Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/fs.h>
   7#include <linux/pagemap.h>
   8#include <linux/time.h>
   9#include <linux/init.h>
  10#include <linux/string.h>
  11#include <linux/backing-dev.h>
  12#include <linux/falloc.h>
  13#include <linux/writeback.h>
  14#include <linux/compat.h>
  15#include <linux/slab.h>
  16#include <linux/btrfs.h>
  17#include <linux/uio.h>
  18#include <linux/iversion.h>
 
 
  19#include "ctree.h"
  20#include "disk-io.h"
  21#include "transaction.h"
  22#include "btrfs_inode.h"
  23#include "print-tree.h"
  24#include "tree-log.h"
  25#include "locking.h"
  26#include "volumes.h"
  27#include "qgroup.h"
  28#include "compression.h"
  29#include "delalloc-space.h"
  30
  31static struct kmem_cache *btrfs_inode_defrag_cachep;
  32/*
  33 * when auto defrag is enabled we
  34 * queue up these defrag structs to remember which
  35 * inodes need defragging passes
  36 */
  37struct inode_defrag {
  38	struct rb_node rb_node;
  39	/* objectid */
  40	u64 ino;
  41	/*
  42	 * transid where the defrag was added, we search for
  43	 * extents newer than this
  44	 */
  45	u64 transid;
  46
  47	/* root objectid */
  48	u64 root;
  49
  50	/* last offset we were able to defrag */
  51	u64 last_offset;
  52
  53	/* if we've wrapped around back to zero once already */
  54	int cycled;
  55};
  56
  57static int __compare_inode_defrag(struct inode_defrag *defrag1,
  58				  struct inode_defrag *defrag2)
  59{
  60	if (defrag1->root > defrag2->root)
  61		return 1;
  62	else if (defrag1->root < defrag2->root)
  63		return -1;
  64	else if (defrag1->ino > defrag2->ino)
  65		return 1;
  66	else if (defrag1->ino < defrag2->ino)
  67		return -1;
  68	else
  69		return 0;
  70}
  71
  72/* pop a record for an inode into the defrag tree.  The lock
  73 * must be held already
  74 *
  75 * If you're inserting a record for an older transid than an
  76 * existing record, the transid already in the tree is lowered
  77 *
  78 * If an existing record is found the defrag item you
  79 * pass in is freed
  80 */
  81static int __btrfs_add_inode_defrag(struct btrfs_inode *inode,
  82				    struct inode_defrag *defrag)
  83{
  84	struct btrfs_fs_info *fs_info = inode->root->fs_info;
  85	struct inode_defrag *entry;
  86	struct rb_node **p;
  87	struct rb_node *parent = NULL;
  88	int ret;
  89
  90	p = &fs_info->defrag_inodes.rb_node;
  91	while (*p) {
  92		parent = *p;
  93		entry = rb_entry(parent, struct inode_defrag, rb_node);
  94
  95		ret = __compare_inode_defrag(defrag, entry);
  96		if (ret < 0)
  97			p = &parent->rb_left;
  98		else if (ret > 0)
  99			p = &parent->rb_right;
 100		else {
 101			/* if we're reinserting an entry for
 102			 * an old defrag run, make sure to
 103			 * lower the transid of our existing record
 104			 */
 105			if (defrag->transid < entry->transid)
 106				entry->transid = defrag->transid;
 107			if (defrag->last_offset > entry->last_offset)
 108				entry->last_offset = defrag->last_offset;
 109			return -EEXIST;
 110		}
 111	}
 112	set_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags);
 113	rb_link_node(&defrag->rb_node, parent, p);
 114	rb_insert_color(&defrag->rb_node, &fs_info->defrag_inodes);
 115	return 0;
 116}
 117
 118static inline int __need_auto_defrag(struct btrfs_fs_info *fs_info)
 119{
 120	if (!btrfs_test_opt(fs_info, AUTO_DEFRAG))
 121		return 0;
 122
 123	if (btrfs_fs_closing(fs_info))
 124		return 0;
 125
 126	return 1;
 127}
 128
 129/*
 130 * insert a defrag record for this inode if auto defrag is
 131 * enabled
 132 */
 133int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
 134			   struct btrfs_inode *inode)
 135{
 136	struct btrfs_root *root = inode->root;
 137	struct btrfs_fs_info *fs_info = root->fs_info;
 138	struct inode_defrag *defrag;
 139	u64 transid;
 140	int ret;
 141
 142	if (!__need_auto_defrag(fs_info))
 143		return 0;
 144
 145	if (test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags))
 146		return 0;
 147
 148	if (trans)
 149		transid = trans->transid;
 150	else
 151		transid = inode->root->last_trans;
 152
 153	defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS);
 154	if (!defrag)
 155		return -ENOMEM;
 156
 157	defrag->ino = btrfs_ino(inode);
 158	defrag->transid = transid;
 159	defrag->root = root->root_key.objectid;
 160
 161	spin_lock(&fs_info->defrag_inodes_lock);
 162	if (!test_bit(BTRFS_INODE_IN_DEFRAG, &inode->runtime_flags)) {
 163		/*
 164		 * If we set IN_DEFRAG flag and evict the inode from memory,
 165		 * and then re-read this inode, this new inode doesn't have
 166		 * IN_DEFRAG flag. At the case, we may find the existed defrag.
 167		 */
 168		ret = __btrfs_add_inode_defrag(inode, defrag);
 169		if (ret)
 170			kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 171	} else {
 172		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 173	}
 174	spin_unlock(&fs_info->defrag_inodes_lock);
 175	return 0;
 176}
 177
 178/*
 179 * Requeue the defrag object. If there is a defrag object that points to
 180 * the same inode in the tree, we will merge them together (by
 181 * __btrfs_add_inode_defrag()) and free the one that we want to requeue.
 182 */
 183static void btrfs_requeue_inode_defrag(struct btrfs_inode *inode,
 184				       struct inode_defrag *defrag)
 185{
 186	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 187	int ret;
 188
 189	if (!__need_auto_defrag(fs_info))
 190		goto out;
 191
 192	/*
 193	 * Here we don't check the IN_DEFRAG flag, because we need merge
 194	 * them together.
 195	 */
 196	spin_lock(&fs_info->defrag_inodes_lock);
 197	ret = __btrfs_add_inode_defrag(inode, defrag);
 198	spin_unlock(&fs_info->defrag_inodes_lock);
 199	if (ret)
 200		goto out;
 201	return;
 202out:
 203	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 204}
 205
 206/*
 207 * pick the defragable inode that we want, if it doesn't exist, we will get
 208 * the next one.
 209 */
 210static struct inode_defrag *
 211btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino)
 212{
 213	struct inode_defrag *entry = NULL;
 214	struct inode_defrag tmp;
 215	struct rb_node *p;
 216	struct rb_node *parent = NULL;
 217	int ret;
 218
 219	tmp.ino = ino;
 220	tmp.root = root;
 221
 222	spin_lock(&fs_info->defrag_inodes_lock);
 223	p = fs_info->defrag_inodes.rb_node;
 224	while (p) {
 225		parent = p;
 226		entry = rb_entry(parent, struct inode_defrag, rb_node);
 227
 228		ret = __compare_inode_defrag(&tmp, entry);
 229		if (ret < 0)
 230			p = parent->rb_left;
 231		else if (ret > 0)
 232			p = parent->rb_right;
 233		else
 234			goto out;
 235	}
 236
 237	if (parent && __compare_inode_defrag(&tmp, entry) > 0) {
 238		parent = rb_next(parent);
 239		if (parent)
 240			entry = rb_entry(parent, struct inode_defrag, rb_node);
 241		else
 242			entry = NULL;
 243	}
 244out:
 245	if (entry)
 246		rb_erase(parent, &fs_info->defrag_inodes);
 247	spin_unlock(&fs_info->defrag_inodes_lock);
 248	return entry;
 249}
 250
 251void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info)
 252{
 253	struct inode_defrag *defrag;
 254	struct rb_node *node;
 255
 256	spin_lock(&fs_info->defrag_inodes_lock);
 257	node = rb_first(&fs_info->defrag_inodes);
 258	while (node) {
 259		rb_erase(node, &fs_info->defrag_inodes);
 260		defrag = rb_entry(node, struct inode_defrag, rb_node);
 261		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 262
 263		cond_resched_lock(&fs_info->defrag_inodes_lock);
 264
 265		node = rb_first(&fs_info->defrag_inodes);
 266	}
 267	spin_unlock(&fs_info->defrag_inodes_lock);
 268}
 269
 270#define BTRFS_DEFRAG_BATCH	1024
 271
 272static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info,
 273				    struct inode_defrag *defrag)
 274{
 275	struct btrfs_root *inode_root;
 276	struct inode *inode;
 277	struct btrfs_key key;
 278	struct btrfs_ioctl_defrag_range_args range;
 279	int num_defrag;
 280	int index;
 281	int ret;
 282
 283	/* get the inode */
 284	key.objectid = defrag->root;
 285	key.type = BTRFS_ROOT_ITEM_KEY;
 286	key.offset = (u64)-1;
 287
 288	index = srcu_read_lock(&fs_info->subvol_srcu);
 289
 290	inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
 291	if (IS_ERR(inode_root)) {
 292		ret = PTR_ERR(inode_root);
 293		goto cleanup;
 294	}
 295
 296	key.objectid = defrag->ino;
 297	key.type = BTRFS_INODE_ITEM_KEY;
 298	key.offset = 0;
 299	inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
 300	if (IS_ERR(inode)) {
 301		ret = PTR_ERR(inode);
 302		goto cleanup;
 303	}
 304	srcu_read_unlock(&fs_info->subvol_srcu, index);
 305
 306	/* do a chunk of defrag */
 307	clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags);
 308	memset(&range, 0, sizeof(range));
 309	range.len = (u64)-1;
 310	range.start = defrag->last_offset;
 311
 312	sb_start_write(fs_info->sb);
 313	num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
 314				       BTRFS_DEFRAG_BATCH);
 315	sb_end_write(fs_info->sb);
 316	/*
 317	 * if we filled the whole defrag batch, there
 318	 * must be more work to do.  Queue this defrag
 319	 * again
 320	 */
 321	if (num_defrag == BTRFS_DEFRAG_BATCH) {
 322		defrag->last_offset = range.start;
 323		btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
 324	} else if (defrag->last_offset && !defrag->cycled) {
 325		/*
 326		 * we didn't fill our defrag batch, but
 327		 * we didn't start at zero.  Make sure we loop
 328		 * around to the start of the file.
 329		 */
 330		defrag->last_offset = 0;
 331		defrag->cycled = 1;
 332		btrfs_requeue_inode_defrag(BTRFS_I(inode), defrag);
 333	} else {
 334		kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 335	}
 336
 337	iput(inode);
 338	return 0;
 339cleanup:
 340	srcu_read_unlock(&fs_info->subvol_srcu, index);
 341	kmem_cache_free(btrfs_inode_defrag_cachep, defrag);
 342	return ret;
 343}
 344
 345/*
 346 * run through the list of inodes in the FS that need
 347 * defragging
 348 */
 349int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
 350{
 351	struct inode_defrag *defrag;
 352	u64 first_ino = 0;
 353	u64 root_objectid = 0;
 354
 355	atomic_inc(&fs_info->defrag_running);
 356	while (1) {
 357		/* Pause the auto defragger. */
 358		if (test_bit(BTRFS_FS_STATE_REMOUNTING,
 359			     &fs_info->fs_state))
 360			break;
 361
 362		if (!__need_auto_defrag(fs_info))
 363			break;
 364
 365		/* find an inode to defrag */
 366		defrag = btrfs_pick_defrag_inode(fs_info, root_objectid,
 367						 first_ino);
 368		if (!defrag) {
 369			if (root_objectid || first_ino) {
 370				root_objectid = 0;
 371				first_ino = 0;
 372				continue;
 373			} else {
 374				break;
 375			}
 376		}
 377
 378		first_ino = defrag->ino + 1;
 379		root_objectid = defrag->root;
 380
 381		__btrfs_run_defrag_inode(fs_info, defrag);
 382	}
 383	atomic_dec(&fs_info->defrag_running);
 384
 385	/*
 386	 * during unmount, we use the transaction_wait queue to
 387	 * wait for the defragger to stop
 388	 */
 389	wake_up(&fs_info->transaction_wait);
 390	return 0;
 391}
 392
 393/* simple helper to fault in pages and copy.  This should go away
 394 * and be replaced with calls into generic code.
 395 */
 396static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
 397					 struct page **prepared_pages,
 398					 struct iov_iter *i)
 399{
 400	size_t copied = 0;
 401	size_t total_copied = 0;
 402	int pg = 0;
 403	int offset = offset_in_page(pos);
 404
 405	while (write_bytes > 0) {
 406		size_t count = min_t(size_t,
 407				     PAGE_SIZE - offset, write_bytes);
 408		struct page *page = prepared_pages[pg];
 409		/*
 410		 * Copy data from userspace to the current page
 411		 */
 412		copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
 413
 414		/* Flush processor's dcache for this page */
 415		flush_dcache_page(page);
 416
 417		/*
 418		 * if we get a partial write, we can end up with
 419		 * partially up to date pages.  These add
 420		 * a lot of complexity, so make sure they don't
 421		 * happen by forcing this copy to be retried.
 422		 *
 423		 * The rest of the btrfs_file_write code will fall
 424		 * back to page at a time copies after we return 0.
 425		 */
 426		if (!PageUptodate(page) && copied < count)
 427			copied = 0;
 
 
 
 
 
 
 428
 429		iov_iter_advance(i, copied);
 430		write_bytes -= copied;
 431		total_copied += copied;
 432
 433		/* Return to btrfs_file_write_iter to fault page */
 434		if (unlikely(copied == 0))
 435			break;
 436
 437		if (copied < PAGE_SIZE - offset) {
 438			offset += copied;
 439		} else {
 440			pg++;
 441			offset = 0;
 442		}
 443	}
 444	return total_copied;
 445}
 446
 447/*
 448 * unlocks pages after btrfs_file_write is done with them
 449 */
 450static void btrfs_drop_pages(struct page **pages, size_t num_pages)
 
 
 451{
 452	size_t i;
 
 
 
 
 453	for (i = 0; i < num_pages; i++) {
 454		/* page checked is some magic around finding pages that
 455		 * have been modified without going through btrfs_set_page_dirty
 456		 * clear it here. There should be no need to mark the pages
 457		 * accessed as prepare_pages should have marked them accessed
 458		 * in prepare_pages via find_or_create_page()
 459		 */
 460		ClearPageChecked(pages[i]);
 
 461		unlock_page(pages[i]);
 462		put_page(pages[i]);
 463	}
 464}
 465
 466static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
 467					 const u64 start,
 468					 const u64 len,
 469					 struct extent_state **cached_state)
 470{
 471	u64 search_start = start;
 472	const u64 end = start + len - 1;
 473
 474	while (search_start < end) {
 475		const u64 search_len = end - search_start + 1;
 476		struct extent_map *em;
 477		u64 em_len;
 478		int ret = 0;
 479
 480		em = btrfs_get_extent(inode, NULL, 0, search_start,
 481				      search_len, 0);
 482		if (IS_ERR(em))
 483			return PTR_ERR(em);
 484
 485		if (em->block_start != EXTENT_MAP_HOLE)
 486			goto next;
 487
 488		em_len = em->len;
 489		if (em->start < search_start)
 490			em_len -= search_start - em->start;
 491		if (em_len > search_len)
 492			em_len = search_len;
 493
 494		ret = set_extent_bit(&inode->io_tree, search_start,
 495				     search_start + em_len - 1,
 496				     EXTENT_DELALLOC_NEW,
 497				     NULL, cached_state, GFP_NOFS);
 498next:
 499		search_start = extent_map_end(em);
 500		free_extent_map(em);
 501		if (ret)
 502			return ret;
 503	}
 504	return 0;
 505}
 506
 507/*
 508 * after copy_from_user, pages need to be dirtied and we need to make
 509 * sure holes are created between the current EOF and the start of
 510 * any next extents (if required).
 511 *
 512 * this also makes the decision about creating an inline extent vs
 513 * doing real data extents, marking pages dirty and delalloc as required.
 514 */
 515int btrfs_dirty_pages(struct inode *inode, struct page **pages,
 516		      size_t num_pages, loff_t pos, size_t write_bytes,
 517		      struct extent_state **cached)
 518{
 519	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 520	int err = 0;
 521	int i;
 522	u64 num_bytes;
 523	u64 start_pos;
 524	u64 end_of_last_block;
 525	u64 end_pos = pos + write_bytes;
 526	loff_t isize = i_size_read(inode);
 527	unsigned int extra_bits = 0;
 528
 529	start_pos = pos & ~((u64) fs_info->sectorsize - 1);
 
 
 
 
 
 
 530	num_bytes = round_up(write_bytes + pos - start_pos,
 531			     fs_info->sectorsize);
 
 532
 533	end_of_last_block = start_pos + num_bytes - 1;
 534
 535	/*
 536	 * The pages may have already been dirty, clear out old accounting so
 537	 * we can set things up properly
 538	 */
 539	clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, end_of_last_block,
 540			 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
 541			 0, 0, cached);
 542
 543	if (!btrfs_is_free_space_inode(BTRFS_I(inode))) {
 544		if (start_pos >= isize &&
 545		    !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)) {
 546			/*
 547			 * There can't be any extents following eof in this case
 548			 * so just set the delalloc new bit for the range
 549			 * directly.
 550			 */
 551			extra_bits |= EXTENT_DELALLOC_NEW;
 552		} else {
 553			err = btrfs_find_new_delalloc_bytes(BTRFS_I(inode),
 554							    start_pos,
 555							    num_bytes, cached);
 556			if (err)
 557				return err;
 558		}
 559	}
 560
 561	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 562					extra_bits, cached);
 563	if (err)
 564		return err;
 565
 566	for (i = 0; i < num_pages; i++) {
 567		struct page *p = pages[i];
 568		SetPageUptodate(p);
 569		ClearPageChecked(p);
 570		set_page_dirty(p);
 
 
 
 
 571	}
 572
 573	/*
 574	 * we've only changed i_size in ram, and we haven't updated
 575	 * the disk i_size.  There is no need to log the inode
 576	 * at this time.
 577	 */
 578	if (end_pos > isize)
 579		i_size_write(inode, end_pos);
 580	return 0;
 581}
 582
 583/*
 584 * this drops all the extents in the cache that intersect the range
 585 * [start, end].  Existing extents are split as required.
 586 */
 587void btrfs_drop_extent_cache(struct btrfs_inode *inode, u64 start, u64 end,
 588			     int skip_pinned)
 589{
 590	struct extent_map *em;
 591	struct extent_map *split = NULL;
 592	struct extent_map *split2 = NULL;
 593	struct extent_map_tree *em_tree = &inode->extent_tree;
 594	u64 len = end - start + 1;
 595	u64 gen;
 596	int ret;
 597	int testend = 1;
 598	unsigned long flags;
 599	int compressed = 0;
 600	bool modified;
 601
 602	WARN_ON(end < start);
 603	if (end == (u64)-1) {
 604		len = (u64)-1;
 605		testend = 0;
 606	}
 607	while (1) {
 608		int no_splits = 0;
 609
 610		modified = false;
 611		if (!split)
 612			split = alloc_extent_map();
 613		if (!split2)
 614			split2 = alloc_extent_map();
 615		if (!split || !split2)
 616			no_splits = 1;
 617
 618		write_lock(&em_tree->lock);
 619		em = lookup_extent_mapping(em_tree, start, len);
 620		if (!em) {
 621			write_unlock(&em_tree->lock);
 622			break;
 623		}
 624		flags = em->flags;
 625		gen = em->generation;
 626		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 627			if (testend && em->start + em->len >= start + len) {
 628				free_extent_map(em);
 629				write_unlock(&em_tree->lock);
 630				break;
 631			}
 632			start = em->start + em->len;
 633			if (testend)
 634				len = start + len - (em->start + em->len);
 635			free_extent_map(em);
 636			write_unlock(&em_tree->lock);
 637			continue;
 638		}
 639		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 640		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 641		clear_bit(EXTENT_FLAG_LOGGING, &flags);
 642		modified = !list_empty(&em->list);
 643		if (no_splits)
 644			goto next;
 645
 646		if (em->start < start) {
 647			split->start = em->start;
 648			split->len = start - em->start;
 649
 650			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 651				split->orig_start = em->orig_start;
 652				split->block_start = em->block_start;
 653
 654				if (compressed)
 655					split->block_len = em->block_len;
 656				else
 657					split->block_len = split->len;
 658				split->orig_block_len = max(split->block_len,
 659						em->orig_block_len);
 660				split->ram_bytes = em->ram_bytes;
 661			} else {
 662				split->orig_start = split->start;
 663				split->block_len = 0;
 664				split->block_start = em->block_start;
 665				split->orig_block_len = 0;
 666				split->ram_bytes = split->len;
 667			}
 668
 669			split->generation = gen;
 670			split->bdev = em->bdev;
 671			split->flags = flags;
 672			split->compress_type = em->compress_type;
 673			replace_extent_mapping(em_tree, em, split, modified);
 674			free_extent_map(split);
 675			split = split2;
 676			split2 = NULL;
 677		}
 678		if (testend && em->start + em->len > start + len) {
 679			u64 diff = start + len - em->start;
 680
 681			split->start = start + len;
 682			split->len = em->start + em->len - (start + len);
 683			split->bdev = em->bdev;
 684			split->flags = flags;
 685			split->compress_type = em->compress_type;
 686			split->generation = gen;
 687
 688			if (em->block_start < EXTENT_MAP_LAST_BYTE) {
 689				split->orig_block_len = max(em->block_len,
 690						    em->orig_block_len);
 691
 692				split->ram_bytes = em->ram_bytes;
 693				if (compressed) {
 694					split->block_len = em->block_len;
 695					split->block_start = em->block_start;
 696					split->orig_start = em->orig_start;
 697				} else {
 698					split->block_len = split->len;
 699					split->block_start = em->block_start
 700						+ diff;
 701					split->orig_start = em->orig_start;
 702				}
 703			} else {
 704				split->ram_bytes = split->len;
 705				split->orig_start = split->start;
 706				split->block_len = 0;
 707				split->block_start = em->block_start;
 708				split->orig_block_len = 0;
 709			}
 710
 711			if (extent_map_in_tree(em)) {
 712				replace_extent_mapping(em_tree, em, split,
 713						       modified);
 714			} else {
 715				ret = add_extent_mapping(em_tree, split,
 716							 modified);
 717				ASSERT(ret == 0); /* Logic error */
 718			}
 719			free_extent_map(split);
 720			split = NULL;
 721		}
 722next:
 723		if (extent_map_in_tree(em))
 724			remove_extent_mapping(em_tree, em);
 725		write_unlock(&em_tree->lock);
 726
 727		/* once for us */
 728		free_extent_map(em);
 729		/* once for the tree*/
 730		free_extent_map(em);
 731	}
 732	if (split)
 733		free_extent_map(split);
 734	if (split2)
 735		free_extent_map(split2);
 736}
 737
 738/*
 739 * this is very complex, but the basic idea is to drop all extents
 740 * in the range start - end.  hint_block is filled in with a block number
 741 * that would be a good hint to the block allocator for this file.
 742 *
 743 * If an extent intersects the range but is not entirely inside the range
 744 * it is either truncated or split.  Anything entirely inside the range
 745 * is deleted from the tree.
 
