7535 7536 7537
	leaf = path->nodes[0];
	ei = btrfs_item_ptr(leaf, path->slots[0],
			    struct btrfs_file_extent_item);
	btrfs_set_file_extent_generation(leaf, ei, trans->transid);
	btrfs_set_file_extent_type(leaf, ei,
C
Chris Mason 已提交
7538
				   BTRFS_FILE_EXTENT_INLINE);
C
Chris Mason 已提交
7539 7540 7541 7542 7543
	btrfs_set_file_extent_encryption(leaf, ei, 0);
	btrfs_set_file_extent_compression(leaf, ei, 0);
	btrfs_set_file_extent_other_encoding(leaf, ei, 0);
	btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);

C
Chris Mason 已提交
7544
	ptr = btrfs_file_extent_inline_start(ei);
7545 7546
	write_extent_buffer(leaf, symname, ptr, name_len);
	btrfs_mark_buffer_dirty(leaf);
C
Chris Mason 已提交
7547
	btrfs_free_path(path);
7548

C
Chris Mason 已提交
7549 7550
	inode->i_op = &btrfs_symlink_inode_operations;
	inode->i_mapping->a_ops = &btrfs_symlink_aops;
C
Chris Mason 已提交
7551
	inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
Y
Yan Zheng 已提交
7552
	inode_set_bytes(inode, name_len);
7553
	btrfs_i_size_write(inode, name_len - 1);
7554 7555 7556
	err = btrfs_update_inode(trans, root, inode);
	if (err)
		drop_inode = 1;
C
Chris Mason 已提交
7557 7558

out_unlock:
7559 7560
	if (!err)
		d_instantiate(dentry, inode);
7561
	nr = trans->blocks_used;
7562
	btrfs_end_transaction(trans, root);
C
Chris Mason 已提交
7563 7564 7565 7566
	if (drop_inode) {
		inode_dec_link_count(inode);
		iput(inode);
	}
7567
	btrfs_btree_balance_dirty(root, nr);
C
Chris Mason 已提交
7568 7569
	return err;
}
7570

7571 7572 7573 7574
static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
				       u64 start, u64 num_bytes, u64 min_size,
				       loff_t actual_len, u64 *alloc_hint,
				       struct btrfs_trans_handle *trans)
Y
Yan Zheng 已提交
7575
{
J
Josef Bacik 已提交
7576 7577
	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
	struct extent_map *em;
Y
Yan Zheng 已提交
7578 7579 7580
	struct btrfs_root *root = BTRFS_I(inode)->root;
	struct btrfs_key ins;
	u64 cur_offset = start;
7581
	u64 i_size;
Y
Yan Zheng 已提交
7582
	int ret = 0;
7583
	bool own_trans = true;
Y
Yan Zheng 已提交
7584

7585 7586
	if (trans)
		own_trans = false;
Y
Yan Zheng 已提交
7587
	while (num_bytes > 0) {
7588 7589 7590 7591 7592 7593
		if (own_trans) {
			trans = btrfs_start_transaction(root, 3);
			if (IS_ERR(trans)) {
				ret = PTR_ERR(trans);
				break;
			}
7594 7595
		}

7596
		ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7597
					   0, *alloc_hint, &ins, 1);
7598
		if (ret) {
7599 7600
			if (own_trans)
				btrfs_end_transaction(trans, root);
7601
			break;
Y
Yan Zheng 已提交
7602
		}
7603

Y
Yan Zheng 已提交
7604 7605 7606
		ret = insert_reserved_file_extent(trans, inode,
						  cur_offset, ins.objectid,
						  ins.offset, ins.offset,
Y
Yan, Zheng 已提交
7607
						  ins.offset, 0, 0, 0,
Y
Yan Zheng 已提交
7608
						  BTRFS_FILE_EXTENT_PREALLOC);
7609 7610 7611 7612 7613 7614
		if (ret) {
			btrfs_abort_transaction(trans, root, ret);
			if (own_trans)
				btrfs_end_transaction(trans, root);
			break;
		}
C
Chris Mason 已提交
7615 7616
		btrfs_drop_extent_cache(inode, cur_offset,
					cur_offset + ins.offset -1, 0);
7617

J
Josef Bacik 已提交
7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648
		em = alloc_extent_map();
		if (!em) {
			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
				&BTRFS_I(inode)->runtime_flags);
			goto next;
		}

		em->start = cur_offset;
		em->orig_start = cur_offset;
		em->len = ins.offset;
		em->block_start = ins.objectid;
		em->block_len = ins.offset;
		em->bdev = root->fs_info->fs_devices->latest_bdev;
		set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
		em->generation = trans->transid;

		while (1) {
			write_lock(&em_tree->lock);
			ret = add_extent_mapping(em_tree, em);
			if (!ret)
				list_move(&em->list,
					  &em_tree->modified_extents);
			write_unlock(&em_tree->lock);
			if (ret != -EEXIST)
				break;
			btrfs_drop_extent_cache(inode, cur_offset,
						cur_offset + ins.offset - 1,
						0);
		}
		free_extent_map(em);
next:
Y
Yan Zheng 已提交
7649 7650
		num_bytes -= ins.offset;
		cur_offset += ins.offset;
7651
		*alloc_hint = ins.objectid + ins.offset;
7652