 
 
 
 
 
 746 */
 747int __btrfs_drop_extents(struct btrfs_trans_handle *trans,
 748			 struct btrfs_root *root, struct inode *inode,
 749			 struct btrfs_path *path, u64 start, u64 end,
 750			 u64 *drop_end, int drop_cache,
 751			 int replace_extent,
 752			 u32 extent_item_size,
 753			 int *key_inserted)
 754{
 755	struct btrfs_fs_info *fs_info = root->fs_info;
 756	struct extent_buffer *leaf;
 757	struct btrfs_file_extent_item *fi;
 758	struct btrfs_ref ref = { 0 };
 759	struct btrfs_key key;
 760	struct btrfs_key new_key;
 761	u64 ino = btrfs_ino(BTRFS_I(inode));
 762	u64 search_start = start;
 763	u64 disk_bytenr = 0;
 764	u64 num_bytes = 0;
 765	u64 extent_offset = 0;
 766	u64 extent_end = 0;
 767	u64 last_end = start;
 768	int del_nr = 0;
 769	int del_slot = 0;
 770	int extent_type;
 771	int recow;
 772	int ret;
 773	int modify_tree = -1;
 774	int update_refs;
 775	int found = 0;
 776	int leafs_visited = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 777
 778	if (drop_cache)
 779		btrfs_drop_extent_cache(BTRFS_I(inode), start, end - 1, 0);
 780
 781	if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent)
 782		modify_tree = 0;
 783
 784	update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
 785		       root == fs_info->tree_root);
 786	while (1) {
 787		recow = 0;
 788		ret = btrfs_lookup_file_extent(trans, root, path, ino,
 789					       search_start, modify_tree);
 790		if (ret < 0)
 791			break;
 792		if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 793			leaf = path->nodes[0];
 794			btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 795			if (key.objectid == ino &&
 796			    key.type == BTRFS_EXTENT_DATA_KEY)
 797				path->slots[0]--;
 798		}
 799		ret = 0;
 800		leafs_visited++;
 801next_slot:
 802		leaf = path->nodes[0];
 803		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 804			BUG_ON(del_nr > 0);
 805			ret = btrfs_next_leaf(root, path);
 806			if (ret < 0)
 807				break;
 808			if (ret > 0) {
 809				ret = 0;
 810				break;
 811			}
 812			leafs_visited++;
 813			leaf = path->nodes[0];
 814			recow = 1;
 815		}
 816
 817		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 818
 819		if (key.objectid > ino)
 820			break;
 821		if (WARN_ON_ONCE(key.objectid < ino) ||
 822		    key.type < BTRFS_EXTENT_DATA_KEY) {
 823			ASSERT(del_nr == 0);
 824			path->slots[0]++;
 825			goto next_slot;
 826		}
 827		if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 828			break;
 829
 830		fi = btrfs_item_ptr(leaf, path->slots[0],
 831				    struct btrfs_file_extent_item);
 832		extent_type = btrfs_file_extent_type(leaf, fi);
 833
 834		if (extent_type == BTRFS_FILE_EXTENT_REG ||
 835		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 836			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 837			num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 838			extent_offset = btrfs_file_extent_offset(leaf, fi);
 839			extent_end = key.offset +
 840				btrfs_file_extent_num_bytes(leaf, fi);
 841		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 842			extent_end = key.offset +
 843				btrfs_file_extent_ram_bytes(leaf, fi);
 844		} else {
 845			/* can't happen */
 846			BUG();
 847		}
 848
 849		/*
 850		 * Don't skip extent items representing 0 byte lengths. They
 851		 * used to be created (bug) if while punching holes we hit
 852		 * -ENOSPC condition. So if we find one here, just ensure we
 853		 * delete it, otherwise we would insert a new file extent item
 854		 * with the same key (offset) as that 0 bytes length file
 855		 * extent item in the call to setup_items_for_insert() later
 856		 * in this function.
 857		 */
 858		if (extent_end == key.offset && extent_end >= search_start) {
 859			last_end = extent_end;
 860			goto delete_extent_item;
 861		}
 862
 863		if (extent_end <= search_start) {
 864			path->slots[0]++;
 865			goto next_slot;
 866		}
 867
 868		found = 1;
 869		search_start = max(key.offset, start);
 870		if (recow || !modify_tree) {
 871			modify_tree = -1;
 872			btrfs_release_path(path);
 873			continue;
 874		}
 875
 876		/*
 877		 *     | - range to drop - |
 878		 *  | -------- extent -------- |
 879		 */
 880		if (start > key.offset && end < extent_end) {
 881			BUG_ON(del_nr > 0);
 882			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 883				ret = -EOPNOTSUPP;
 884				break;
 885			}
 886
 887			memcpy(&new_key, &key, sizeof(new_key));
 888			new_key.offset = start;
 889			ret = btrfs_duplicate_item(trans, root, path,
 890						   &new_key);
 891			if (ret == -EAGAIN) {
 892				btrfs_release_path(path);
 893				continue;
 894			}
 895			if (ret < 0)
 896				break;
 897
 898			leaf = path->nodes[0];
 899			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 900					    struct btrfs_file_extent_item);
 901			btrfs_set_file_extent_num_bytes(leaf, fi,
 902							start - key.offset);
 903
 904			fi = btrfs_item_ptr(leaf, path->slots[0],
 905					    struct btrfs_file_extent_item);
 906
 907			extent_offset += start - key.offset;
 908			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 909			btrfs_set_file_extent_num_bytes(leaf, fi,
 910							extent_end - start);
 911			btrfs_mark_buffer_dirty(leaf);
 912
 913			if (update_refs && disk_bytenr > 0) {
 914				btrfs_init_generic_ref(&ref,
 915						BTRFS_ADD_DELAYED_REF,
 916						disk_bytenr, num_bytes, 0);
 
 917				btrfs_init_data_ref(&ref,
 918						root->root_key.objectid,
 919						new_key.objectid,
 920						start - extent_offset);
 
 921				ret = btrfs_inc_extent_ref(trans, &ref);
 922				BUG_ON(ret); /* -ENOMEM */
 
 
 
 923			}
 924			key.offset = start;
 925		}
 926		/*
 927		 * From here on out we will have actually dropped something, so
 928		 * last_end can be updated.
 929		 */
 930		last_end = extent_end;
 931
 932		/*
 933		 *  | ---- range to drop ----- |
 934		 *      | -------- extent -------- |
 935		 */
 936		if (start <= key.offset && end < extent_end) {
 937			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 938				ret = -EOPNOTSUPP;
 939				break;
 940			}
 941
 942			memcpy(&new_key, &key, sizeof(new_key));
 943			new_key.offset = end;
 944			btrfs_set_item_key_safe(fs_info, path, &new_key);
 945
 946			extent_offset += end - key.offset;
 947			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 948			btrfs_set_file_extent_num_bytes(leaf, fi,
 949							extent_end - end);
 950			btrfs_mark_buffer_dirty(leaf);
 951			if (update_refs && disk_bytenr > 0)
 952				inode_sub_bytes(inode, end - key.offset);
 953			break;
 954		}
 955
 956		search_start = extent_end;
 957		/*
 958		 *       | ---- range to drop ----- |
 959		 *  | -------- extent -------- |
 960		 */
 961		if (start > key.offset && end >= extent_end) {
 962			BUG_ON(del_nr > 0);
 963			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 964				ret = -EOPNOTSUPP;
 965				break;
 966			}
 967
 968			btrfs_set_file_extent_num_bytes(leaf, fi,
 969							start - key.offset);
 970			btrfs_mark_buffer_dirty(leaf);
 971			if (update_refs && disk_bytenr > 0)
 972				inode_sub_bytes(inode, extent_end - start);
 973			if (end == extent_end)
 974				break;
 975
 976			path->slots[0]++;
 977			goto next_slot;
 978		}
 979
 980		/*
 981		 *  | ---- range to drop ----- |
 982		 *    | ------ extent ------ |
 983		 */
 984		if (start <= key.offset && end >= extent_end) {
 985delete_extent_item:
 986			if (del_nr == 0) {
 987				del_slot = path->slots[0];
 988				del_nr = 1;
 989			} else {
 990				BUG_ON(del_slot + del_nr != path->slots[0]);
 991				del_nr++;
 992			}
 993
 994			if (update_refs &&
 995			    extent_type == BTRFS_FILE_EXTENT_INLINE) {
 996				inode_sub_bytes(inode,
 997						extent_end - key.offset);
 998				extent_end = ALIGN(extent_end,
 999						   fs_info->sectorsize);
1000			} else if (update_refs && disk_bytenr > 0) {
1001				btrfs_init_generic_ref(&ref,
1002						BTRFS_DROP_DELAYED_REF,
1003						disk_bytenr, num_bytes, 0);
 
1004				btrfs_init_data_ref(&ref,
1005						root->root_key.objectid,
1006						key.objectid,
1007						key.offset - extent_offset);
 
1008				ret = btrfs_free_extent(trans, &ref);
1009				BUG_ON(ret); /* -ENOMEM */
1010				inode_sub_bytes(inode,
1011						extent_end - key.offset);
 
 
1012			}
1013
1014			if (end == extent_end)
1015				break;
1016
1017			if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
1018				path->slots[0]++;
1019				goto next_slot;
1020			}
1021
1022			ret = btrfs_del_items(trans, root, path, del_slot,
1023					      del_nr);
1024			if (ret) {
1025				btrfs_abort_transaction(trans, ret);
1026				break;
1027			}
1028
1029			del_nr = 0;
1030			del_slot = 0;
1031
1032			btrfs_release_path(path);
1033			continue;
1034		}
1035
1036		BUG();
1037	}
1038
1039	if (!ret && del_nr > 0) {
1040		/*
1041		 * Set path->slots[0] to first slot, so that after the delete
1042		 * if items are move off from our leaf to its immediate left or
1043		 * right neighbor leafs, we end up with a correct and adjusted
1044		 * path->slots[0] for our insertion (if replace_extent != 0).
1045		 */
1046		path->slots[0] = del_slot;
1047		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1048		if (ret)
1049			btrfs_abort_transaction(trans, ret);
1050	}
1051
1052	leaf = path->nodes[0];
1053	/*
1054	 * If btrfs_del_items() was called, it might have deleted a leaf, in
1055	 * which case it unlocked our path, so check path->locks[0] matches a
1056	 * write lock.
1057	 */
1058	if (!ret && replace_extent && leafs_visited == 1 &&
1059	    (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING ||
1060	     path->locks[0] == BTRFS_WRITE_LOCK) &&
1061	    btrfs_leaf_free_space(leaf) >=
1062	    sizeof(struct btrfs_item) + extent_item_size) {
1063
1064		key.objectid = ino;
1065		key.type = BTRFS_EXTENT_DATA_KEY;
1066		key.offset = start;
1067		if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
1068			struct btrfs_key slot_key;
1069
1070			btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
1071			if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
1072				path->slots[0]++;
1073		}
1074		setup_items_for_insert(root, path, &key,
1075				       &extent_item_size,
1076				       extent_item_size,
1077				       sizeof(struct btrfs_item) +
1078				       extent_item_size, 1);
1079		*key_inserted = 1;
1080	}
1081
1082	if (!replace_extent || !(*key_inserted))
 
 
1083		btrfs_release_path(path);
1084	if (drop_end)
1085		*drop_end = found ? min(end, last_end) : end;
1086	return ret;
1087}
1088
1089int btrfs_drop_extents(struct btrfs_trans_handle *trans,
1090		       struct btrfs_root *root, struct inode *inode, u64 start,
1091		       u64 end, int drop_cache)
1092{
1093	struct btrfs_path *path;
1094	int ret;
1095
1096	path = btrfs_alloc_path();
1097	if (!path)
1098		return -ENOMEM;
1099	ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL,
1100				   drop_cache, 0, 0, NULL);
1101	btrfs_free_path(path);
1102	return ret;
1103}
1104
1105static int extent_mergeable(struct extent_buffer *leaf, int slot,
1106			    u64 objectid, u64 bytenr, u64 orig_offset,
1107			    u64 *start, u64 *end)
1108{
1109	struct btrfs_file_extent_item *fi;
1110	struct btrfs_key key;
1111	u64 extent_end;
1112
1113	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
1114		return 0;
1115
1116	btrfs_item_key_to_cpu(leaf, &key, slot);
1117	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
1118		return 0;
1119
1120	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
1121	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
1122	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
1123	    btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
1124	    btrfs_file_extent_compression(leaf, fi) ||
1125	    btrfs_file_extent_encryption(leaf, fi) ||
1126	    btrfs_file_extent_other_encoding(leaf, fi))
1127		return 0;
1128
1129	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1130	if ((*start && *start != key.offset) || (*end && *end != extent_end))
1131		return 0;
1132
1133	*start = key.offset;
1134	*end = extent_end;
1135	return 1;
1136}
1137
1138/*
1139 * Mark extent in the range start - end as written.
1140 *
1141 * This changes extent type from 'pre-allocated' to 'regular'. If only
1142 * part of extent is marked as written, the extent will be split into
1143 * two or three.
1144 */
1145int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
1146			      struct btrfs_inode *inode, u64 start, u64 end)
1147{
1148	struct btrfs_fs_info *fs_info = trans->fs_info;
1149	struct btrfs_root *root = inode->root;
1150	struct extent_buffer *leaf;
1151	struct btrfs_path *path;
1152	struct btrfs_file_extent_item *fi;
1153	struct btrfs_ref ref = { 0 };
1154	struct btrfs_key key;
1155	struct btrfs_key new_key;
1156	u64 bytenr;
1157	u64 num_bytes;
1158	u64 extent_end;
1159	u64 orig_offset;
1160	u64 other_start;
1161	u64 other_end;
1162	u64 split;
1163	int del_nr = 0;
1164	int del_slot = 0;
1165	int recow;
1166	int ret;
1167	u64 ino = btrfs_ino(inode);
1168
1169	path = btrfs_alloc_path();
1170	if (!path)
1171		return -ENOMEM;
1172again:
1173	recow = 0;
1174	split = start;
1175	key.objectid = ino;
1176	key.type = BTRFS_EXTENT_DATA_KEY;
1177	key.offset = split;
1178
1179	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1180	if (ret < 0)
1181		goto out;
1182	if (ret > 0 && path->slots[0] > 0)
1183		path->slots[0]--;
1184
1185	leaf = path->nodes[0];
1186	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1187	if (key.objectid != ino ||
1188	    key.type != BTRFS_EXTENT_DATA_KEY) {
1189		ret = -EINVAL;
1190		btrfs_abort_transaction(trans, ret);
1191		goto out;
1192	}
1193	fi = btrfs_item_ptr(leaf, path->slots[0],
1194			    struct btrfs_file_extent_item);
1195	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
1196		ret = -EINVAL;
1197		btrfs_abort_transaction(trans, ret);
1198		goto out;
1199	}
1200	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
1201	if (key.offset > start || extent_end < end) {
1202		ret = -EINVAL;
1203		btrfs_abort_transaction(trans, ret);
1204		goto out;
1205	}
1206
1207	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1208	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1209	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
1210	memcpy(&new_key, &key, sizeof(new_key));
1211
1212	if (start == key.offset && end < extent_end) {
1213		other_start = 0;
1214		other_end = start;
1215		if (extent_mergeable(leaf, path->slots[0] - 1,
1216				     ino, bytenr, orig_offset,
1217				     &other_start, &other_end)) {
1218			new_key.offset = end;
1219			btrfs_set_item_key_safe(fs_info, path, &new_key);
1220			fi = btrfs_item_ptr(leaf, path->slots[0],
1221					    struct btrfs_file_extent_item);
1222			btrfs_set_file_extent_generation(leaf, fi,
1223							 trans->transid);
1224			btrfs_set_file_extent_num_bytes(leaf, fi,
1225							extent_end - end);
1226			btrfs_set_file_extent_offset(leaf, fi,
1227						     end - orig_offset);
1228			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1229					    struct btrfs_file_extent_item);
1230			btrfs_set_file_extent_generation(leaf, fi,
1231							 trans->transid);
1232			btrfs_set_file_extent_num_bytes(leaf, fi,
1233							end - other_start);
1234			btrfs_mark_buffer_dirty(leaf);
1235			goto out;
1236		}
1237	}
1238
1239	if (start > key.offset && end == extent_end) {
1240		other_start = end;
1241		other_end = 0;
1242		if (extent_mergeable(leaf, path->slots[0] + 1,
1243				     ino, bytenr, orig_offset,
1244				     &other_start, &other_end)) {
1245			fi = btrfs_item_ptr(leaf, path->slots[0],
1246					    struct btrfs_file_extent_item);
1247			btrfs_set_file_extent_num_bytes(leaf, fi,
1248							start - key.offset);
1249			btrfs_set_file_extent_generation(leaf, fi,
1250							 trans->transid);
1251			path->slots[0]++;
1252			new_key.offset = start;
1253			btrfs_set_item_key_safe(fs_info, path, &new_key);
1254
1255			fi = btrfs_item_ptr(leaf, path->slots[0],
1256					    struct btrfs_file_extent_item);
1257			btrfs_set_file_extent_generation(leaf, fi,
1258							 trans->transid);
1259			btrfs_set_file_extent_num_bytes(leaf, fi,
1260							other_end - start);
1261			btrfs_set_file_extent_offset(leaf, fi,
1262						     start - orig_offset);
1263			btrfs_mark_buffer_dirty(leaf);
1264			goto out;
1265		}
1266	}
1267
1268	while (start > key.offset || end < extent_end) {
1269		if (key.offset == start)
1270			split = end;
1271
1272		new_key.offset = split;
1273		ret = btrfs_duplicate_item(trans, root, path, &new_key);
1274		if (ret == -EAGAIN) {
1275			btrfs_release_path(path);
1276			goto again;
1277		}
1278		if (ret < 0) {
1279			btrfs_abort_transaction(trans, ret);
1280			goto out;
1281		}
1282
1283		leaf = path->nodes[0];
1284		fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
1285				    struct btrfs_file_extent_item);
1286		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1287		btrfs_set_file_extent_num_bytes(leaf, fi,
1288						split - key.offset);
1289
1290		fi = btrfs_item_ptr(leaf, path->slots[0],
1291				    struct btrfs_file_extent_item);
1292
1293		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1294		btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
1295		btrfs_set_file_extent_num_bytes(leaf, fi,
1296						extent_end - split);
1297		btrfs_mark_buffer_dirty(leaf);
1298
1299		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
1300				       num_bytes, 0);
1301		btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
1302				    orig_offset);
1303		ret = btrfs_inc_extent_ref(trans, &ref);
1304		if (ret) {
1305			btrfs_abort_transaction(trans, ret);
1306			goto out;
1307		}
1308
1309		if (split == start) {
1310			key.offset = start;
1311		} else {
1312			if (start != key.offset) {
1313				ret = -EINVAL;
1314				btrfs_abort_transaction(trans, ret);
1315				goto out;
1316			}
1317			path->slots[0]--;
1318			extent_end = end;
1319		}
1320		recow = 1;
1321	}
1322
1323	other_start = end;
1324	other_end = 0;
1325	btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1326			       num_bytes, 0);
1327	btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset);
 
1328	if (extent_mergeable(leaf, path->slots[0] + 1,
1329			     ino, bytenr, orig_offset,
1330			     &other_start, &other_end)) {
1331		if (recow) {
1332			btrfs_release_path(path);
1333			goto again;
1334		}
1335		extent_end = other_end;
1336		del_slot = path->slots[0] + 1;
1337		del_nr++;
1338		ret = btrfs_free_extent(trans, &ref);
1339		if (ret) {
1340			btrfs_abort_transaction(trans, ret);
1341			goto out;
1342		}
1343	}
1344	other_start = 0;
1345	other_end = start;
1346	if (extent_mergeable(leaf, path->slots[0] - 1,
1347			     ino, bytenr, orig_offset,
1348			     &other_start, &other_end)) {
1349		if (recow) {
1350			btrfs_release_path(path);
1351			goto again;
1352		}
1353		key.offset = other_start;
1354		del_slot = path->slots[0];
1355		del_nr++;
1356		ret = btrfs_free_extent(trans, &ref);
1357		if (ret) {
1358			btrfs_abort_transaction(trans, ret);
1359			goto out;
1360		}
1361	}
1362	if (del_nr == 0) {
1363		fi = btrfs_item_ptr(leaf, path->slots[0],
1364			   struct btrfs_file_extent_item);
1365		btrfs_set_file_extent_type(leaf, fi,
1366					   BTRFS_FILE_EXTENT_REG);
1367		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1368		btrfs_mark_buffer_dirty(leaf);
1369	} else {
1370		fi = btrfs_item_ptr(leaf, del_slot - 1,
1371			   struct btrfs_file_extent_item);
1372		btrfs_set_file_extent_type(leaf, fi,
1373					   BTRFS_FILE_EXTENT_REG);
1374		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1375		btrfs_set_file_extent_num_bytes(leaf, fi,
1376						extent_end - key.offset);
1377		btrfs_mark_buffer_dirty(leaf);
1378
1379		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1380		if (ret < 0) {
1381			btrfs_abort_transaction(trans, ret);
1382			goto out;
1383		}
1384	}
1385out:
1386	btrfs_free_path(path);
1387	return 0;
1388}
1389
1390/*
1391 * on error we return an unlocked page and the error value
1392 * on success we return a locked page and 0
1393 */
1394static int prepare_uptodate_page(struct inode *inode,
1395				 struct page *page, u64 pos,
1396				 bool force_uptodate)
1397{
 
1398	int ret = 0;
1399
1400	if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
1401	    !PageUptodate(page)) {
1402		ret = btrfs_readpage(NULL, page);
1403		if (ret)
1404			return ret;
1405		lock_page(page);
1406		if (!PageUptodate(page)) {
1407			unlock_page(page);
1408			return -EIO;
1409		}
1410		if (page->mapping != inode->i_mapping) {
 
 
 
 
 
 
 
 
 
 
 
1411			unlock_page(page);
1412			return -EAGAIN;
1413		}
1414	}
1415	return 0;
1416}
1417
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1418/*
1419 * this just gets pages into the page cache and locks them down.
1420 */
1421static noinline int prepare_pages(struct inode *inode, struct page **pages,
1422				  size_t num_pages, loff_t pos,
1423				  size_t write_bytes, bool force_uptodate)
 
1424{
1425	int i;
1426	unsigned long index = pos >> PAGE_SHIFT;
1427	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
 
1428	int err = 0;
1429	int faili;
1430
1431	for (i = 0; i < num_pages; i++) {
1432again:
1433		pages[i] = find_or_create_page(inode->i_mapping, index + i,
1434					       mask | __GFP_WRITE);
1435		if (!pages[i]) {
1436			faili = i - 1;
1437			err = -ENOMEM;
 
 
 
 
 
 
 
 
 
1438			goto fail;
1439		}
1440
1441		if (i == 0)
1442			err = prepare_uptodate_page(inode, pages[i], pos,
1443						    force_uptodate);
1444		if (!err && i == num_pages - 1)
1445			err = prepare_uptodate_page(inode, pages[i],
1446						    pos + write_bytes, false);
1447		if (err) {
1448			put_page(pages[i]);
1449			if (err == -EAGAIN) {
1450				err = 0;
1451				goto again;
1452			}
1453			faili = i - 1;
1454			goto fail;
1455		}
1456		wait_on_page_writeback(pages[i]);
1457	}
1458
1459	return 0;
1460fail:
1461	while (faili >= 0) {
1462		unlock_page(pages[faili]);
1463		put_page(pages[faili]);
1464		faili--;
1465	}
1466	return err;
1467
1468}
1469
1470/*
1471 * This function locks the extent and properly waits for data=ordered extents
1472 * to finish before allowing the pages to be modified if need.
1473 *
1474 * The return value:
1475 * 1 - the extent is locked
1476 * 0 - the extent is not locked, and everything is OK
1477 * -EAGAIN - need re-prepare the pages
1478 * the other < 0 number - Something wrong happens
1479 */
1480static noinline int
1481lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
1482				size_t num_pages, loff_t pos,
1483				size_t write_bytes,
1484				u64 *lockstart, u64 *lockend,
1485				struct extent_state **cached_state)
1486{
1487	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1488	u64 start_pos;
1489	u64 last_pos;
1490	int i;
1491	int ret = 0;
1492
1493	start_pos = round_down(pos, fs_info->sectorsize);
1494	last_pos = start_pos
1495		+ round_up(pos + write_bytes - start_pos,
1496			   fs_info->sectorsize) - 1;
1497
1498	if (start_pos < inode->vfs_inode.i_size) {
1499		struct btrfs_ordered_extent *ordered;
1500
1501		lock_extent_bits(&inode->io_tree, start_pos, last_pos,
1502				cached_state);
 
 
 
 
 
 
 
 
 
 
 
 
 
1503		ordered = btrfs_lookup_ordered_range(inode, start_pos,
1504						     last_pos - start_pos + 1);
1505		if (ordered &&
1506		    ordered->file_offset + ordered->len > start_pos &&
1507		    ordered->file_offset <= last_pos) {
1508			unlock_extent_cached(&inode->io_tree, start_pos,
1509					last_pos, cached_state);
1510			for (i = 0; i < num_pages; i++) {
1511				unlock_page(pages[i]);
1512				put_page(pages[i]);
1513			}
1514			btrfs_start_ordered_extent(&inode->vfs_inode,
1515					ordered, 1);
1516			btrfs_put_ordered_extent(ordered);
1517			return -EAGAIN;
1518		}
1519		if (ordered)
1520			btrfs_put_ordered_extent(ordered);
1521
1522		*lockstart = start_pos;
1523		*lockend = last_pos;
1524		ret = 1;
1525	}
1526
1527	/*
1528	 * It's possible the pages are dirty right now, but we don't want
1529	 * to clean them yet because copy_from_user may catch a page fault
1530	 * and we might have to fall back to one page at a time.  If that
1531	 * happens, we'll unlock these pages and we'd have a window where
1532	 * reclaim could sneak in and drop the once-dirty page on the floor
1533	 * without writing it.
1534	 *
1535	 * We have the pages locked and the extent range locked, so there's
1536	 * no way someone can start IO on any dirty pages in this range.
1537	 *
1538	 * We'll call btrfs_dirty_pages() later on, and that will flip around
1539	 * delalloc bits and dirty the pages as required.
1540	 */
1541	for (i = 0; i < num_pages; i++) {
1542		set_page_extent_mapped(pages[i]);
1543		WARN_ON(!PageLocked(pages[i]));
1544	}
1545
1546	return ret;
1547}
1548
1549static noinline int check_can_nocow(struct btrfs_inode *inode, loff_t pos,
1550				    size_t *write_bytes)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1551{
1552	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1553	struct btrfs_root *root = inode->root;
 
1554	u64 lockstart, lockend;
1555	u64 num_bytes;
1556	int ret;
1557
1558	ret = btrfs_start_write_no_snapshotting(root);
1559	if (!ret)
 
 
1560		return -EAGAIN;
1561
1562	lockstart = round_down(pos, fs_info->sectorsize);
1563	lockend = round_up(pos + *write_bytes,
1564			   fs_info->sectorsize) - 1;
1565
1566	btrfs_lock_and_flush_ordered_range(&inode->io_tree, inode, lockstart,
1567					   lockend, NULL);
1568
1569	num_bytes = lockend - lockstart + 1;
1570	ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1571			NULL, NULL, NULL);
1572	if (ret <= 0) {
1573		ret = 0;
1574		btrfs_end_write_no_snapshotting(root);
 
 
1575	} else {
 
 
 
 
 
 
 
 
1576		*write_bytes = min_t(size_t, *write_bytes ,
1577				     num_bytes - pos + lockstart);
1578	}
1579
1580	unlock_extent(&inode->io_tree, lockstart, lockend);
1581
1582	return ret;
1583}
1584
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1585static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1586					       struct iov_iter *i)
1587{
1588	struct file *file = iocb->ki_filp;
1589	loff_t pos = iocb->ki_pos;
1590	struct inode *inode = file_inode(file);
1591	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1592	struct btrfs_root *root = BTRFS_I(inode)->root;
1593	struct page **pages = NULL;
1594	struct extent_changeset *data_reserved = NULL;
1595	u64 release_bytes = 0;
1596	u64 lockstart;
1597	u64 lockend;
1598	size_t num_written = 0;
1599	int nrptrs;
1600	int ret = 0;
1601	bool only_release_metadata = false;
1602	bool force_page_uptodate = false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1603
 
1604	nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1605			PAGE_SIZE / (sizeof(struct page *)));
1606	nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1607	nrptrs = max(nrptrs, 8);
1608	pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1609	if (!pages)
1610		return -ENOMEM;
 
 
1611
1612	while (iov_iter_count(i) > 0) {
1613		struct extent_state *cached_state = NULL;
1614		size_t offset = offset_in_page(pos);
1615		size_t sector_offset;
1616		size_t write_bytes = min(iov_iter_count(i),
1617					 nrptrs * (size_t)PAGE_SIZE -
1618					 offset);
1619		size_t num_pages = DIV_ROUND_UP(write_bytes + offset,
1620						PAGE_SIZE);
1621		size_t reserve_bytes;
1622		size_t dirty_pages;
1623		size_t copied;
1624		size_t dirty_sectors;
1625		size_t num_sectors;
1626		int extents_locked;
1627
1628		WARN_ON(num_pages > nrptrs);
1629
1630		/*
1631		 * Fault pages before locking them in prepare_pages
1632		 * to avoid recursive lock
1633		 */
1634		if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1635			ret = -EFAULT;
1636			break;
1637		}
1638
 
1639		sector_offset = pos & (fs_info->sectorsize - 1);
1640		reserve_bytes = round_up(write_bytes + sector_offset,
1641				fs_info->sectorsize);
1642
1643		extent_changeset_release(data_reserved);
1644		ret = btrfs_check_data_free_space(inode, &data_reserved, pos,
1645						  write_bytes);
 
1646		if (ret < 0) {
1647			if ((BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1648						      BTRFS_INODE_PREALLOC)) &&
1649			    check_can_nocow(BTRFS_I(inode), pos,
1650					&write_bytes) > 0) {
1651				/*
1652				 * For nodata cow case, no need to reserve
1653				 * data space.
1654				 */
1655				only_release_metadata = true;
1656				/*
1657				 * our prealloc extent may be smaller than
1658				 * write_bytes, so scale down.
1659				 */
1660				num_pages = DIV_ROUND_UP(write_bytes + offset,
1661							 PAGE_SIZE);
1662				reserve_bytes = round_up(write_bytes +
1663							 sector_offset,
1664							 fs_info->sectorsize);
1665			} else {
1666				break;
1667			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1668		}
1669
 
 
 
 
1670		WARN_ON(reserve_bytes == 0);
1671		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1672				reserve_bytes);
 
1673		if (ret) {
1674			if (!only_release_metadata)
1675				btrfs_free_reserved_data_space(inode,
1676						data_reserved, pos,
1677						write_bytes);
1678			else
1679				btrfs_end_write_no_snapshotting(root);
 
 
 
1680			break;
1681		}
1682
1683		release_bytes = reserve_bytes;
1684again:
 
 
 
 
 
 
1685		/*
1686		 * This is going to setup the pages array with the number of
1687		 * pages we want, so we don't really need to worry about the
1688		 * contents of pages from loop to loop
1689		 */
1690		ret = prepare_pages(inode, pages, num_pages,
1691				    pos, write_bytes,
1692				    force_page_uptodate);
1693		if (ret) {
1694			btrfs_delalloc_release_extents(BTRFS_I(inode),
1695						       reserve_bytes);
1696			break;
1697		}
1698
1699		extents_locked = lock_and_cleanup_extent_if_need(
1700				BTRFS_I(inode), pages,
1701				num_pages, pos, write_bytes, &lockstart,
1702				&lockend, &cached_state);
1703		if (extents_locked < 0) {
1704			if (extents_locked == -EAGAIN)
1705				goto again;
 
1706			btrfs_delalloc_release_extents(BTRFS_I(inode),
1707						       reserve_bytes);
1708			ret = extents_locked;
1709			break;
1710		}
1711
1712		copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1713
1714		num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1715		dirty_sectors = round_up(copied + sector_offset,
1716					fs_info->sectorsize);
1717		dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1718
1719		/*
1720		 * if we have trouble faulting in the pages, fall
1721		 * back to one page at a time
1722		 */
1723		if (copied < write_bytes)
1724			nrptrs = 1;
1725
1726		if (copied == 0) {
1727			force_page_uptodate = true;
1728			dirty_sectors = 0;
1729			dirty_pages = 0;
1730		} else {
1731			force_page_uptodate = false;
1732			dirty_pages = DIV_ROUND_UP(copied + offset,
1733						   PAGE_SIZE);
1734		}
1735
1736		if (num_sectors > dirty_sectors) {
1737			/* release everything except the sectors we dirtied */
1738			release_bytes -= dirty_sectors <<
1739						fs_info->sb->s_blocksize_bits;
1740			if (only_release_metadata) {
1741				btrfs_delalloc_release_metadata(BTRFS_I(inode),
1742							release_bytes, true);
1743			} else {
1744				u64 __pos;
1745
1746				__pos = round_down(pos,
1747						   fs_info->sectorsize) +
1748					(dirty_pages << PAGE_SHIFT);
1749				btrfs_delalloc_release_space(inode,
1750						data_reserved, __pos,
1751						release_bytes, true);
1752			}
1753		}
1754
1755		release_bytes = round_up(copied + sector_offset,
1756					fs_info->sectorsize);
1757
1758		if (copied > 0)
1759			ret = btrfs_dirty_pages(inode, pages, dirty_pages,
1760						pos, copied, &cached_state);
1761
1762		/*
1763		 * If we have not locked the extent range, because the range's
1764		 * start offset is >= i_size, we might still have a non-NULL
1765		 * cached extent state, acquired while marking the extent range
1766		 * as delalloc through btrfs_dirty_pages(). Therefore free any
1767		 * possible cached extent state to avoid a memory leak.
1768		 */
1769		if (extents_locked)
1770			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1771					     lockstart, lockend, &cached_state);
1772		else
1773			free_extent_state(cached_state);
1774
1775		btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1776		if (ret) {
1777			btrfs_drop_pages(pages, num_pages);
1778			break;
1779		}
1780
1781		release_bytes = 0;
1782		if (only_release_metadata)
1783			btrfs_end_write_no_snapshotting(root);
1784
1785		if (only_release_metadata && copied > 0) {
1786			lockstart = round_down(pos,
1787					       fs_info->sectorsize);
1788			lockend = round_up(pos + copied,
1789					   fs_info->sectorsize) - 1;
1790
1791			set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
1792				       lockend, EXTENT_NORESERVE, NULL,
1793				       NULL, GFP_NOFS);
1794			only_release_metadata = false;
1795		}
1796
1797		btrfs_drop_pages(pages, num_pages);
1798
1799		cond_resched();
1800
1801		balance_dirty_pages_ratelimited(inode->i_mapping);
1802		if (dirty_pages < (fs_info->nodesize >> PAGE_SHIFT) + 1)
1803			btrfs_btree_balance_dirty(fs_info);
1804
1805		pos += copied;
1806		num_written += copied;
1807	}
1808
1809	kfree(pages);
1810
1811	if (release_bytes) {
1812		if (only_release_metadata) {
1813			btrfs_end_write_no_snapshotting(root);
1814			btrfs_delalloc_release_metadata(BTRFS_I(inode),
1815					release_bytes, true);
1816		} else {
1817			btrfs_delalloc_release_space(inode, data_reserved,
 
1818					round_down(pos, fs_info->sectorsize),
1819					release_bytes, true);
1820		}
1821	}
1822
1823	extent_changeset_free(data_reserved);
 
 
 
 
 
 
1824	return num_written ? num_written : ret;
1825}
1826
1827static ssize_t __btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1828{
1829	struct file *file = iocb->ki_filp;
1830	struct inode *inode = file_inode(file);
 
1831	loff_t pos;
1832	ssize_t written;
1833	ssize_t written_buffered;
 
1834	loff_t endbyte;
1835	int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1836
1837	written = generic_file_direct_write(iocb, from);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1838
1839	if (written < 0 || !iov_iter_count(from))
1840		return written;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1841
1842	pos = iocb->ki_pos;
1843	written_buffered = btrfs_buffered_write(iocb, from);
1844	if (written_buffered < 0) {
1845		err = written_buffered;
1846		goto out;
1847	}
1848	/*
1849	 * Ensure all data is persisted. We want the next direct IO read to be
1850	 * able to read what was just written.
1851	 */
1852	endbyte = pos + written_buffered - 1;
1853	err = btrfs_fdatawrite_range(inode, pos, endbyte);
1854	if (err)
1855		goto out;
1856	err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1857	if (err)
1858		goto out;
1859	written += written_buffered;
1860	iocb->ki_pos = pos + written_buffered;
1861	invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1862				 endbyte >> PAGE_SHIFT);
1863out:
1864	return written ? written : err;
1865}
1866
1867static void update_time_for_write(struct inode *inode)
1868{
1869	struct timespec64 now;
1870
1871	if (IS_NOCMTIME(inode))
1872		return;
1873
1874	now = current_time(inode);
1875	if (!timespec64_equal(&inode->i_mtime, &now))
1876		inode->i_mtime = now;
1877
1878	if (!timespec64_equal(&inode->i_ctime, &now))
1879		inode->i_ctime = now;
1880
1881	if (IS_I_VERSION(inode))
1882		inode_inc_iversion(inode);
1883}
1884
1885static ssize_t btrfs_file_write_iter(struct kiocb *iocb,
1886				    struct iov_iter *from)
1887{
1888	struct file *file = iocb->ki_filp;
1889	struct inode *inode = file_inode(file);
1890	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1891	struct btrfs_root *root = BTRFS_I(inode)->root;
1892	u64 start_pos;
1893	u64 end_pos;
1894	ssize_t num_written = 0;
1895	const bool sync = iocb->ki_flags & IOCB_DSYNC;
1896	ssize_t err;
1897	loff_t pos;
1898	size_t count;
1899	loff_t oldsize;
1900	int clean_page = 0;
1901
1902	if (!(iocb->ki_flags & IOCB_DIRECT) &&
1903	    (iocb->ki_flags & IOCB_NOWAIT))
1904		return -EOPNOTSUPP;
1905
1906	if (!inode_trylock(inode)) {
1907		if (iocb->ki_flags & IOCB_NOWAIT)
1908			return -EAGAIN;
1909		inode_lock(inode);
1910	}
1911
1912	err = generic_write_checks(iocb, from);
1913	if (err <= 0) {
1914		inode_unlock(inode);
1915		return err;
1916	}
1917
1918	pos = iocb->ki_pos;
1919	count = iov_iter_count(from);
1920	if (iocb->ki_flags & IOCB_NOWAIT) {
1921		/*
1922		 * We will allocate space in case nodatacow is not set,
1923		 * so bail
1924		 */
1925		if (!(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW |
1926					      BTRFS_INODE_PREALLOC)) ||
1927		    check_can_nocow(BTRFS_I(inode), pos, &count) <= 0) {
1928			inode_unlock(inode);
1929			return -EAGAIN;
1930		}
1931	}
1932
1933	current->backing_dev_info = inode_to_bdi(inode);
1934	err = file_remove_privs(file);
1935	if (err) {
1936		inode_unlock(inode);
1937		goto out;
1938	}
1939
1940	/*
1941	 * If BTRFS flips readonly due to some impossible error
1942	 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1943	 * although we have opened a file as writable, we have
1944	 * to stop this write operation to ensure FS consistency.
1945	 */
1946	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
1947		inode_unlock(inode);
1948		err = -EROFS;
1949		goto out;
1950	}
 