7653
		inode_inc_iversion(inode);
Y
Yan Zheng 已提交
7654
		inode->i_ctime = CURRENT_TIME;
7655
		BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
Y
Yan Zheng 已提交
7656
		if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7657 7658
		    (actual_len > inode->i_size) &&
		    (cur_offset > inode->i_size)) {
7659
			if (cur_offset > actual_len)
7660
				i_size = actual_len;
7661
			else
7662 7663 7664
				i_size = cur_offset;
			i_size_write(inode, i_size);
			btrfs_ordered_update_i_size(inode, i_size, NULL);
7665 7666
		}

Y
Yan Zheng 已提交
7667
		ret = btrfs_update_inode(trans, root, inode);
7668 7669 7670 7671 7672 7673 7674

		if (ret) {
			btrfs_abort_transaction(trans, root, ret);
			if (own_trans)
				btrfs_end_transaction(trans, root);
			break;
		}
Y
Yan Zheng 已提交
7675

7676 7677
		if (own_trans)
			btrfs_end_transaction(trans, root);
7678
	}
Y
Yan Zheng 已提交
7679 7680 7681
	return ret;
}

7682 7683 7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695 7696 7697 7698 7699
int btrfs_prealloc_file_range(struct inode *inode, int mode,
			      u64 start, u64 num_bytes, u64 min_size,
			      loff_t actual_len, u64 *alloc_hint)
{
	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
					   min_size, actual_len, alloc_hint,
					   NULL);
}

int btrfs_prealloc_file_range_trans(struct inode *inode,
				    struct btrfs_trans_handle *trans, int mode,
				    u64 start, u64 num_bytes, u64 min_size,
				    loff_t actual_len, u64 *alloc_hint)
{
	return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
					   min_size, actual_len, alloc_hint, trans);
}

7700 7701 7702 7703 7704
static int btrfs_set_page_dirty(struct page *page)
{
	return __set_page_dirty_nobuffers(page);
}

7705
static int btrfs_permission(struct inode *inode, int mask)
Y
Yan 已提交
7706
{
L
Li Zefan 已提交
7707
	struct btrfs_root *root = BTRFS_I(inode)->root;
7708
	umode_t mode = inode->i_mode;
L
Li Zefan 已提交
7709

7710 7711 7712 7713 7714 7715 7716
	if (mask & MAY_WRITE &&
	    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
		if (btrfs_root_readonly(root))
			return -EROFS;
		if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
			return -EACCES;
	}
7717
	return generic_permission(inode, mask);
Y
Yan 已提交
7718
}
C
Chris Mason 已提交
7719

7720
static const struct inode_operations btrfs_dir_inode_operations = {
7721
	.getattr	= btrfs_getattr,
C
Chris Mason 已提交
7722 7723 7724 7725 7726 7727 7728 7729 7730
	.lookup		= btrfs_lookup,
	.create		= btrfs_create,
	.unlink		= btrfs_unlink,
	.link		= btrfs_link,
	.mkdir		= btrfs_mkdir,
	.rmdir		= btrfs_rmdir,
	.rename		= btrfs_rename,
	.symlink	= btrfs_symlink,
	.setattr	= btrfs_setattr,
J
Josef Bacik 已提交
7731
	.mknod		= btrfs_mknod,
7732 7733
	.setxattr	= btrfs_setxattr,
	.getxattr	= btrfs_getxattr,
J
Josef Bacik 已提交
7734
	.listxattr	= btrfs_listxattr,
7735
	.removexattr	= btrfs_removexattr,
Y
Yan 已提交
7736
	.permission	= btrfs_permission,
7737
	.get_acl	= btrfs_get_acl,
C
Chris Mason 已提交
7738
};
7739
static const struct inode_operations btrfs_dir_ro_inode_operations = {
C
Chris Mason 已提交
7740
	.lookup		= btrfs_lookup,
Y
Yan 已提交
7741
	.permission	= btrfs_permission,
7742
	.get_acl	= btrfs_get_acl,
C
Chris Mason 已提交
7743
};
7744

7745
static const struct file_operations btrfs_dir_file_operations = {
C
Chris Mason 已提交
7746 7747
	.llseek		= generic_file_llseek,
	.read		= generic_read_dir,
7748
	.readdir	= btrfs_real_readdir,
C
Christoph Hellwig 已提交
7749
	.unlocked_ioctl	= btrfs_ioctl,
C
Chris Mason 已提交
7750
#ifdef CONFIG_COMPAT
C
Christoph Hellwig 已提交
7751
	.compat_ioctl	= btrfs_ioctl,
C
Chris Mason 已提交
7752
#endif
S
Sage Weil 已提交
7753
	.release        = btrfs_release_file,
7754
	.fsync		= btrfs_sync_file,
C
Chris Mason 已提交
7755 7756
};