 
 
 
 
 
 
 
 
 
 
 
1951
1952	/*
1953	 * We reserve space for updating the inode when we reserve space for the
1954	 * extent we are going to write, so we will enospc out there.  We don't
1955	 * need to start yet another transaction to update the inode as we will
1956	 * update the inode when we finish writing whatever data we write.
1957	 */
1958	update_time_for_write(inode);
 
1959
1960	start_pos = round_down(pos, fs_info->sectorsize);
1961	oldsize = i_size_read(inode);
1962	if (start_pos > oldsize) {
1963		/* Expand hole size to cover write data, preventing empty gap */
1964		end_pos = round_up(pos + count,
1965				   fs_info->sectorsize);
1966		err = btrfs_cont_expand(inode, oldsize, end_pos);
1967		if (err) {
1968			inode_unlock(inode);
1969			goto out;
1970		}
1971		if (start_pos > round_up(oldsize, fs_info->sectorsize))
1972			clean_page = 1;
1973	}
1974
1975	if (sync)
1976		atomic_inc(&BTRFS_I(inode)->sync_writers);
1977
1978	if (iocb->ki_flags & IOCB_DIRECT) {
1979		num_written = __btrfs_direct_write(iocb, from);
 
 
 
 
1980	} else {
1981		num_written = btrfs_buffered_write(iocb, from);
1982		if (num_written > 0)
1983			iocb->ki_pos = pos + num_written;
1984		if (clean_page)
1985			pagecache_isize_extended(inode, oldsize,
1986						i_size_read(inode));
1987	}
1988
1989	inode_unlock(inode);
1990
1991	/*
1992	 * We also have to set last_sub_trans to the current log transid,
1993	 * otherwise subsequent syncs to a file that's been synced in this
1994	 * transaction will appear to have already occurred.
1995	 */
1996	spin_lock(&BTRFS_I(inode)->lock);
1997	BTRFS_I(inode)->last_sub_trans = root->log_transid;
1998	spin_unlock(&BTRFS_I(inode)->lock);
1999	if (num_written > 0)
2000		num_written = generic_write_sync(iocb, num_written);
2001
2002	if (sync)
2003		atomic_dec(&BTRFS_I(inode)->sync_writers);
2004out:
2005	current->backing_dev_info = NULL;
2006	return num_written ? num_written : err;
 
2007}
2008
2009int btrfs_release_file(struct inode *inode, struct file *filp)
2010{
2011	struct btrfs_file_private *private = filp->private_data;
2012
2013	if (private && private->filldir_buf)
2014		kfree(private->filldir_buf);
2015	kfree(private);
2016	filp->private_data = NULL;
 
 
2017
2018	/*
2019	 * ordered_data_close is set by setattr when we are about to truncate
2020	 * a file from a non-zero size to a zero size.  This tries to
2021	 * flush down new bytes that may have been written if the
2022	 * application were using truncate to replace a file in place.
2023	 */
2024	if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
2025			       &BTRFS_I(inode)->runtime_flags))
2026			filemap_flush(inode->i_mapping);
2027	return 0;
2028}
2029
2030static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
2031{
2032	int ret;
2033	struct blk_plug plug;
2034
2035	/*
2036	 * This is only called in fsync, which would do synchronous writes, so
2037	 * a plug can merge adjacent IOs as much as possible.  Esp. in case of
2038	 * multiple disks using raid profile, a large IO can be split to
2039	 * several segments of stripe length (currently 64K).
2040	 */
2041	blk_start_plug(&plug);
2042	atomic_inc(&BTRFS_I(inode)->sync_writers);
2043	ret = btrfs_fdatawrite_range(inode, start, end);
2044	atomic_dec(&BTRFS_I(inode)->sync_writers);
2045	blk_finish_plug(&plug);
2046
2047	return ret;
2048}
2049
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2050/*
2051 * fsync call for both files and directories.  This logs the inode into
2052 * the tree log instead of forcing full commits whenever possible.
2053 *
2054 * It needs to call filemap_fdatawait so that all ordered extent updates are
2055 * in the metadata btree are up to date for copying to the log.
2056 *
2057 * It drops the inode mutex before doing the tree log commit.  This is an
2058 * important optimization for directories because holding the mutex prevents
2059 * new operations on the dir while we write to disk.
2060 */
2061int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
2062{
2063	struct dentry *dentry = file_dentry(file);
2064	struct inode *inode = d_inode(dentry);
2065	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2066	struct btrfs_root *root = BTRFS_I(inode)->root;
2067	struct btrfs_trans_handle *trans;
2068	struct btrfs_log_ctx ctx;
2069	int ret = 0, err;
 
 
2070
2071	trace_btrfs_sync_file(file, datasync);
2072
2073	btrfs_init_log_ctx(&ctx, inode);
2074
2075	/*
 
 
 
 
 
 
 
 
 
 
 
 
2076	 * We write the dirty pages in the range and wait until they complete
2077	 * out of the ->i_mutex. If so, we can flush the dirty pages by
2078	 * multi-task, and make the performance up.  See
2079	 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
2080	 */
2081	ret = start_ordered_ops(inode, start, end);
2082	if (ret)
2083		goto out;
2084
2085	inode_lock(inode);
2086
2087	/*
2088	 * We take the dio_sem here because the tree log stuff can race with
2089	 * lockless dio writes and get an extent map logged for an extent we
2090	 * never waited on.  We need it this high up for lockdep reasons.
2091	 */
2092	down_write(&BTRFS_I(inode)->dio_sem);
2093
2094	atomic_inc(&root->log_batch);
2095
2096	/*
2097	 * If the inode needs a full sync, make sure we use a full range to
2098	 * avoid log tree corruption, due to hole detection racing with ordered
2099	 * extent completion for adjacent ranges, and assertion failures during
2100	 * hole detection. Do this while holding the inode lock, to avoid races
2101	 * with other tasks.
2102	 */
2103	if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2104		     &BTRFS_I(inode)->runtime_flags)) {
2105		start = 0;
2106		end = LLONG_MAX;
2107	}
2108
2109	/*
2110	 * Before we acquired the inode's lock, someone may have dirtied more
2111	 * pages in the target range. We need to make sure that writeback for
2112	 * any such pages does not start while we are logging the inode, because
2113	 * if it does, any of the following might happen when we are not doing a
2114	 * full inode sync:
2115	 *
2116	 * 1) We log an extent after its writeback finishes but before its
2117	 *    checksums are added to the csum tree, leading to -EIO errors
2118	 *    when attempting to read the extent after a log replay.
2119	 *
2120	 * 2) We can end up logging an extent before its writeback finishes.
2121	 *    Therefore after the log replay we will have a file extent item
2122	 *    pointing to an unwritten extent (and no data checksums as well).
2123	 *
2124	 * So trigger writeback for any eventual new dirty pages and then we
2125	 * wait for all ordered extents to complete below.
2126	 */
2127	ret = start_ordered_ops(inode, start, end);
2128	if (ret) {
2129		inode_unlock(inode);
2130		goto out;
2131	}
2132
2133	/*
 
 
 
 
 
 
 
 
 
 
 
2134	 * We have to do this here to avoid the priority inversion of waiting on
2135	 * IO of a lower priority task while holding a transaction open.
2136	 *
2137	 * Also, the range length can be represented by u64, we have to do the
2138	 * typecasts to avoid signed overflow if it's [0, LLONG_MAX].
 
 
 
 
 
 
 
 
2139	 */
2140	ret = btrfs_wait_ordered_range(inode, start, (u64)end - (u64)start + 1);
2141	if (ret) {
2142		up_write(&BTRFS_I(inode)->dio_sem);
2143		inode_unlock(inode);
2144		goto out;
 
 
 
 
 
 
2145	}
 
 
 
 
2146	atomic_inc(&root->log_batch);
2147
2148	smp_mb();
2149	if (btrfs_inode_in_log(BTRFS_I(inode), fs_info->generation) ||
2150	    BTRFS_I(inode)->last_trans <= fs_info->last_trans_committed) {
2151		/*
2152		 * We've had everything committed since the last time we were
2153		 * modified so clear this flag in case it was set for whatever
2154		 * reason, it's no longer relevant.
2155		 */
2156		clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2157			  &BTRFS_I(inode)->runtime_flags);
2158		/*
2159		 * An ordered extent might have started before and completed
2160		 * already with io errors, in which case the inode was not
2161		 * updated and we end up here. So check the inode's mapping
2162		 * for any errors that might have happened since we last
2163		 * checked called fsync.
2164		 */
2165		ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
2166		up_write(&BTRFS_I(inode)->dio_sem);
2167		inode_unlock(inode);
2168		goto out;
2169	}
2170
 
 
2171	/*
2172	 * We use start here because we will need to wait on the IO to complete
2173	 * in btrfs_sync_log, which could require joining a transaction (for
2174	 * example checking cross references in the nocow path).  If we use join
2175	 * here we could get into a situation where we're waiting on IO to
2176	 * happen that is blocked on a transaction trying to commit.  With start
2177	 * we inc the extwriter counter, so we wait for all extwriters to exit
2178	 * before we start blocking joiners.  This comment is to keep somebody
2179	 * from thinking they are super smart and changing this to
2180	 * btrfs_join_transaction *cough*Josef*cough*.
2181	 */
2182	trans = btrfs_start_transaction(root, 0);
2183	if (IS_ERR(trans)) {
2184		ret = PTR_ERR(trans);
2185		up_write(&BTRFS_I(inode)->dio_sem);
2186		inode_unlock(inode);
2187		goto out;
2188	}
 
2189
2190	ret = btrfs_log_dentry_safe(trans, dentry, start, end, &ctx);
 
 
 
 
 
 
 
 
 
 
2191	if (ret < 0) {
2192		/* Fallthrough and commit/free transaction. */
2193		ret = 1;
2194	}
2195
2196	/* we've logged all the items and now have a consistent
2197	 * version of the file in the log.  It is possible that
2198	 * someone will come in and modify the file, but that's
2199	 * fine because the log is consistent on disk, and we
2200	 * have references to all of the file's extents
2201	 *
2202	 * It is possible that someone will come in and log the
2203	 * file again, but that will end up using the synchronization
2204	 * inside btrfs_sync_log to keep things safe.
2205	 */
2206	up_write(&BTRFS_I(inode)->dio_sem);
2207	inode_unlock(inode);
 
 
 
 
2208
2209	if (ret != BTRFS_NO_LOG_SYNC) {
 
 
2210		if (!ret) {
2211			ret = btrfs_sync_log(trans, root, &ctx);
2212			if (!ret) {
2213				ret = btrfs_end_transaction(trans);
2214				goto out;
2215			}
2216		}
2217		ret = btrfs_commit_transaction(trans);
2218	} else {
 
 
 
 
 
 
 
 
 
 
 
 
 
2219		ret = btrfs_end_transaction(trans);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2220	}
 
 
2221out:
 
2222	ASSERT(list_empty(&ctx.list));
 
2223	err = file_check_and_advance_wb_err(file);
2224	if (!ret)
2225		ret = err;
2226	return ret > 0 ? -EIO : ret;
 
 
 
 
 
2227}
2228
2229static const struct vm_operations_struct btrfs_file_vm_ops = {
2230	.fault		= filemap_fault,
2231	.map_pages	= filemap_map_pages,
2232	.page_mkwrite	= btrfs_page_mkwrite,
2233};
2234
2235static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
2236{
2237	struct address_space *mapping = filp->f_mapping;
2238
2239	if (!mapping->a_ops->readpage)
2240		return -ENOEXEC;
2241
2242	file_accessed(filp);
2243	vma->vm_ops = &btrfs_file_vm_ops;
2244
2245	return 0;
2246}
2247
2248static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2249			  int slot, u64 start, u64 end)
2250{
2251	struct btrfs_file_extent_item *fi;
2252	struct btrfs_key key;
2253
2254	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2255		return 0;
2256
2257	btrfs_item_key_to_cpu(leaf, &key, slot);
2258	if (key.objectid != btrfs_ino(inode) ||
2259	    key.type != BTRFS_EXTENT_DATA_KEY)
2260		return 0;
2261
2262	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2263
2264	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2265		return 0;
2266
2267	if (btrfs_file_extent_disk_bytenr(leaf, fi))
2268		return 0;
2269
2270	if (key.offset == end)
2271		return 1;
2272	if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2273		return 1;
2274	return 0;
2275}
2276
2277static int fill_holes(struct btrfs_trans_handle *trans,
2278		struct btrfs_inode *inode,
2279		struct btrfs_path *path, u64 offset, u64 end)
2280{
2281	struct btrfs_fs_info *fs_info = trans->fs_info;
2282	struct btrfs_root *root = inode->root;
2283	struct extent_buffer *leaf;
2284	struct btrfs_file_extent_item *fi;
2285	struct extent_map *hole_em;
2286	struct extent_map_tree *em_tree = &inode->extent_tree;
2287	struct btrfs_key key;
2288	int ret;
2289
2290	if (btrfs_fs_incompat(fs_info, NO_HOLES))
2291		goto out;
2292
2293	key.objectid = btrfs_ino(inode);
2294	key.type = BTRFS_EXTENT_DATA_KEY;
2295	key.offset = offset;
2296
2297	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2298	if (ret <= 0) {
2299		/*
2300		 * We should have dropped this offset, so if we find it then
2301		 * something has gone horribly wrong.
2302		 */
2303		if (ret == 0)
2304			ret = -EINVAL;
2305		return ret;
2306	}
2307
2308	leaf = path->nodes[0];
2309	if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2310		u64 num_bytes;
2311
2312		path->slots[0]--;
2313		fi = btrfs_item_ptr(leaf, path->slots[0],
2314				    struct btrfs_file_extent_item);
2315		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2316			end - offset;
2317		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2318		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2319		btrfs_set_file_extent_offset(leaf, fi, 0);
2320		btrfs_mark_buffer_dirty(leaf);
 
2321		goto out;
2322	}
2323
2324	if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2325		u64 num_bytes;
2326
2327		key.offset = offset;
2328		btrfs_set_item_key_safe(fs_info, path, &key);
2329		fi = btrfs_item_ptr(leaf, path->slots[0],
2330				    struct btrfs_file_extent_item);
2331		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2332			offset;
2333		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2334		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2335		btrfs_set_file_extent_offset(leaf, fi, 0);
2336		btrfs_mark_buffer_dirty(leaf);
 
2337		goto out;
2338	}
2339	btrfs_release_path(path);
2340
2341	ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
2342			offset, 0, 0, end - offset, 0, end - offset, 0, 0, 0);
2343	if (ret)
2344		return ret;
2345
2346out:
2347	btrfs_release_path(path);
2348
2349	hole_em = alloc_extent_map();
2350	if (!hole_em) {
2351		btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2352		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
2353	} else {
2354		hole_em->start = offset;
2355		hole_em->len = end - offset;
2356		hole_em->ram_bytes = hole_em->len;
2357		hole_em->orig_start = offset;
2358
2359		hole_em->block_start = EXTENT_MAP_HOLE;
2360		hole_em->block_len = 0;
2361		hole_em->orig_block_len = 0;
2362		hole_em->bdev = fs_info->fs_devices->latest_bdev;
2363		hole_em->compress_type = BTRFS_COMPRESS_NONE;
2364		hole_em->generation = trans->transid;
2365
2366		do {
2367			btrfs_drop_extent_cache(inode, offset, end - 1, 0);
2368			write_lock(&em_tree->lock);
2369			ret = add_extent_mapping(em_tree, hole_em, 1);
2370			write_unlock(&em_tree->lock);
2371		} while (ret == -EEXIST);
2372		free_extent_map(hole_em);
2373		if (ret)
2374			set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2375					&inode->runtime_flags);
2376	}
2377
2378	return 0;
2379}
2380
2381/*
2382 * Find a hole extent on given inode and change start/len to the end of hole
2383 * extent.(hole/vacuum extent whose em->start <= start &&
2384 *	   em->start + em->len > start)
2385 * When a hole extent is found, return 1 and modify start/len.
2386 */
2387static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len)
2388{
2389	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2390	struct extent_map *em;
2391	int ret = 0;
2392
2393	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2394			      round_down(*start, fs_info->sectorsize),
2395			      round_up(*len, fs_info->sectorsize), 0);
2396	if (IS_ERR(em))
2397		return PTR_ERR(em);
2398
2399	/* Hole or vacuum extent(only exists in no-hole mode) */
2400	if (em->block_start == EXTENT_MAP_HOLE) {
2401		ret = 1;
2402		*len = em->start + em->len > *start + *len ?
2403		       0 : *start + *len - em->start - em->len;
2404		*start = em->start + em->len;
2405	}
2406	free_extent_map(em);
2407	return ret;
2408}
2409
2410static int btrfs_punch_hole_lock_range(struct inode *inode,
2411				       const u64 lockstart,
2412				       const u64 lockend,
2413				       struct extent_state **cached_state)
2414{
2415	while (1) {
2416		struct btrfs_ordered_extent *ordered;
2417		int ret;
 
 
 
 
 
 
 
2418
 
2419		truncate_pagecache_range(inode, lockstart, lockend);
2420
2421		lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2422				 cached_state);
2423		ordered = btrfs_lookup_first_ordered_extent(inode, lockend);
2424
2425		/*
2426		 * We need to make sure we have no ordered extents in this range
2427		 * and nobody raced in and read a page in this range, if we did
2428		 * we need to try again.
2429		 */
2430		if ((!ordered ||
2431		    (ordered->file_offset + ordered->len <= lockstart ||
2432		     ordered->file_offset > lockend)) &&
2433		     !filemap_range_has_page(inode->i_mapping,
2434					     lockstart, lockend)) {
2435			if (ordered)
2436				btrfs_put_ordered_extent(ordered);
2437			break;
2438		}
2439		if (ordered)
2440			btrfs_put_ordered_extent(ordered);
2441		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
2442				     lockend, cached_state);
2443		ret = btrfs_wait_ordered_range(inode, lockstart,
2444					       lockend - lockstart + 1);
2445		if (ret)
2446			return ret;
2447	}
2448	return 0;
 
2449}
2450
2451static int btrfs_insert_clone_extent(struct btrfs_trans_handle *trans,
2452				     struct inode *inode,
2453				     struct btrfs_path *path,
2454				     struct btrfs_clone_extent_info *clone_info,
2455				     const u64 clone_len)
 
2456{
2457	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2458	struct btrfs_root *root = BTRFS_I(inode)->root;
2459	struct btrfs_file_extent_item *extent;
2460	struct extent_buffer *leaf;
2461	struct btrfs_key key;
2462	int slot;
2463	struct btrfs_ref ref = { 0 };
2464	u64 ref_offset;
2465	int ret;
2466
2467	if (clone_len == 0)
2468		return 0;
2469
2470	if (clone_info->disk_offset == 0 &&
2471	    btrfs_fs_incompat(fs_info, NO_HOLES))
 
2472		return 0;
 
2473
2474	key.objectid = btrfs_ino(BTRFS_I(inode));
2475	key.type = BTRFS_EXTENT_DATA_KEY;
2476	key.offset = clone_info->file_offset;
2477	ret = btrfs_insert_empty_item(trans, root, path, &key,
2478				      clone_info->item_size);
2479	if (ret)
2480		return ret;
2481	leaf = path->nodes[0];
2482	slot = path->slots[0];
2483	write_extent_buffer(leaf, clone_info->extent_buf,
2484			    btrfs_item_ptr_offset(leaf, slot),
2485			    clone_info->item_size);
2486	extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2487	btrfs_set_file_extent_offset(leaf, extent, clone_info->data_offset);
2488	btrfs_set_file_extent_num_bytes(leaf, extent, clone_len);
2489	btrfs_mark_buffer_dirty(leaf);
 
 
 
2490	btrfs_release_path(path);
2491
 
 
 
 
 
2492	/* If it's a hole, nothing more needs to be done. */
2493	if (clone_info->disk_offset == 0)
 
2494		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2495
2496	inode_add_bytes(inode, clone_len);
2497	btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2498			       clone_info->disk_offset,
2499			       clone_info->disk_len, 0);
2500	ref_offset = clone_info->file_offset - clone_info->data_offset;
2501	btrfs_init_data_ref(&ref, root->root_key.objectid,
2502			    btrfs_ino(BTRFS_I(inode)), ref_offset);
2503	ret = btrfs_inc_extent_ref(trans, &ref);
2504
2505	return ret;
2506}
2507
2508/*
2509 * The respective range must have been previously locked, as well as the inode.
2510 * The end offset is inclusive (last byte of the range).
2511 * @clone_info is NULL for fallocate's hole punching and non-NULL for extent
2512 * cloning.
2513 * When cloning, we don't want to end up in a state where we dropped extents
2514 * without inserting a new one, so we must abort the transaction to avoid a
2515 * corruption.
2516 */
2517int btrfs_punch_hole_range(struct inode *inode, struct btrfs_path *path,
2518			   const u64 start, const u64 end,
2519			   struct btrfs_clone_extent_info *clone_info,
2520			   struct btrfs_trans_handle **trans_out)
 
2521{
2522	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 
 
2523	u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2524	u64 ino_size = round_up(inode->i_size, fs_info->sectorsize);
2525	struct btrfs_root *root = BTRFS_I(inode)->root;
2526	struct btrfs_trans_handle *trans = NULL;
2527	struct btrfs_block_rsv *rsv;
2528	unsigned int rsv_count;
2529	u64 cur_offset;
2530	u64 drop_end;
2531	u64 len = end - start;
2532	int ret = 0;
2533
2534	if (end <= start)
2535		return -EINVAL;
2536
2537	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2538	if (!rsv) {
2539		ret = -ENOMEM;
2540		goto out;
2541	}
2542	rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2543	rsv->failfast = 1;
2544
2545	/*
2546	 * 1 - update the inode
2547	 * 1 - removing the extents in the range
2548	 * 1 - adding the hole extent if no_holes isn't set or if we are cloning
2549	 *     an extent
2550	 */
2551	if (!btrfs_fs_incompat(fs_info, NO_HOLES) || clone_info)
2552		rsv_count = 3;
2553	else
2554		rsv_count = 2;
2555
2556	trans = btrfs_start_transaction(root, rsv_count);
2557	if (IS_ERR(trans)) {
2558		ret = PTR_ERR(trans);
2559		trans = NULL;
2560		goto out_free;
2561	}
2562
2563	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2564				      min_size, false);
2565	BUG_ON(ret);
 
2566	trans->block_rsv = rsv;
2567
2568	cur_offset = start;
 
 
 
2569	while (cur_offset < end) {
2570		ret = __btrfs_drop_extents(trans, root, inode, path,
2571					   cur_offset, end + 1, &drop_end,
2572					   1, 0, 0, NULL);
 
 
 
2573		if (ret != -ENOSPC) {
2574			/*
2575			 * When cloning we want to avoid transaction aborts when
2576			 * nothing was done and we are attempting to clone parts
2577			 * of inline extents, in such cases -EOPNOTSUPP is
2578			 * returned by __btrfs_drop_extents() without having
2579			 * changed anything in the file.
 
2580			 */
2581			if (clone_info && ret && ret != -EOPNOTSUPP)
 
 
2582				btrfs_abort_transaction(trans, ret);
2583			break;
2584		}
2585
2586		trans->block_rsv = &fs_info->trans_block_rsv;
2587
2588		if (!clone_info && cur_offset < drop_end &&
2589		    cur_offset < ino_size) {
2590			ret = fill_holes(trans, BTRFS_I(inode), path,
2591					cur_offset, drop_end);
2592			if (ret) {
2593				/*
2594				 * If we failed then we didn't insert our hole
2595				 * entries for the area we dropped, so now the
2596				 * fs is corrupted, so we must abort the
2597				 * transaction.
2598				 */
2599				btrfs_abort_transaction(trans, ret);
2600				break;
2601			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2602		}
2603
2604		if (clone_info) {
2605			u64 clone_len = drop_end - cur_offset;
2606
2607			ret = btrfs_insert_clone_extent(trans, inode, path,
2608							clone_info, clone_len);
 
 
 
2609			if (ret) {
2610				btrfs_abort_transaction(trans, ret);
2611				break;
2612			}
2613			clone_info->data_len -= clone_len;
2614			clone_info->data_offset += clone_len;
2615			clone_info->file_offset += clone_len;
2616		}
2617
2618		cur_offset = drop_end;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2619
2620		ret = btrfs_update_inode(trans, root, inode);
2621		if (ret)
2622			break;
2623
2624		btrfs_end_transaction(trans);
2625		btrfs_btree_balance_dirty(fs_info);
2626
2627		trans = btrfs_start_transaction(root, rsv_count);
2628		if (IS_ERR(trans)) {
2629			ret = PTR_ERR(trans);
2630			trans = NULL;
2631			break;
2632		}
2633
2634		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2635					      rsv, min_size, false);
2636		BUG_ON(ret);	/* shouldn't happen */
 
2637		trans->block_rsv = rsv;
2638
2639		if (!clone_info) {
 
 
2640			ret = find_first_non_hole(inode, &cur_offset, &len);
2641			if (unlikely(ret < 0))
2642				break;
2643			if (ret && !len) {
2644				ret = 0;
2645				break;
2646			}
2647		}
2648	}
2649
2650	/*
2651	 * If we were cloning, force the next fsync to be a full one since we
2652	 * we replaced (or just dropped in the case of cloning holes when
2653	 * NO_HOLES is enabled) extents and extent maps.
2654	 * This is for the sake of simplicity, and cloning into files larger
2655	 * than 16Mb would force the full fsync any way (when
2656	 * try_release_extent_mapping() is invoked during page cache truncation.
2657	 */
2658	if (clone_info)
2659		set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2660			&BTRFS_I(inode)->runtime_flags);
2661
2662	if (ret)
2663		goto out_trans;
2664
2665	trans->block_rsv = &fs_info->trans_block_rsv;
2666	/*
2667	 * If we are using the NO_HOLES feature we might have had already an
2668	 * hole that overlaps a part of the region [lockstart, lockend] and
2669	 * ends at (or beyond) lockend. Since we have no file extent items to
2670	 * represent holes, drop_end can be less than lockend and so we must
2671	 * make sure we have an extent map representing the existing hole (the
2672	 * call to __btrfs_drop_extents() might have dropped the existing extent
2673	 * map representing the existing hole), otherwise the fast fsync path
2674	 * will not record the existence of the hole region
2675	 * [existing_hole_start, lockend].
2676	 */
2677	if (drop_end <= end)
2678		drop_end = end + 1;
2679	/*
2680	 * Don't insert file hole extent item if it's for a range beyond eof
2681	 * (because it's useless) or if it represents a 0 bytes range (when
2682	 * cur_offset == drop_end).
2683	 */
2684	if (!clone_info && cur_offset < ino_size && cur_offset < drop_end) {
2685		ret = fill_holes(trans, BTRFS_I(inode), path,
2686				cur_offset, drop_end);
 
2687		if (ret) {
2688			/* Same comment as above. */
2689			btrfs_abort_transaction(trans, ret);
2690			goto out_trans;
2691		}
 
 
 
 
 
 
 
 
 
2692	}
2693	if (clone_info) {
2694		ret = btrfs_insert_clone_extent(trans, inode, path, clone_info,
2695						clone_info->data_len);
 
2696		if (ret) {
2697			btrfs_abort_transaction(trans, ret);
2698			goto out_trans;
2699		}
2700	}
2701
2702out_trans:
2703	if (!trans)
2704		goto out_free;
2705
2706	trans->block_rsv = &fs_info->trans_block_rsv;
2707	if (ret)
2708		btrfs_end_transaction(trans);
2709	else
2710		*trans_out = trans;
2711out_free:
2712	btrfs_free_block_rsv(fs_info, rsv);
2713out:
2714	return ret;
2715}
2716
2717static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
2718{
2719	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 
2720	struct btrfs_root *root = BTRFS_I(inode)->root;
2721	struct extent_state *cached_state = NULL;
2722	struct btrfs_path *path;
2723	struct btrfs_trans_handle *trans = NULL;
2724	u64 lockstart;
2725	u64 lockend;
2726	u64 tail_start;
2727	u64 tail_len;
2728	u64 orig_start = offset;
2729	int ret = 0;
2730	bool same_block;
2731	u64 ino_size;
2732	bool truncated_block = false;
2733	bool updated_inode = false;
2734
 
 
2735	ret = btrfs_wait_ordered_range(inode, offset, len);
2736	if (ret)
2737		return ret;
2738
2739	inode_lock(inode);
2740	ino_size = round_up(inode->i_size, fs_info->sectorsize);
2741	ret = find_first_non_hole(inode, &offset, &len);
2742	if (ret < 0)
2743		goto out_only_mutex;
2744	if (ret && !len) {
2745		/* Already in a large hole */
2746		ret = 0;
2747		goto out_only_mutex;
2748	}
2749
2750	lockstart = round_up(offset, btrfs_inode_sectorsize(inode));
2751	lockend = round_down(offset + len,
2752			     btrfs_inode_sectorsize(inode)) - 1;
 
 
 
2753	same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2754		== (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2755	/*
2756	 * We needn't truncate any block which is beyond the end of the file
2757	 * because we are sure there is no data there.
2758	 */
2759	/*
2760	 * Only do this if we are in the same block and we aren't doing the
2761	 * entire block.
2762	 */
2763	if (same_block && len < fs_info->sectorsize) {
2764		if (offset < ino_size) {
2765			truncated_block = true;
2766			ret = btrfs_truncate_block(inode, offset, len, 0);
 
2767		} else {
2768			ret = 0;
2769		}
2770		goto out_only_mutex;
2771	}
2772
2773	/* zero back part of the first block */
2774	if (offset < ino_size) {
2775		truncated_block = true;
2776		ret = btrfs_truncate_block(inode, offset, 0, 0);
2777		if (ret) {
2778			inode_unlock(inode);
2779			return ret;
2780		}
2781	}
2782
2783	/* Check the aligned pages after the first unaligned page,
2784	 * if offset != orig_start, which means the first unaligned page
2785	 * including several following pages are already in holes,
2786	 * the extra check can be skipped */
2787	if (offset == orig_start) {
2788		/* after truncate page, check hole again */
2789		len = offset + len - lockstart;
2790		offset = lockstart;
2791		ret = find_first_non_hole(inode, &offset, &len);
2792		if (ret < 0)
2793			goto out_only_mutex;
2794		if (ret && !len) {
2795			ret = 0;
2796			goto out_only_mutex;
2797		}
2798		lockstart = offset;
2799	}
2800
2801	/* Check the tail unaligned part is in a hole */
2802	tail_start = lockend + 1;
2803	tail_len = offset + len - tail_start;
2804	if (tail_len) {
2805		ret = find_first_non_hole(inode, &tail_start, &tail_len);
2806		if (unlikely(ret < 0))
2807			goto out_only_mutex;
2808		if (!ret) {
2809			/* zero the front end of the last page */
2810			if (tail_start + tail_len < ino_size) {
2811				truncated_block = true;
2812				ret = btrfs_truncate_block(inode,
2813							tail_start + tail_len,
2814							0, 1);
2815				if (ret)
2816					goto out_only_mutex;
2817			}
2818		}
2819	}
2820
2821	if (lockend < lockstart) {
2822		ret = 0;
2823		goto out_only_mutex;
2824	}
2825
2826	ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
2827					  &cached_state);
2828	if (ret)
2829		goto out_only_mutex;
2830
2831	path = btrfs_alloc_path();
2832	if (!path) {
2833		ret = -ENOMEM;
2834		goto out;
2835	}
2836
2837	ret = btrfs_punch_hole_range(inode, path, lockstart, lockend, NULL,
2838				     &trans);
2839	btrfs_free_path(path);
2840	if (ret)
2841		goto out;
2842
2843	ASSERT(trans != NULL);
2844	inode_inc_iversion(inode);
2845	inode->i_mtime = inode->i_ctime = current_time(inode);
2846	ret = btrfs_update_inode(trans, root, inode);
2847	updated_inode = true;
2848	btrfs_end_transaction(trans);
2849	btrfs_btree_balance_dirty(fs_info);
2850out:
2851	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2852			     &cached_state);
2853out_only_mutex:
2854	if (!updated_inode && truncated_block && !ret) {
2855		/*
2856		 * If we only end up zeroing part of a page, we still need to
2857		 * update the inode item, so that all the time fields are
2858		 * updated as well as the necessary btrfs inode in memory fields
2859		 * for detecting, at fsync time, if the inode isn't yet in the
2860		 * log tree or it's there but not up to date.
2861		 */
2862		struct timespec64 now = current_time(inode);
2863
2864		inode_inc_iversion(inode);
2865		inode->i_mtime = now;
2866		inode->i_ctime = now;
2867		trans = btrfs_start_transaction(root, 1);
2868		if (IS_ERR(trans)) {
2869			ret = PTR_ERR(trans);
2870		} else {
2871			int ret2;
2872
2873			ret = btrfs_update_inode(trans, root, inode);
2874			ret2 = btrfs_end_transaction(trans);
2875			if (!ret)
2876				ret = ret2;
2877		}
2878	}
2879	inode_unlock(inode);
2880	return ret;
2881}
2882
2883/* Helper structure to record which range is already reserved */
2884struct falloc_range {
2885	struct list_head list;
2886	u64 start;
2887	u64 len;
2888};
2889
2890/*
2891 * Helper function to add falloc range
2892 *
2893 * Caller should have locked the larger range of extent containing
2894 * [start, len)
2895 */
2896static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2897{
2898	struct falloc_range *prev = NULL;
2899	struct falloc_range *range = NULL;
2900
2901	if (list_empty(head))
2902		goto insert;
2903
2904	/*
2905	 * As fallocate iterate by bytenr order, we only need to check
2906	 * the last range.
2907	 */
2908	prev = list_entry(head->prev, struct falloc_range, list);
2909	if (prev->start + prev->len == start) {
2910		prev->len += len;
2911		return 0;
2912	}
2913insert:
2914	range = kmalloc(sizeof(*range), GFP_KERNEL);
2915	if (!range)
2916		return -ENOMEM;
2917	range->start = start;
2918	range->len = len;
2919	list_add_tail(&range->list, head);
2920	return 0;
2921}
2922
2923static int btrfs_fallocate_update_isize(struct inode *inode,
2924					const u64 end,
2925					const int mode)
2926{
2927	struct btrfs_trans_handle *trans;
2928	struct btrfs_root *root = BTRFS_I(inode)->root;
2929	int ret;
2930	int ret2;
2931
2932	if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2933		return 0;
2934
2935	trans = btrfs_start_transaction(root, 1);
2936	if (IS_ERR(trans))
2937		return PTR_ERR(trans);
2938
2939	inode->i_ctime = current_time(inode);
2940	i_size_write(inode, end);
2941	btrfs_ordered_update_i_size(inode, end, NULL);
2942	ret = btrfs_update_inode(trans, root, inode);
2943	ret2 = btrfs_end_transaction(trans);
2944
2945	return ret ? ret : ret2;
2946}
2947
2948enum {
2949	RANGE_BOUNDARY_WRITTEN_EXTENT,
2950	RANGE_BOUNDARY_PREALLOC_EXTENT,
2951	RANGE_BOUNDARY_HOLE,
2952};
2953
2954static int btrfs_zero_range_check_range_boundary(struct inode *inode,
2955						 u64 offset)
2956{
2957	const u64 sectorsize = btrfs_inode_sectorsize(inode);
2958	struct extent_map *em;
2959	int ret;
2960
2961	offset = round_down(offset, sectorsize);
2962	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, offset, sectorsize, 0);
2963	if (IS_ERR(em))
2964		return PTR_ERR(em);
2965
2966	if (em->block_start == EXTENT_MAP_HOLE)
2967		ret = RANGE_BOUNDARY_HOLE;
2968	else if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
2969		ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2970	else
2971		ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2972
2973	free_extent_map(em);
2974	return ret;
2975}
2976
2977static int btrfs_zero_range(struct inode *inode,
2978			    loff_t offset,
2979			    loff_t len,
2980			    const int mode)
2981{
2982	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2983	struct extent_map *em;
2984	struct extent_changeset *data_reserved = NULL;
2985	int ret;
2986	u64 alloc_hint = 0;
2987	const u64 sectorsize = btrfs_inode_sectorsize(inode);
2988	u64 alloc_start = round_down(offset, sectorsize);
2989	u64 alloc_end = round_up(offset + len, sectorsize);
2990	u64 bytes_to_reserve = 0;
2991	bool space_reserved = false;
2992
2993	inode_dio_wait(inode);
2994
2995	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
2996			      alloc_start, alloc_end - alloc_start, 0);
2997	if (IS_ERR(em)) {
2998		ret = PTR_ERR(em);
2999		goto out;
3000	}
3001
3002	/*
3003	 * Avoid hole punching and extent allocation for some cases. More cases
3004	 * could be considered, but these are unlikely common and we keep things
3005	 * as simple as possible for now. Also, intentionally, if the target
3006	 * range contains one or more prealloc extents together with regular
3007	 * extents and holes, we drop all the existing extents and allocate a
3008	 * new prealloc extent, so that we get a larger contiguous disk extent.
3009	 */
3010	if (em->start <= alloc_start &&
3011	    test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3012		const u64 em_end = em->start + em->len;
3013
3014		if (em_end >= offset + len) {
3015			/*
3016			 * The whole range is already a prealloc extent,
3017			 * do nothing except updating the inode's i_size if
3018			 * needed.
3019			 */
3020			free_extent_map(em);
3021			ret = btrfs_fallocate_update_isize(inode, offset + len,
3022							   mode);
3023			goto out;
3024		}
3025		/*
3026		 * Part of the range is already a prealloc extent, so operate
3027		 * only on the remaining part of the range.
3028		 */
3029		alloc_start = em_end;
3030		ASSERT(IS_ALIGNED(alloc_start, sectorsize));
3031		len = offset + len - alloc_start;
3032		offset = alloc_start;
3033		alloc_hint = em->block_start + em->len;
3034	}
3035	free_extent_map(em);
3036
3037	if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
3038	    BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
3039		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0,
3040				      alloc_start, sectorsize, 0);
3041		if (IS_ERR(em)) {
3042			ret = PTR_ERR(em);
3043			goto out;
3044		}
3045
3046		if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3047			free_extent_map(em);
3048			ret = btrfs_fallocate_update_isize(inode, offset + len,
3049							   mode);
3050			goto out;
3051		}
3052		if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
3053			free_extent_map(em);
3054			ret = btrfs_truncate_block(inode, offset, len, 0);
 
3055			if (!ret)
3056				ret = btrfs_fallocate_update_isize(inode,
3057								   offset + len,
3058								   mode);
3059			return ret;
3060		}
3061		free_extent_map(em);
3062		alloc_start = round_down(offset, sectorsize);
3063		alloc_end = alloc_start + sectorsize;
3064		goto reserve_space;
3065	}
3066
3067	alloc_start = round_up(offset, sectorsize);
3068	alloc_end = round_down(offset + len, sectorsize);
3069
3070	/*
3071	 * For unaligned ranges, check the pages at the boundaries, they might
3072	 * map to an extent, in which case we need to partially zero them, or
3073	 * they might map to a hole, in which case we need our allocation range
3074	 * to cover them.
3075	 */
3076	if (!IS_ALIGNED(offset, sectorsize)) {
3077		ret = btrfs_zero_range_check_range_boundary(inode, offset);
 
3078		if (ret < 0)
3079			goto out;
3080		if (ret == RANGE_BOUNDARY_HOLE) {
3081			alloc_start = round_down(offset, sectorsize);
3082			ret = 0;
3083		} else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3084			ret = btrfs_truncate_block(inode, offset, 0, 0);
3085			if (ret)
3086				goto out;
3087		} else {
3088			ret = 0;
3089		}
3090	}
3091
3092	if (!IS_ALIGNED(offset + len, sectorsize)) {
3093		ret = btrfs_zero_range_check_range_boundary(inode,
3094							    offset + len);
3095		if (ret < 0)
3096			goto out;
3097		if (ret == RANGE_BOUNDARY_HOLE) {
3098			alloc_end = round_up(offset + len, sectorsize);
3099			ret = 0;
3100		} else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
3101			ret = btrfs_truncate_block(inode, offset + len, 0, 1);
 
3102			if (ret)
3103				goto out;
3104		} else {
3105			ret = 0;
3106		}
3107	}
3108
3109reserve_space:
3110	if (alloc_start < alloc_end) {
3111		struct extent_state *cached_state = NULL;
3112		const u64 lockstart = alloc_start;
3113		const u64 lockend = alloc_end - 1;
3114
3115		bytes_to_reserve = alloc_end - alloc_start;
3116		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3117						      bytes_to_reserve);
3118		if (ret < 0)
3119			goto out;
3120		space_reserved = true;
3121		ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
 
 
3122						alloc_start, bytes_to_reserve);
3123		if (ret)
3124			goto out;
3125		ret = btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3126						  &cached_state);
3127		if (ret)
3128			goto out;
 
3129		ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3130						alloc_end - alloc_start,
3131						i_blocksize(inode),
3132						offset + len, &alloc_hint);
3133		unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart,
3134				     lockend, &cached_state);
3135		/* btrfs_prealloc_file_range releases reserved space on error */
3136		if (ret) {
3137			space_reserved = false;
3138			goto out;
3139		}
3140	}
3141	ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3142 out:
3143	if (ret && space_reserved)
3144		btrfs_free_reserved_data_space(inode, data_reserved,
3145					       alloc_start, bytes_to_reserve);
3146	extent_changeset_free(data_reserved);
3147
3148	return ret;
3149}
3150
3151static long btrfs_fallocate(struct file *file, int mode,
3152			    loff_t offset, loff_t len)
3153{
3154	struct inode *inode = file_inode(file);
3155	struct extent_state *cached_state = NULL;
3156	struct extent_changeset *data_reserved = NULL;
3157	struct falloc_range *range;
3158	struct falloc_range *tmp;
3159	struct list_head reserve_list;
3160	u64 cur_offset;
3161	u64 last_byte;
3162	u64 alloc_start;
3163	u64 alloc_end;
3164	u64 alloc_hint = 0;
3165	u64 locked_end;
3166	u64 actual_end = 0;
 
 
 
3167	struct extent_map *em;
3168	int blocksize = btrfs_inode_sectorsize(inode);
3169	int ret;
3170
 
 
 
 
3171	alloc_start = round_down(offset, blocksize);
3172	alloc_end = round_up(offset + len, blocksize);
3173	cur_offset = alloc_start;
3174
3175	/* Make sure we aren't being give some crap mode */
3176	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3177		     FALLOC_FL_ZERO_RANGE))
3178		return -EOPNOTSUPP;
3179
3180	if (mode & FALLOC_FL_PUNCH_HOLE)
3181		return btrfs_punch_hole(inode, offset, len);
3182
3183	/*
3184	 * Only trigger disk allocation, don't trigger qgroup reserve
3185	 *
3186	 * For qgroup space, it will be checked later.
3187	 */
3188	if (!(mode & FALLOC_FL_ZERO_RANGE)) {
3189		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3190						      alloc_end - alloc_start);
3191		if (ret < 0)
3192			return ret;
3193	}
3194
3195	inode_lock(inode);
3196
3197	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3198		ret = inode_newsize_ok(inode, offset + len);
3199		if (ret)
3200			goto out;
3201	}
3202
 
 
 
 
3203	/*
3204	 * TODO: Move these two operations after we have checked
3205	 * accurate reserved space, or fallocate can still fail but
3206	 * with page truncated or size expanded.
3207	 *
3208	 * But that's a minor problem and won't do much harm BTW.
3209	 */
3210	if (alloc_start > inode->i_size) {
3211		ret = btrfs_cont_expand(inode, i_size_read(inode),
3212					alloc_start);
3213		if (ret)
3214			goto out;
3215	} else if (offset + len > inode->i_size) {
3216		/*
3217		 * If we are fallocating from the end of the file onward we
3218		 * need to zero out the end of the block if i_size lands in the
3219		 * middle of a block.
3220		 */
3221		ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
3222		if (ret)
3223			goto out;
3224	}
3225
3226	/*
3227	 * wait for ordered IO before we have any locks.  We'll loop again
3228	 * below with the locks held.
 