7757
static struct extent_io_ops btrfs_extent_io_ops = {
7758
	.fill_delalloc = run_delalloc_range,
7759
	.submit_bio_hook = btrfs_submit_bio_hook,
7760
	.merge_bio_hook = btrfs_merge_bio_hook,
7761
	.readpage_end_io_hook = btrfs_readpage_end_io_hook,
7762
	.writepage_end_io_hook = btrfs_writepage_end_io_hook,
7763
	.writepage_start_hook = btrfs_writepage_start_hook,
C
Chris Mason 已提交
7764 7765
	.set_bit_hook = btrfs_set_bit_hook,
	.clear_bit_hook = btrfs_clear_bit_hook,
J
Josef Bacik 已提交
7766 7767
	.merge_extent_hook = btrfs_merge_extent_hook,
	.split_extent_hook = btrfs_split_extent_hook,
7768 7769
};

7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781
/*
 * btrfs doesn't support the bmap operation because swapfiles
 * use bmap to make a mapping of extents in the file.  They assume
 * these extents won't change over the life of the file and they
 * use the bmap result to do IO directly to the drive.
 *
 * the btrfs bmap call would return logical addresses that aren't
 * suitable for IO and they also will change frequently as COW
 * operations happen.  So, swapfile + btrfs == corruption.
 *
 * For now we're avoiding this by dropping bmap.
 */
7782
static const struct address_space_operations btrfs_aops = {
C
Chris Mason 已提交
7783 7784
	.readpage	= btrfs_readpage,
	.writepage	= btrfs_writepage,
C
Chris Mason 已提交
7785
	.writepages	= btrfs_writepages,
C
Chris Mason 已提交
7786
	.readpages	= btrfs_readpages,
7787
	.direct_IO	= btrfs_direct_IO,
7788 7789
	.invalidatepage = btrfs_invalidatepage,
	.releasepage	= btrfs_releasepage,
7790
	.set_page_dirty	= btrfs_set_page_dirty,
7791
	.error_remove_page = generic_error_remove_page,
C
Chris Mason 已提交
7792 7793
};

7794
static const struct address_space_operations btrfs_symlink_aops = {
C
Chris Mason 已提交
7795 7796
	.readpage	= btrfs_readpage,
	.writepage	= btrfs_writepage,
C
Chris Mason 已提交
7797 7798
	.invalidatepage = btrfs_invalidatepage,
	.releasepage	= btrfs_releasepage,
C
Chris Mason 已提交
7799 7800
};

7801
static const struct inode_operations btrfs_file_inode_operations = {
C
Chris Mason 已提交
7802 7803
	.getattr	= btrfs_getattr,
	.setattr	= btrfs_setattr,
7804 7805
	.setxattr	= btrfs_setxattr,
	.getxattr	= btrfs_getxattr,
J
Josef Bacik 已提交
7806
	.listxattr      = btrfs_listxattr,
7807
	.removexattr	= btrfs_removexattr,
Y
Yan 已提交
7808
	.permission	= btrfs_permission,
Y
Yehuda Sadeh 已提交
7809
	.fiemap		= btrfs_fiemap,
7810
	.get_acl	= btrfs_get_acl,
7811
	.update_time	= btrfs_update_time,
C
Chris Mason 已提交
7812
};
7813
static const struct inode_operations btrfs_special_inode_operations = {
J
Josef Bacik 已提交
7814 7815
	.getattr	= btrfs_getattr,
	.setattr	= btrfs_setattr,
Y
Yan 已提交
7816
	.permission	= btrfs_permission,
7817 7818
	.setxattr	= btrfs_setxattr,
	.getxattr	= btrfs_getxattr,
J
Josef Bacik 已提交
7819
	.listxattr	= btrfs_listxattr,
7820
	.removexattr	= btrfs_removexattr,
7821
	.get_acl	= btrfs_get_acl,
7822
	.update_time	= btrfs_update_time,
J
Josef Bacik 已提交
7823
};
7824
static const struct inode_operations btrfs_symlink_inode_operations = {
C
Chris Mason 已提交
7825 7826 7827
	.readlink	= generic_readlink,
	.follow_link	= page_follow_link_light,
	.put_link	= page_put_link,
7828
	.getattr	= btrfs_getattr,
7829
	.setattr	= btrfs_setattr,
Y
Yan 已提交
7830
	.permission	= btrfs_permission,
J
Jim Owens 已提交
7831 7832 7833 7834
	.setxattr	= btrfs_setxattr,
	.getxattr	= btrfs_getxattr,
	.listxattr	= btrfs_listxattr,
	.removexattr	= btrfs_removexattr,
7835
	.get_acl	= btrfs_get_acl,
7836
	.update_time	= btrfs_update_time,
C
Chris Mason 已提交
7837
};
7838

7839
const struct dentry_operations btrfs_dentry_operations = {
7840
	.d_delete	= btrfs_dentry_delete,
7841
	.d_release	= btrfs_dentry_release,
7842
};