 
 
 
3229	 */
3230	ret = btrfs_wait_ordered_range(inode, alloc_start,
3231				       alloc_end - alloc_start);
3232	if (ret)
3233		goto out;
3234
3235	if (mode & FALLOC_FL_ZERO_RANGE) {
3236		ret = btrfs_zero_range(inode, offset, len, mode);
3237		inode_unlock(inode);
3238		return ret;
3239	}
3240
3241	locked_end = alloc_end - 1;
3242	while (1) {
3243		struct btrfs_ordered_extent *ordered;
3244
3245		/* the extent lock is ordered inside the running
3246		 * transaction
3247		 */
3248		lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
3249				 locked_end, &cached_state);
3250		ordered = btrfs_lookup_first_ordered_extent(inode, locked_end);
3251
3252		if (ordered &&
3253		    ordered->file_offset + ordered->len > alloc_start &&
3254		    ordered->file_offset < alloc_end) {
3255			btrfs_put_ordered_extent(ordered);
3256			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
3257					     alloc_start, locked_end,
3258					     &cached_state);
3259			/*
3260			 * we can't wait on the range with the transaction
3261			 * running or with the extent lock held
3262			 */
3263			ret = btrfs_wait_ordered_range(inode, alloc_start,
3264						       alloc_end - alloc_start);
3265			if (ret)
3266				goto out;
3267		} else {
3268			if (ordered)
3269				btrfs_put_ordered_extent(ordered);
3270			break;
3271		}
3272	}
3273
3274	/* First, check if we exceed the qgroup limit */
3275	INIT_LIST_HEAD(&reserve_list);
3276	while (cur_offset < alloc_end) {
3277		em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
3278				      alloc_end - cur_offset, 0);
3279		if (IS_ERR(em)) {
3280			ret = PTR_ERR(em);
3281			break;
3282		}
3283		last_byte = min(extent_map_end(em), alloc_end);
3284		actual_end = min_t(u64, extent_map_end(em), offset + len);
3285		last_byte = ALIGN(last_byte, blocksize);
3286		if (em->block_start == EXTENT_MAP_HOLE ||
3287		    (cur_offset >= inode->i_size &&
3288		     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
3289			ret = add_falloc_range(&reserve_list, cur_offset,
3290					       last_byte - cur_offset);
 
3291			if (ret < 0) {
3292				free_extent_map(em);
3293				break;
3294			}
3295			ret = btrfs_qgroup_reserve_data(inode, &data_reserved,
3296					cur_offset, last_byte - cur_offset);
3297			if (ret < 0) {
3298				cur_offset = last_byte;
3299				free_extent_map(em);
3300				break;
3301			}
3302		} else {
3303			/*
3304			 * Do not need to reserve unwritten extent for this
3305			 * range, free reserved data space first, otherwise
3306			 * it'll result in false ENOSPC error.
3307			 */
3308			btrfs_free_reserved_data_space(inode, data_reserved,
3309					cur_offset, last_byte - cur_offset);
3310		}
3311		free_extent_map(em);
3312		cur_offset = last_byte;
3313	}
3314
 
 
 
 
 
 
 
 
 
 
 
3315	/*
3316	 * If ret is still 0, means we're OK to fallocate.
3317	 * Or just cleanup the list and exit.
3318	 */
3319	list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3320		if (!ret)
3321			ret = btrfs_prealloc_file_range(inode, mode,
3322					range->start,
3323					range->len, i_blocksize(inode),
3324					offset + len, &alloc_hint);
3325		else
3326			btrfs_free_reserved_data_space(inode,
3327					data_reserved, range->start,
3328					range->len);
 
 
 
 
 
 
 
 
 
 
 
 
 
3329		list_del(&range->list);
3330		kfree(range);
3331	}
3332	if (ret < 0)
3333		goto out_unlock;
3334
3335	/*
3336	 * We didn't need to allocate any more space, but we still extended the
3337	 * size of the file so we need to update i_size and the inode item.
3338	 */
3339	ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3340out_unlock:
3341	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3342			     &cached_state);
3343out:
3344	inode_unlock(inode);
3345	/* Let go of our reservation. */
3346	if (ret != 0 && !(mode & FALLOC_FL_ZERO_RANGE))
3347		btrfs_free_reserved_data_space(inode, data_reserved,
3348				cur_offset, alloc_end - cur_offset);
3349	extent_changeset_free(data_reserved);
3350	return ret;
3351}
3352
3353static int find_desired_extent(struct inode *inode, loff_t *offset, int whence)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3354{
3355	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3356	struct extent_map *em = NULL;
 
3357	struct extent_state *cached_state = NULL;
 
 
 
 
 
 
 
3358	u64 lockstart;
3359	u64 lockend;
3360	u64 start;
3361	u64 len;
3362	int ret = 0;
3363
3364	if (inode->i_size == 0)
3365		return -ENXIO;
3366
3367	/*
3368	 * *offset can be negative, in this case we start finding DATA/HOLE from
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3369	 * the very start of the file.
3370	 */
3371	start = max_t(loff_t, 0, *offset);
3372
3373	lockstart = round_down(start, fs_info->sectorsize);
3374	lockend = round_up(i_size_read(inode),
3375			   fs_info->sectorsize);
3376	if (lockend <= lockstart)
3377		lockend = lockstart + fs_info->sectorsize;
3378	lockend--;
3379	len = lockend - lockstart + 1;
3380
3381	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3382			 &cached_state);
 
 
3383
3384	while (start < inode->i_size) {
3385		em = btrfs_get_extent_fiemap(BTRFS_I(inode), start, len);
3386		if (IS_ERR(em)) {
3387			ret = PTR_ERR(em);
3388			em = NULL;
3389			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3390		}
3391
3392		if (whence == SEEK_HOLE &&
3393		    (em->block_start == EXTENT_MAP_HOLE ||
3394		     test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3395			break;
3396		else if (whence == SEEK_DATA &&
3397			   (em->block_start != EXTENT_MAP_HOLE &&
3398			    !test_bit(EXTENT_FLAG_PREALLOC, &em->flags)))
3399			break;
3400
3401		start = em->start + em->len;
3402		free_extent_map(em);
3403		em = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3404		cond_resched();
3405	}
3406	free_extent_map(em);
3407	if (!ret) {
3408		if (whence == SEEK_DATA && start >= inode->i_size)
3409			ret = -ENXIO;
3410		else
3411			*offset = min_t(loff_t, start, inode->i_size);
 
 
3412	}
3413	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3414			     &cached_state);
3415	return ret;
 
 
 
 
 
 
 
 
 
3416}
3417
3418static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3419{
3420	struct inode *inode = file->f_mapping->host;
3421	int ret;
3422
3423	inode_lock(inode);
3424	switch (whence) {
3425	case SEEK_END:
3426	case SEEK_CUR:
3427		offset = generic_file_llseek(file, offset, whence);
3428		goto out;
3429	case SEEK_DATA:
3430	case SEEK_HOLE:
3431		if (offset >= i_size_read(inode)) {
3432			inode_unlock(inode);
3433			return -ENXIO;
3434		}
3435
3436		ret = find_desired_extent(inode, &offset, whence);
3437		if (ret) {
3438			inode_unlock(inode);
3439			return ret;
3440		}
3441	}
3442
3443	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3444out:
3445	inode_unlock(inode);
3446	return offset;
3447}
3448
3449static int btrfs_file_open(struct inode *inode, struct file *filp)
3450{
3451	filp->f_mode |= FMODE_NOWAIT;
 
 
 
 
 
 
 
3452	return generic_file_open(inode, filp);
3453}
3454
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3455const struct file_operations btrfs_file_operations = {
3456	.llseek		= btrfs_file_llseek,
3457	.read_iter      = generic_file_read_iter,
3458	.splice_read	= generic_file_splice_read,
3459	.write_iter	= btrfs_file_write_iter,
 
3460	.mmap		= btrfs_file_mmap,
3461	.open		= btrfs_file_open,
3462	.release	= btrfs_release_file,
 
3463	.fsync		= btrfs_sync_file,
3464	.fallocate	= btrfs_fallocate,
3465	.unlocked_ioctl	= btrfs_ioctl,
3466#ifdef CONFIG_COMPAT
3467	.compat_ioctl	= btrfs_compat_ioctl,
3468#endif
3469	.remap_file_range = btrfs_remap_file_range,
3470};
3471
3472void __cold btrfs_auto_defrag_exit(void)
3473{
3474	kmem_cache_destroy(btrfs_inode_defrag_cachep);
3475}
3476
3477int __init btrfs_auto_defrag_init(void)
3478{
3479	btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag",
3480					sizeof(struct inode_defrag), 0,
3481					SLAB_MEM_SPREAD,
3482					NULL);
3483	if (!btrfs_inode_defrag_cachep)
3484		return -ENOMEM;
3485
3486	return 0;
3487}
3488
3489int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3490{
3491	int ret;
3492
3493	/*
3494	 * So with compression we will find and lock a dirty page and clear the
3495	 * first one as dirty, setup an async extent, and immediately return
3496	 * with the entire range locked but with nobody actually marked with
3497	 * writeback.  So we can't just filemap_write_and_wait_range() and
3498	 * expect it to work since it will just kick off a thread to do the
3499	 * actual work.  So we need to call filemap_fdatawrite_range _again_
3500	 * since it will wait on the page lock, which won't be unlocked until
3501	 * after the pages have been marked as writeback and so we're good to go
3502	 * from there.  We have to do this otherwise we'll miss the ordered
3503	 * extents and that results in badness.  Please Josef, do not think you
3504	 * know better and pull this out at some point in the future, it is
3505	 * right and you are wrong.
3506	 */
3507	ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3508	if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3509			     &BTRFS_I(inode)->runtime_flags))
3510		ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3511
3512	return ret;
3513}
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/fs.h>
   7#include <linux/pagemap.h>
   8#include <linux/time.h>
   9#include <linux/init.h>
  10#include <linux/string.h>
  11#include <linux/backing-dev.h>
  12#include <linux/falloc.h>
  13#include <linux/writeback.h>
  14#include <linux/compat.h>
  15#include <linux/slab.h>
  16#include <linux/btrfs.h>
  17#include <linux/uio.h>
  18#include <linux/iversion.h>
  19#include <linux/fsverity.h>
  20#include <linux/iomap.h>
  21#include "ctree.h"
  22#include "disk-io.h"
  23#include "transaction.h"
  24#include "btrfs_inode.h"
 
  25#include "tree-log.h"
  26#include "locking.h"
 
  27#include "qgroup.h"
  28#include "compression.h"
  29#include "delalloc-space.h"
  30#include "reflink.h"
  31#include "subpage.h"
  32#include "fs.h"
  33#include "accessors.h"
  34#include "extent-tree.h"
  35#include "file-item.h"
  36#include "ioctl.h"
  37#include "file.h"
  38#include "super.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  39
  40/* simple helper to fault in pages and copy.  This should go away
  41 * and be replaced with calls into generic code.
  42 */
  43static noinline int btrfs_copy_from_user(loff_t pos, size_t write_bytes,
  44					 struct page **prepared_pages,
  45					 struct iov_iter *i)
  46{
  47	size_t copied = 0;
  48	size_t total_copied = 0;
  49	int pg = 0;
  50	int offset = offset_in_page(pos);
  51
  52	while (write_bytes > 0) {
  53		size_t count = min_t(size_t,
  54				     PAGE_SIZE - offset, write_bytes);
  55		struct page *page = prepared_pages[pg];
  56		/*
  57		 * Copy data from userspace to the current page
  58		 */
  59		copied = copy_page_from_iter_atomic(page, offset, count, i);
  60
  61		/* Flush processor's dcache for this page */
  62		flush_dcache_page(page);
  63
  64		/*
  65		 * if we get a partial write, we can end up with
  66		 * partially up to date pages.  These add
  67		 * a lot of complexity, so make sure they don't
  68		 * happen by forcing this copy to be retried.
  69		 *
  70		 * The rest of the btrfs_file_write code will fall
  71		 * back to page at a time copies after we return 0.
  72		 */
  73		if (unlikely(copied < count)) {
  74			if (!PageUptodate(page)) {
  75				iov_iter_revert(i, copied);
  76				copied = 0;
  77			}
  78			if (!copied)
  79				break;
  80		}
  81
 
  82		write_bytes -= copied;
  83		total_copied += copied;
  84		offset += copied;
  85		if (offset == PAGE_SIZE) {
 
 
 
 
 
 
  86			pg++;
  87			offset = 0;
  88		}
  89	}
  90	return total_copied;
  91}
  92
  93/*
  94 * unlocks pages after btrfs_file_write is done with them
  95 */
  96static void btrfs_drop_pages(struct btrfs_fs_info *fs_info,
  97			     struct page **pages, size_t num_pages,
  98			     u64 pos, u64 copied)
  99{
 100	size_t i;
 101	u64 block_start = round_down(pos, fs_info->sectorsize);
 102	u64 block_len = round_up(pos + copied, fs_info->sectorsize) - block_start;
 103
 104	ASSERT(block_len <= U32_MAX);
 105	for (i = 0; i < num_pages; i++) {
 106		/* page checked is some magic around finding pages that
 107		 * have been modified without going through btrfs_set_page_dirty
 108		 * clear it here. There should be no need to mark the pages
 109		 * accessed as prepare_pages should have marked them accessed
 110		 * in prepare_pages via find_or_create_page()
 111		 */
 112		btrfs_folio_clamp_clear_checked(fs_info, page_folio(pages[i]),
 113						block_start, block_len);
 114		unlock_page(pages[i]);
 115		put_page(pages[i]);
 116	}
 117}
 118
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 119/*
 120 * After btrfs_copy_from_user(), update the following things for delalloc:
 121 * - Mark newly dirtied pages as DELALLOC in the io tree.
 122 *   Used to advise which range is to be written back.
 123 * - Mark modified pages as Uptodate/Dirty and not needing COW fixup
 124 * - Update inode size for past EOF write
 
 125 */
 126int btrfs_dirty_pages(struct btrfs_inode *inode, struct page **pages,
 127		      size_t num_pages, loff_t pos, size_t write_bytes,
 128		      struct extent_state **cached, bool noreserve)
 129{
 130	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 131	int err = 0;
 132	int i;
 133	u64 num_bytes;
 134	u64 start_pos;
 135	u64 end_of_last_block;
 136	u64 end_pos = pos + write_bytes;
 137	loff_t isize = i_size_read(&inode->vfs_inode);
 138	unsigned int extra_bits = 0;
 139
 140	if (write_bytes == 0)
 141		return 0;
 142
 143	if (noreserve)
 144		extra_bits |= EXTENT_NORESERVE;
 145
 146	start_pos = round_down(pos, fs_info->sectorsize);
 147	num_bytes = round_up(write_bytes + pos - start_pos,
 148			     fs_info->sectorsize);
 149	ASSERT(num_bytes <= U32_MAX);
 150
 151	end_of_last_block = start_pos + num_bytes - 1;
 152
 153	/*
 154	 * The pages may have already been dirty, clear out old accounting so
 155	 * we can set things up properly
 156	 */
 157	clear_extent_bit(&inode->io_tree, start_pos, end_of_last_block,
 158			 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
 159			 cached);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 160
 161	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 162					extra_bits, cached);
 163	if (err)
 164		return err;
 165
 166	for (i = 0; i < num_pages; i++) {
 167		struct page *p = pages[i];
 168
 169		btrfs_folio_clamp_set_uptodate(fs_info, page_folio(p),
 170					       start_pos, num_bytes);
 171		btrfs_folio_clamp_clear_checked(fs_info, page_folio(p),
 172						start_pos, num_bytes);
 173		btrfs_folio_clamp_set_dirty(fs_info, page_folio(p),
 174					    start_pos, num_bytes);
 175	}
 176
 177	/*
 178	 * we've only changed i_size in ram, and we haven't updated
 179	 * the disk i_size.  There is no need to log the inode
 180	 * at this time.
 181	 */
 182	if (end_pos > isize)
 183		i_size_write(&inode->vfs_inode, end_pos);
 184	return 0;
 185}
 186
 187/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 188 * this is very complex, but the basic idea is to drop all extents
 189 * in the range start - end.  hint_block is filled in with a block number
 190 * that would be a good hint to the block allocator for this file.
 191 *
 192 * If an extent intersects the range but is not entirely inside the range
 193 * it is either truncated or split.  Anything entirely inside the range
 194 * is deleted from the tree.
 195 *
 196 * Note: the VFS' inode number of bytes is not updated, it's up to the caller
 197 * to deal with that. We set the field 'bytes_found' of the arguments structure
 198 * with the number of allocated bytes found in the target range, so that the
 199 * caller can update the inode's number of bytes in an atomic way when
 200 * replacing extents in a range to avoid races with stat(2).
 201 */
 202int btrfs_drop_extents(struct btrfs_trans_handle *trans,
 203		       struct btrfs_root *root, struct btrfs_inode *inode,
 204		       struct btrfs_drop_extents_args *args)
 
 
 
 
 205{
 206	struct btrfs_fs_info *fs_info = root->fs_info;
 207	struct extent_buffer *leaf;
 208	struct btrfs_file_extent_item *fi;
 209	struct btrfs_ref ref = { 0 };
 210	struct btrfs_key key;
 211	struct btrfs_key new_key;
 212	u64 ino = btrfs_ino(inode);
 213	u64 search_start = args->start;
 214	u64 disk_bytenr = 0;
 215	u64 num_bytes = 0;
 216	u64 extent_offset = 0;
 217	u64 extent_end = 0;
 218	u64 last_end = args->start;
 219	int del_nr = 0;
 220	int del_slot = 0;
 221	int extent_type;
 222	int recow;
 223	int ret;
 224	int modify_tree = -1;
 225	int update_refs;
 226	int found = 0;
 227	struct btrfs_path *path = args->path;
 228
 229	args->bytes_found = 0;
 230	args->extent_inserted = false;
 231
 232	/* Must always have a path if ->replace_extent is true */
 233	ASSERT(!(args->replace_extent && !args->path));
 234
 235	if (!path) {
 236		path = btrfs_alloc_path();
 237		if (!path) {
 238			ret = -ENOMEM;
 239			goto out;
 240		}
 241	}
 242
 243	if (args->drop_cache)
 244		btrfs_drop_extent_map_range(inode, args->start, args->end - 1, false);
 245
 246	if (args->start >= inode->disk_i_size && !args->replace_extent)
 247		modify_tree = 0;
 248
 249	update_refs = (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID);
 
 250	while (1) {
 251		recow = 0;
 252		ret = btrfs_lookup_file_extent(trans, root, path, ino,
 253					       search_start, modify_tree);
 254		if (ret < 0)
 255			break;
 256		if (ret > 0 && path->slots[0] > 0 && search_start == args->start) {
 257			leaf = path->nodes[0];
 258			btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 259			if (key.objectid == ino &&
 260			    key.type == BTRFS_EXTENT_DATA_KEY)
 261				path->slots[0]--;
 262		}
 263		ret = 0;
 
 264next_slot:
 265		leaf = path->nodes[0];
 266		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 267			BUG_ON(del_nr > 0);
 268			ret = btrfs_next_leaf(root, path);
 269			if (ret < 0)
 270				break;
 271			if (ret > 0) {
 272				ret = 0;
 273				break;
 274			}
 
 275			leaf = path->nodes[0];
 276			recow = 1;
 277		}
 278
 279		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 280
 281		if (key.objectid > ino)
 282			break;
 283		if (WARN_ON_ONCE(key.objectid < ino) ||
 284		    key.type < BTRFS_EXTENT_DATA_KEY) {
 285			ASSERT(del_nr == 0);
 286			path->slots[0]++;
 287			goto next_slot;
 288		}
 289		if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= args->end)
 290			break;
 291
 292		fi = btrfs_item_ptr(leaf, path->slots[0],
 293				    struct btrfs_file_extent_item);
 294		extent_type = btrfs_file_extent_type(leaf, fi);
 295
 296		if (extent_type == BTRFS_FILE_EXTENT_REG ||
 297		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 298			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 299			num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 300			extent_offset = btrfs_file_extent_offset(leaf, fi);
 301			extent_end = key.offset +
 302				btrfs_file_extent_num_bytes(leaf, fi);
 303		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 304			extent_end = key.offset +
 305				btrfs_file_extent_ram_bytes(leaf, fi);
 306		} else {
 307			/* can't happen */
 308			BUG();
 309		}
 310
 311		/*
 312		 * Don't skip extent items representing 0 byte lengths. They
 313		 * used to be created (bug) if while punching holes we hit
 314		 * -ENOSPC condition. So if we find one here, just ensure we
 315		 * delete it, otherwise we would insert a new file extent item
 316		 * with the same key (offset) as that 0 bytes length file
 317		 * extent item in the call to setup_items_for_insert() later
 318		 * in this function.
 319		 */
 320		if (extent_end == key.offset && extent_end >= search_start) {
 321			last_end = extent_end;
 322			goto delete_extent_item;
 323		}
 324
 325		if (extent_end <= search_start) {
 326			path->slots[0]++;
 327			goto next_slot;
 328		}
 329
 330		found = 1;
 331		search_start = max(key.offset, args->start);
 332		if (recow || !modify_tree) {
 333			modify_tree = -1;
 334			btrfs_release_path(path);
 335			continue;
 336		}
 337
 338		/*
 339		 *     | - range to drop - |
 340		 *  | -------- extent -------- |
 341		 */
 342		if (args->start > key.offset && args->end < extent_end) {
 343			BUG_ON(del_nr > 0);
 344			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 345				ret = -EOPNOTSUPP;
 346				break;
 347			}
 348
 349			memcpy(&new_key, &key, sizeof(new_key));
 350			new_key.offset = args->start;
 351			ret = btrfs_duplicate_item(trans, root, path,
 352						   &new_key);
 353			if (ret == -EAGAIN) {
 354				btrfs_release_path(path);
 355				continue;
 356			}
 357			if (ret < 0)
 358				break;
 359
 360			leaf = path->nodes[0];
 361			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 362					    struct btrfs_file_extent_item);
 363			btrfs_set_file_extent_num_bytes(leaf, fi,
 364							args->start - key.offset);
 365
 366			fi = btrfs_item_ptr(leaf, path->slots[0],
 367					    struct btrfs_file_extent_item);
 368
 369			extent_offset += args->start - key.offset;
 370			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 371			btrfs_set_file_extent_num_bytes(leaf, fi,
 372							extent_end - args->start);
 373			btrfs_mark_buffer_dirty(trans, leaf);
 374
 375			if (update_refs && disk_bytenr > 0) {
 376				btrfs_init_generic_ref(&ref,
 377						BTRFS_ADD_DELAYED_REF,
 378						disk_bytenr, num_bytes, 0,
 379						root->root_key.objectid);
 380				btrfs_init_data_ref(&ref,
 381						root->root_key.objectid,
 382						new_key.objectid,
 383						args->start - extent_offset,
 384						0, false);
 385				ret = btrfs_inc_extent_ref(trans, &ref);
 386				if (ret) {
 387					btrfs_abort_transaction(trans, ret);
 388					break;
 389				}
 390			}
 391			key.offset = args->start;
 392		}
 393		/*
 394		 * From here on out we will have actually dropped something, so
 395		 * last_end can be updated.
 396		 */
 397		last_end = extent_end;
 398
 399		/*
 400		 *  | ---- range to drop ----- |
 401		 *      | -------- extent -------- |
 402		 */
 403		if (args->start <= key.offset && args->end < extent_end) {
 404			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 405				ret = -EOPNOTSUPP;
 406				break;
 407			}
 408
 409			memcpy(&new_key, &key, sizeof(new_key));
 410			new_key.offset = args->end;
 411			btrfs_set_item_key_safe(trans, path, &new_key);
 412
 413			extent_offset += args->end - key.offset;
 414			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 415			btrfs_set_file_extent_num_bytes(leaf, fi,
 416							extent_end - args->end);
 417			btrfs_mark_buffer_dirty(trans, leaf);
 418			if (update_refs && disk_bytenr > 0)
 419				args->bytes_found += args->end - key.offset;
 420			break;
 421		}
 422
 423		search_start = extent_end;
 424		/*
 425		 *       | ---- range to drop ----- |
 426		 *  | -------- extent -------- |
 427		 */
 428		if (args->start > key.offset && args->end >= extent_end) {
 429			BUG_ON(del_nr > 0);
 430			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 431				ret = -EOPNOTSUPP;
 432				break;
 433			}
 434
 435			btrfs_set_file_extent_num_bytes(leaf, fi,
 436							args->start - key.offset);
 437			btrfs_mark_buffer_dirty(trans, leaf);
 438			if (update_refs && disk_bytenr > 0)
 439				args->bytes_found += extent_end - args->start;
 440			if (args->end == extent_end)
 441				break;
 442
 443			path->slots[0]++;
 444			goto next_slot;
 445		}
 446
 447		/*
 448		 *  | ---- range to drop ----- |
 449		 *    | ------ extent ------ |
 450		 */
 451		if (args->start <= key.offset && args->end >= extent_end) {
 452delete_extent_item:
 453			if (del_nr == 0) {
 454				del_slot = path->slots[0];
 455				del_nr = 1;
 456			} else {
 457				BUG_ON(del_slot + del_nr != path->slots[0]);
 458				del_nr++;
 459			}
 460
 461			if (update_refs &&
 462			    extent_type == BTRFS_FILE_EXTENT_INLINE) {
 463				args->bytes_found += extent_end - key.offset;
 
 464				extent_end = ALIGN(extent_end,
 465						   fs_info->sectorsize);
 466			} else if (update_refs && disk_bytenr > 0) {
 467				btrfs_init_generic_ref(&ref,
 468						BTRFS_DROP_DELAYED_REF,
 469						disk_bytenr, num_bytes, 0,
 470						root->root_key.objectid);
 471				btrfs_init_data_ref(&ref,
 472						root->root_key.objectid,
 473						key.objectid,
 474						key.offset - extent_offset, 0,
 475						false);
 476				ret = btrfs_free_extent(trans, &ref);
 477				if (ret) {
 478					btrfs_abort_transaction(trans, ret);
 479					break;
 480				}
 481				args->bytes_found += extent_end - key.offset;
 482			}
 483
 484			if (args->end == extent_end)
 485				break;
 486
 487			if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 488				path->slots[0]++;
 489				goto next_slot;
 490			}
 491
 492			ret = btrfs_del_items(trans, root, path, del_slot,
 493					      del_nr);
 494			if (ret) {
 495				btrfs_abort_transaction(trans, ret);
 496				break;
 497			}
 498
 499			del_nr = 0;
 500			del_slot = 0;
 501
 502			btrfs_release_path(path);
 503			continue;
 504		}
 505
 506		BUG();
 507	}
 508
 509	if (!ret && del_nr > 0) {
 510		/*
 511		 * Set path->slots[0] to first slot, so that after the delete
 512		 * if items are move off from our leaf to its immediate left or
 513		 * right neighbor leafs, we end up with a correct and adjusted
 514		 * path->slots[0] for our insertion (if args->replace_extent).
 515		 */
 516		path->slots[0] = del_slot;
 517		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 518		if (ret)
 519			btrfs_abort_transaction(trans, ret);
 520	}
 521
 522	leaf = path->nodes[0];
 523	/*
 524	 * If btrfs_del_items() was called, it might have deleted a leaf, in
 525	 * which case it unlocked our path, so check path->locks[0] matches a
 526	 * write lock.
 527	 */
 528	if (!ret && args->replace_extent &&
 529	    path->locks[0] == BTRFS_WRITE_LOCK &&
 
 530	    btrfs_leaf_free_space(leaf) >=
 531	    sizeof(struct btrfs_item) + args->extent_item_size) {
 532
 533		key.objectid = ino;
 534		key.type = BTRFS_EXTENT_DATA_KEY;
 535		key.offset = args->start;
 536		if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) {
 537			struct btrfs_key slot_key;
 538
 539			btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]);
 540			if (btrfs_comp_cpu_keys(&key, &slot_key) > 0)
 541				path->slots[0]++;
 542		}
 543		btrfs_setup_item_for_insert(trans, root, path, &key,
 544					    args->extent_item_size);
 545		args->extent_inserted = true;
 
 
 
 546	}
 547
 548	if (!args->path)
 549		btrfs_free_path(path);
 550	else if (!args->extent_inserted)
 551		btrfs_release_path(path);
 552out:
 553	args->drop_end = found ? min(args->end, last_end) : args->end;
 
 
 
 
 
 
 
 
 
 554
 
 
 
 
 
 
 555	return ret;
 556}
 557
 558static int extent_mergeable(struct extent_buffer *leaf, int slot,
 559			    u64 objectid, u64 bytenr, u64 orig_offset,
 560			    u64 *start, u64 *end)
 561{
 562	struct btrfs_file_extent_item *fi;
 563	struct btrfs_key key;
 564	u64 extent_end;
 565
 566	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 567		return 0;
 568
 569	btrfs_item_key_to_cpu(leaf, &key, slot);
 570	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 571		return 0;
 572
 573	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 574	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 575	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 576	    btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 577	    btrfs_file_extent_compression(leaf, fi) ||
 578	    btrfs_file_extent_encryption(leaf, fi) ||
 579	    btrfs_file_extent_other_encoding(leaf, fi))
 580		return 0;
 581
 582	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 583	if ((*start && *start != key.offset) || (*end && *end != extent_end))
 584		return 0;
 585
 586	*start = key.offset;
 587	*end = extent_end;
 588	return 1;
 589}
 590
 591/*
 592 * Mark extent in the range start - end as written.
 593 *
 594 * This changes extent type from 'pre-allocated' to 'regular'. If only
 595 * part of extent is marked as written, the extent will be split into
 596 * two or three.
 597 */
 598int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 599			      struct btrfs_inode *inode, u64 start, u64 end)
 600{
 
 601	struct btrfs_root *root = inode->root;
 602	struct extent_buffer *leaf;
 603	struct btrfs_path *path;
 604	struct btrfs_file_extent_item *fi;
 605	struct btrfs_ref ref = { 0 };
 606	struct btrfs_key key;
 607	struct btrfs_key new_key;
 608	u64 bytenr;
 609	u64 num_bytes;
 610	u64 extent_end;
 611	u64 orig_offset;
 612	u64 other_start;
 613	u64 other_end;
 614	u64 split;
 615	int del_nr = 0;
 616	int del_slot = 0;
 617	int recow;
 618	int ret = 0;
 619	u64 ino = btrfs_ino(inode);
 620
 621	path = btrfs_alloc_path();
 622	if (!path)
 623		return -ENOMEM;
 624again:
 625	recow = 0;
 626	split = start;
 627	key.objectid = ino;
 628	key.type = BTRFS_EXTENT_DATA_KEY;
 629	key.offset = split;
 630
 631	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 632	if (ret < 0)
 633		goto out;
 634	if (ret > 0 && path->slots[0] > 0)
 635		path->slots[0]--;
 636
 637	leaf = path->nodes[0];
 638	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 639	if (key.objectid != ino ||
 640	    key.type != BTRFS_EXTENT_DATA_KEY) {
 641		ret = -EINVAL;
 642		btrfs_abort_transaction(trans, ret);
 643		goto out;
 644	}
 645	fi = btrfs_item_ptr(leaf, path->slots[0],
 646			    struct btrfs_file_extent_item);
 647	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_PREALLOC) {
 648		ret = -EINVAL;
 649		btrfs_abort_transaction(trans, ret);
 650		goto out;
 651	}
 652	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 653	if (key.offset > start || extent_end < end) {
 654		ret = -EINVAL;
 655		btrfs_abort_transaction(trans, ret);
 656		goto out;
 657	}
 658
 659	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 660	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 661	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 662	memcpy(&new_key, &key, sizeof(new_key));
 663
 664	if (start == key.offset && end < extent_end) {
 665		other_start = 0;
 666		other_end = start;
 667		if (extent_mergeable(leaf, path->slots[0] - 1,
 668				     ino, bytenr, orig_offset,
 669				     &other_start, &other_end)) {
 670			new_key.offset = end;
 671			btrfs_set_item_key_safe(trans, path, &new_key);
 672			fi = btrfs_item_ptr(leaf, path->slots[0],
 673					    struct btrfs_file_extent_item);
 674			btrfs_set_file_extent_generation(leaf, fi,
 675							 trans->transid);
 676			btrfs_set_file_extent_num_bytes(leaf, fi,
 677							extent_end - end);
 678			btrfs_set_file_extent_offset(leaf, fi,
 679						     end - orig_offset);
 680			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 681					    struct btrfs_file_extent_item);
 682			btrfs_set_file_extent_generation(leaf, fi,
 683							 trans->transid);
 684			btrfs_set_file_extent_num_bytes(leaf, fi,
 685							end - other_start);
 686			btrfs_mark_buffer_dirty(trans, leaf);
 687			goto out;
 688		}
 689	}
 690
 691	if (start > key.offset && end == extent_end) {
 692		other_start = end;
 693		other_end = 0;
 694		if (extent_mergeable(leaf, path->slots[0] + 1,
 695				     ino, bytenr, orig_offset,
 696				     &other_start, &other_end)) {
 697			fi = btrfs_item_ptr(leaf, path->slots[0],
 698					    struct btrfs_file_extent_item);
 699			btrfs_set_file_extent_num_bytes(leaf, fi,
 700							start - key.offset);
 701			btrfs_set_file_extent_generation(leaf, fi,
 702							 trans->transid);
 703			path->slots[0]++;
 704			new_key.offset = start;
 705			btrfs_set_item_key_safe(trans, path, &new_key);
 706
 707			fi = btrfs_item_ptr(leaf, path->slots[0],
 708					    struct btrfs_file_extent_item);
 709			btrfs_set_file_extent_generation(leaf, fi,
 710							 trans->transid);
 711			btrfs_set_file_extent_num_bytes(leaf, fi,
 712							other_end - start);
 713			btrfs_set_file_extent_offset(leaf, fi,
 714						     start - orig_offset);
 715			btrfs_mark_buffer_dirty(trans, leaf);
 716			goto out;
 717		}
 718	}
 719
 720	while (start > key.offset || end < extent_end) {
 721		if (key.offset == start)
 722			split = end;
 723
 724		new_key.offset = split;
 725		ret = btrfs_duplicate_item(trans, root, path, &new_key);
 726		if (ret == -EAGAIN) {
 727			btrfs_release_path(path);
 728			goto again;
 729		}
 730		if (ret < 0) {
 731			btrfs_abort_transaction(trans, ret);
 732			goto out;
 733		}
 734
 735		leaf = path->nodes[0];
 736		fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 737				    struct btrfs_file_extent_item);
 738		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 739		btrfs_set_file_extent_num_bytes(leaf, fi,
 740						split - key.offset);
 741
 742		fi = btrfs_item_ptr(leaf, path->slots[0],
 743				    struct btrfs_file_extent_item);
 744
 745		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 746		btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 747		btrfs_set_file_extent_num_bytes(leaf, fi,
 748						extent_end - split);
 749		btrfs_mark_buffer_dirty(trans, leaf);
 750
 751		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, bytenr,
 752				       num_bytes, 0, root->root_key.objectid);
 753		btrfs_init_data_ref(&ref, root->root_key.objectid, ino,
 754				    orig_offset, 0, false);
 755		ret = btrfs_inc_extent_ref(trans, &ref);
 756		if (ret) {
 757			btrfs_abort_transaction(trans, ret);
 758			goto out;
 759		}
 760
 761		if (split == start) {
 762			key.offset = start;
 763		} else {
 764			if (start != key.offset) {
 765				ret = -EINVAL;
 766				btrfs_abort_transaction(trans, ret);
 767				goto out;
 768			}
 769			path->slots[0]--;
 770			extent_end = end;
 771		}
 772		recow = 1;
 773	}
 774
 775	other_start = end;
 776	other_end = 0;
 777	btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
 778			       num_bytes, 0, root->root_key.objectid);
 779	btrfs_init_data_ref(&ref, root->root_key.objectid, ino, orig_offset,
 780			    0, false);
 781	if (extent_mergeable(leaf, path->slots[0] + 1,
 782			     ino, bytenr, orig_offset,
 783			     &other_start, &other_end)) {
 784		if (recow) {
 785			btrfs_release_path(path);
 786			goto again;
 787		}
 788		extent_end = other_end;
 789		del_slot = path->slots[0] + 1;
 790		del_nr++;
 791		ret = btrfs_free_extent(trans, &ref);
 792		if (ret) {
 793			btrfs_abort_transaction(trans, ret);
 794			goto out;
 795		}
 796	}
 797	other_start = 0;
 798	other_end = start;
 799	if (extent_mergeable(leaf, path->slots[0] - 1,
 800			     ino, bytenr, orig_offset,
 801			     &other_start, &other_end)) {
 802		if (recow) {
 803			btrfs_release_path(path);
 804			goto again;
 805		}
 806		key.offset = other_start;
 807		del_slot = path->slots[0];
 808		del_nr++;
 809		ret = btrfs_free_extent(trans, &ref);
 810		if (ret) {
 811			btrfs_abort_transaction(trans, ret);
 812			goto out;
 813		}
 814	}
 815	if (del_nr == 0) {
 816		fi = btrfs_item_ptr(leaf, path->slots[0],
 817			   struct btrfs_file_extent_item);
 818		btrfs_set_file_extent_type(leaf, fi,
 819					   BTRFS_FILE_EXTENT_REG);
 820		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 821		btrfs_mark_buffer_dirty(trans, leaf);
 822	} else {
 823		fi = btrfs_item_ptr(leaf, del_slot - 1,
 824			   struct btrfs_file_extent_item);
 825		btrfs_set_file_extent_type(leaf, fi,
 826					   BTRFS_FILE_EXTENT_REG);
 827		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
 828		btrfs_set_file_extent_num_bytes(leaf, fi,
 829						extent_end - key.offset);
 830		btrfs_mark_buffer_dirty(trans, leaf);
 831
 832		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 833		if (ret < 0) {
 834			btrfs_abort_transaction(trans, ret);
 835			goto out;
 836		}
 837	}
 838out:
 839	btrfs_free_path(path);
 840	return ret;
 841}
 842
 843/*
 844 * on error we return an unlocked page and the error value
 845 * on success we return a locked page and 0
 846 */
 847static int prepare_uptodate_page(struct inode *inode,
 848				 struct page *page, u64 pos,
 849				 bool force_uptodate)
 850{
 851	struct folio *folio = page_folio(page);
 852	int ret = 0;
 853
 854	if (((pos & (PAGE_SIZE - 1)) || force_uptodate) &&
 855	    !PageUptodate(page)) {
 856		ret = btrfs_read_folio(NULL, folio);
 857		if (ret)
 858			return ret;
 859		lock_page(page);
 860		if (!PageUptodate(page)) {
 861			unlock_page(page);
 862			return -EIO;
 863		}
 864
 865		/*
 866		 * Since btrfs_read_folio() will unlock the folio before it
 867		 * returns, there is a window where btrfs_release_folio() can be
 868		 * called to release the page.  Here we check both inode
 869		 * mapping and PagePrivate() to make sure the page was not
 870		 * released.
 871		 *
 872		 * The private flag check is essential for subpage as we need
 873		 * to store extra bitmap using folio private.
 874		 */
 875		if (page->mapping != inode->i_mapping || !folio_test_private(folio)) {
 876			unlock_page(page);
 877			return -EAGAIN;
 878		}
 879	}
 880	return 0;
 881}
 882
 883static fgf_t get_prepare_fgp_flags(bool nowait)
 884{
 885	fgf_t fgp_flags = FGP_LOCK | FGP_ACCESSED | FGP_CREAT;
 886
 887	if (nowait)
 888		fgp_flags |= FGP_NOWAIT;
 889
 890	return fgp_flags;
 891}
 892
 893static gfp_t get_prepare_gfp_flags(struct inode *inode, bool nowait)
 894{
 895	gfp_t gfp;
 896
 897	gfp = btrfs_alloc_write_mask(inode->i_mapping);
 898	if (nowait) {
 899		gfp &= ~__GFP_DIRECT_RECLAIM;
 900		gfp |= GFP_NOWAIT;
 901	}
 902
 903	return gfp;
 904}
 905
 906/*
 907 * this just gets pages into the page cache and locks them down.
 908 */
 909static noinline int prepare_pages(struct inode *inode, struct page **pages,
 910				  size_t num_pages, loff_t pos,
 911				  size_t write_bytes, bool force_uptodate,
 912				  bool nowait)
 913{
 914	int i;
 915	unsigned long index = pos >> PAGE_SHIFT;
 916	gfp_t mask = get_prepare_gfp_flags(inode, nowait);
 917	fgf_t fgp_flags = get_prepare_fgp_flags(nowait);
 918	int err = 0;
 919	int faili;
 920
 921	for (i = 0; i < num_pages; i++) {
 922again:
 923		pages[i] = pagecache_get_page(inode->i_mapping, index + i,
 924					      fgp_flags, mask | __GFP_WRITE);
 925		if (!pages[i]) {
 926			faili = i - 1;
 927			if (nowait)
 928				err = -EAGAIN;
 929			else
 930				err = -ENOMEM;
 931			goto fail;
 932		}
 933
 934		err = set_page_extent_mapped(pages[i]);
 935		if (err < 0) {
 936			faili = i;
 937			goto fail;
 938		}
 939
 940		if (i == 0)
 941			err = prepare_uptodate_page(inode, pages[i], pos,
 942						    force_uptodate);
 943		if (!err && i == num_pages - 1)
 944			err = prepare_uptodate_page(inode, pages[i],
 945						    pos + write_bytes, false);
 946		if (err) {
 947			put_page(pages[i]);
 948			if (!nowait && err == -EAGAIN) {
 949				err = 0;
 950				goto again;
 951			}
 952			faili = i - 1;
 953			goto fail;
 954		}
 955		wait_on_page_writeback(pages[i]);
 956	}
 957
 958	return 0;
 959fail:
 960	while (faili >= 0) {
 961		unlock_page(pages[faili]);
 962		put_page(pages[faili]);
 963		faili--;
 964	}
 965	return err;
 966
 967}
 968
 969/*
 970 * This function locks the extent and properly waits for data=ordered extents
 971 * to finish before allowing the pages to be modified if need.
 972 *
 973 * The return value:
 974 * 1 - the extent is locked
 975 * 0 - the extent is not locked, and everything is OK
 976 * -EAGAIN - need re-prepare the pages
 977 * the other < 0 number - Something wrong happens
 978 */
 979static noinline int
 980lock_and_cleanup_extent_if_need(struct btrfs_inode *inode, struct page **pages,
 981				size_t num_pages, loff_t pos,
 982				size_t write_bytes,
 983				u64 *lockstart, u64 *lockend, bool nowait,
 984				struct extent_state **cached_state)
 985{
 986	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 987	u64 start_pos;
 988	u64 last_pos;
 989	int i;
 990	int ret = 0;
 991
 992	start_pos = round_down(pos, fs_info->sectorsize);
 993	last_pos = round_up(pos + write_bytes, fs_info->sectorsize) - 1;
 
 
 994
 995	if (start_pos < inode->vfs_inode.i_size) {
 996		struct btrfs_ordered_extent *ordered;
 997
 998		if (nowait) {
 999			if (!try_lock_extent(&inode->io_tree, start_pos, last_pos,
1000					     cached_state)) {
1001				for (i = 0; i < num_pages; i++) {
1002					unlock_page(pages[i]);
1003					put_page(pages[i]);
1004					pages[i] = NULL;
1005				}
1006
1007				return -EAGAIN;
1008			}
1009		} else {
1010			lock_extent(&inode->io_tree, start_pos, last_pos, cached_state);
1011		}
1012
1013		ordered = btrfs_lookup_ordered_range(inode, start_pos,
1014						     last_pos - start_pos + 1);
1015		if (ordered &&
1016		    ordered->file_offset + ordered->num_bytes > start_pos &&
1017		    ordered->file_offset <= last_pos) {
1018			unlock_extent(&inode->io_tree, start_pos, last_pos,
1019				      cached_state);
1020			for (i = 0; i < num_pages; i++) {
1021				unlock_page(pages[i]);
1022				put_page(pages[i]);
1023			}
1024			btrfs_start_ordered_extent(ordered);
 
1025			btrfs_put_ordered_extent(ordered);
1026			return -EAGAIN;
1027		}
1028		if (ordered)
1029			btrfs_put_ordered_extent(ordered);
1030
1031		*lockstart = start_pos;
1032		*lockend = last_pos;
1033		ret = 1;
1034	}
1035
1036	/*
1037	 * We should be called after prepare_pages() which should have locked
1038	 * all pages in the range.
 
 
 
 
 
 
 
 
 
 
1039	 */
1040	for (i = 0; i < num_pages; i++)
 
1041		WARN_ON(!PageLocked(pages[i]));
 
1042
1043	return ret;
1044}
1045
1046/*
1047 * Check if we can do nocow write into the range [@pos, @pos + @write_bytes)
1048 *
1049 * @pos:         File offset.
1050 * @write_bytes: The length to write, will be updated to the nocow writeable
1051 *               range.
1052 *
1053 * This function will flush ordered extents in the range to ensure proper
1054 * nocow checks.
1055 *
1056 * Return:
1057 * > 0          If we can nocow, and updates @write_bytes.
1058 *  0           If we can't do a nocow write.
1059 * -EAGAIN      If we can't do a nocow write because snapshoting of the inode's
1060 *              root is in progress.
1061 * < 0          If an error happened.
1062 *
1063 * NOTE: Callers need to call btrfs_check_nocow_unlock() if we return > 0.
1064 */
1065int btrfs_check_nocow_lock(struct btrfs_inode *inode, loff_t pos,
1066			   size_t *write_bytes, bool nowait)
1067{
1068	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1069	struct btrfs_root *root = inode->root;
1070	struct extent_state *cached_state = NULL;
1071	u64 lockstart, lockend;
1072	u64 num_bytes;
1073	int ret;
1074
1075	if (!(inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1076		return 0;
1077
1078	if (!btrfs_drew_try_write_lock(&root->snapshot_lock))
1079		return -EAGAIN;
1080
1081	lockstart = round_down(pos, fs_info->sectorsize);
1082	lockend = round_up(pos + *write_bytes,
1083			   fs_info->sectorsize) - 1;
 
 
 
 
1084	num_bytes = lockend - lockstart + 1;
1085
1086	if (nowait) {
1087		if (!btrfs_try_lock_ordered_range(inode, lockstart, lockend,
1088						  &cached_state)) {
1089			btrfs_drew_write_unlock(&root->snapshot_lock);
1090			return -EAGAIN;
1091		}
1092	} else {
1093		btrfs_lock_and_flush_ordered_range(inode, lockstart, lockend,
1094						   &cached_state);
1095	}
1096	ret = can_nocow_extent(&inode->vfs_inode, lockstart, &num_bytes,
1097			NULL, NULL, NULL, nowait, false);
1098	if (ret <= 0)
1099		btrfs_drew_write_unlock(&root->snapshot_lock);
1100	else
1101		*write_bytes = min_t(size_t, *write_bytes ,
1102				     num_bytes - pos + lockstart);
1103	unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
 
 
1104
1105	return ret;
1106}
1107
1108void btrfs_check_nocow_unlock(struct btrfs_inode *inode)
1109{
1110	btrfs_drew_write_unlock(&inode->root->snapshot_lock);
1111}
1112
1113static void update_time_for_write(struct inode *inode)
1114{
1115	struct timespec64 now, ts;
1116
1117	if (IS_NOCMTIME(inode))
1118		return;
1119
1120	now = current_time(inode);
1121	ts = inode_get_mtime(inode);
1122	if (!timespec64_equal(&ts, &now))
1123		inode_set_mtime_to_ts(inode, now);
1124
1125	ts = inode_get_ctime(inode);
1126	if (!timespec64_equal(&ts, &now))
1127		inode_set_ctime_to_ts(inode, now);
1128
1129	if (IS_I_VERSION(inode))
1130		inode_inc_iversion(inode);
1131}
1132
1133static int btrfs_write_check(struct kiocb *iocb, struct iov_iter *from,
1134			     size_t count)
1135{
1136	struct file *file = iocb->ki_filp;
1137	struct inode *inode = file_inode(file);
1138	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1139	loff_t pos = iocb->ki_pos;
1140	int ret;
1141	loff_t oldsize;
1142	loff_t start_pos;
1143
1144	/*
1145	 * Quickly bail out on NOWAIT writes if we don't have the nodatacow or
1146	 * prealloc flags, as without those flags we always have to COW. We will
1147	 * later check if we can really COW into the target range (using
1148	 * can_nocow_extent() at btrfs_get_blocks_direct_write()).
1149	 */
1150	if ((iocb->ki_flags & IOCB_NOWAIT) &&
1151	    !(BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
1152		return -EAGAIN;
1153
1154	ret = file_remove_privs(file);
1155	if (ret)
1156		return ret;
1157
1158	/*
1159	 * We reserve space for updating the inode when we reserve space for the
1160	 * extent we are going to write, so we will enospc out there.  We don't
1161	 * need to start yet another transaction to update the inode as we will
1162	 * update the inode when we finish writing whatever data we write.
1163	 */
1164	update_time_for_write(inode);
1165
1166	start_pos = round_down(pos, fs_info->sectorsize);
1167	oldsize = i_size_read(inode);
1168	if (start_pos > oldsize) {
1169		/* Expand hole size to cover write data, preventing empty gap */
1170		loff_t end_pos = round_up(pos + count, fs_info->sectorsize);
1171
1172		ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, end_pos);
1173		if (ret)
1174			return ret;
1175	}
1176
1177	return 0;
1178}
1179
1180static noinline ssize_t btrfs_buffered_write(struct kiocb *iocb,
1181					       struct iov_iter *i)
1182{
1183	struct file *file = iocb->ki_filp;
1184	loff_t pos;
1185	struct inode *inode = file_inode(file);
1186	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
 
1187	struct page **pages = NULL;
1188	struct extent_changeset *data_reserved = NULL;
1189	u64 release_bytes = 0;
1190	u64 lockstart;
1191	u64 lockend;
1192	size_t num_written = 0;
1193	int nrptrs;
1194	ssize_t ret;
1195	bool only_release_metadata = false;
1196	bool force_page_uptodate = false;
1197	loff_t old_isize = i_size_read(inode);
1198	unsigned int ilock_flags = 0;
1199	const bool nowait = (iocb->ki_flags & IOCB_NOWAIT);
1200	unsigned int bdp_flags = (nowait ? BDP_ASYNC : 0);
1201
1202	if (nowait)
1203		ilock_flags |= BTRFS_ILOCK_TRY;
1204
1205	ret = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1206	if (ret < 0)
1207		return ret;
1208
1209	ret = generic_write_checks(iocb, i);
1210	if (ret <= 0)
1211		goto out;
1212
1213	ret = btrfs_write_check(iocb, i, ret);
1214	if (ret < 0)
1215		goto out;
1216
1217	pos = iocb->ki_pos;
1218	nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_SIZE),
1219			PAGE_SIZE / (sizeof(struct page *)));
1220	nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied);
1221	nrptrs = max(nrptrs, 8);
1222	pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL);
1223	if (!pages) {
1224		ret = -ENOMEM;
1225		goto out;
1226	}
1227
1228	while (iov_iter_count(i) > 0) {
1229		struct extent_state *cached_state = NULL;
1230		size_t offset = offset_in_page(pos);
1231		size_t sector_offset;
1232		size_t write_bytes = min(iov_iter_count(i),
1233					 nrptrs * (size_t)PAGE_SIZE -
1234					 offset);
1235		size_t num_pages;
 
1236		size_t reserve_bytes;
1237		size_t dirty_pages;
1238		size_t copied;
1239		size_t dirty_sectors;
1240		size_t num_sectors;
1241		int extents_locked;
1242
 
 
1243		/*
1244		 * Fault pages before locking them in prepare_pages
1245		 * to avoid recursive lock
1246		 */
1247		if (unlikely(fault_in_iov_iter_readable(i, write_bytes))) {
1248			ret = -EFAULT;
1249			break;
1250		}
1251
1252		only_release_metadata = false;
1253		sector_offset = pos & (fs_info->sectorsize - 1);
 
 
1254
1255		extent_changeset_release(data_reserved);
1256		ret = btrfs_check_data_free_space(BTRFS_I(inode),
1257						  &data_reserved, pos,
1258						  write_bytes, nowait);
1259		if (ret < 0) {
1260			int can_nocow;
1261
1262			if (nowait && (ret == -ENOSPC || ret == -EAGAIN)) {
1263				ret = -EAGAIN;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1264				break;
1265			}
1266
1267			/*
1268			 * If we don't have to COW at the offset, reserve
1269			 * metadata only. write_bytes may get smaller than
1270			 * requested here.
1271			 */
1272			can_nocow = btrfs_check_nocow_lock(BTRFS_I(inode), pos,
1273							   &write_bytes, nowait);
1274			if (can_nocow < 0)
1275				ret = can_nocow;
1276			if (can_nocow > 0)
1277				ret = 0;
1278			if (ret)
1279				break;
1280			only_release_metadata = true;
1281		}
1282
1283		num_pages = DIV_ROUND_UP(write_bytes + offset, PAGE_SIZE);
1284		WARN_ON(num_pages > nrptrs);
1285		reserve_bytes = round_up(write_bytes + sector_offset,
1286					 fs_info->sectorsize);
1287		WARN_ON(reserve_bytes == 0);
1288		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
1289						      reserve_bytes,
1290						      reserve_bytes, nowait);
1291		if (ret) {
1292			if (!only_release_metadata)
1293				btrfs_free_reserved_data_space(BTRFS_I(inode),
1294						data_reserved, pos,
1295						write_bytes);
1296			else
1297				btrfs_check_nocow_unlock(BTRFS_I(inode));
1298
1299			if (nowait && ret == -ENOSPC)
1300				ret = -EAGAIN;
1301			break;
1302		}
1303
1304		release_bytes = reserve_bytes;
1305again:
1306		ret = balance_dirty_pages_ratelimited_flags(inode->i_mapping, bdp_flags);
1307		if (ret) {
1308			btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1309			break;
1310		}
1311
1312		/*
1313		 * This is going to setup the pages array with the number of
1314		 * pages we want, so we don't really need to worry about the
1315		 * contents of pages from loop to loop
1316		 */
1317		ret = prepare_pages(inode, pages, num_pages,
1318				    pos, write_bytes, force_page_uptodate, false);
 
1319		if (ret) {
1320			btrfs_delalloc_release_extents(BTRFS_I(inode),
1321						       reserve_bytes);
1322			break;
1323		}
1324
1325		extents_locked = lock_and_cleanup_extent_if_need(
1326				BTRFS_I(inode), pages,
1327				num_pages, pos, write_bytes, &lockstart,
1328				&lockend, nowait, &cached_state);
1329		if (extents_locked < 0) {
1330			if (!nowait && extents_locked == -EAGAIN)
1331				goto again;
1332
1333			btrfs_delalloc_release_extents(BTRFS_I(inode),
1334						       reserve_bytes);
1335			ret = extents_locked;
1336			break;
1337		}
1338
1339		copied = btrfs_copy_from_user(pos, write_bytes, pages, i);
1340
1341		num_sectors = BTRFS_BYTES_TO_BLKS(fs_info, reserve_bytes);
1342		dirty_sectors = round_up(copied + sector_offset,
1343					fs_info->sectorsize);
1344		dirty_sectors = BTRFS_BYTES_TO_BLKS(fs_info, dirty_sectors);
1345
1346		/*
1347		 * if we have trouble faulting in the pages, fall
1348		 * back to one page at a time
1349		 */
1350		if (copied < write_bytes)
1351			nrptrs = 1;
1352
1353		if (copied == 0) {
1354			force_page_uptodate = true;
1355			dirty_sectors = 0;
1356			dirty_pages = 0;
1357		} else {
1358			force_page_uptodate = false;
1359			dirty_pages = DIV_ROUND_UP(copied + offset,
1360						   PAGE_SIZE);
1361		}
1362
1363		if (num_sectors > dirty_sectors) {
1364			/* release everything except the sectors we dirtied */
1365			release_bytes -= dirty_sectors << fs_info->sectorsize_bits;
 
1366			if (only_release_metadata) {
1367				btrfs_delalloc_release_metadata(BTRFS_I(inode),
1368							release_bytes, true);
1369			} else {
1370				u64 __pos;
1371
1372				__pos = round_down(pos,
1373						   fs_info->sectorsize) +
1374					(dirty_pages << PAGE_SHIFT);
1375				btrfs_delalloc_release_space(BTRFS_I(inode),
1376						data_reserved, __pos,
1377						release_bytes, true);
1378			}
1379		}
1380
1381		release_bytes = round_up(copied + sector_offset,
1382					fs_info->sectorsize);
1383
1384		ret = btrfs_dirty_pages(BTRFS_I(inode), pages,
1385					dirty_pages, pos, copied,
1386					&cached_state, only_release_metadata);
1387
1388		/*
1389		 * If we have not locked the extent range, because the range's
1390		 * start offset is >= i_size, we might still have a non-NULL
1391		 * cached extent state, acquired while marking the extent range
1392		 * as delalloc through btrfs_dirty_pages(). Therefore free any
1393		 * possible cached extent state to avoid a memory leak.
1394		 */
1395		if (extents_locked)
1396			unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
1397				      lockend, &cached_state);
1398		else
1399			free_extent_state(cached_state);
1400
1401		btrfs_delalloc_release_extents(BTRFS_I(inode), reserve_bytes);
1402		if (ret) {
1403			btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
1404			break;
1405		}
1406
1407		release_bytes = 0;
1408		if (only_release_metadata)
1409			btrfs_check_nocow_unlock(BTRFS_I(inode));
1410
1411		btrfs_drop_pages(fs_info, pages, num_pages, pos, copied);
 
 
 
 
 
 
 
 
 
 
 
 
1412
1413		cond_resched();
1414
 
 
 
 
1415		pos += copied;
1416		num_written += copied;
1417	}
1418
1419	kfree(pages);
1420
1421	if (release_bytes) {
1422		if (only_release_metadata) {
1423			btrfs_check_nocow_unlock(BTRFS_I(inode));
1424			btrfs_delalloc_release_metadata(BTRFS_I(inode),
1425					release_bytes, true);
1426		} else {
1427			btrfs_delalloc_release_space(BTRFS_I(inode),
1428					data_reserved,
1429					round_down(pos, fs_info->sectorsize),
1430					release_bytes, true);
1431		}
1432	}
1433
1434	extent_changeset_free(data_reserved);
1435	if (num_written > 0) {
1436		pagecache_isize_extended(inode, old_isize, iocb->ki_pos);
1437		iocb->ki_pos += num_written;
1438	}
1439out:
1440	btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1441	return num_written ? num_written : ret;
1442}
1443
1444static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
1445			       const struct iov_iter *iter, loff_t offset)
1446{
1447	const u32 blocksize_mask = fs_info->sectorsize - 1;
1448
1449	if (offset & blocksize_mask)
1450		return -EINVAL;
1451
1452	if (iov_iter_alignment(iter) & blocksize_mask)
1453		return -EINVAL;
1454
1455	return 0;
1456}
1457
1458static ssize_t btrfs_direct_write(struct kiocb *iocb, struct iov_iter *from)
1459{
1460	struct file *file = iocb->ki_filp;
1461	struct inode *inode = file_inode(file);
1462	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1463	loff_t pos;
1464	ssize_t written = 0;
1465	ssize_t written_buffered;
1466	size_t prev_left = 0;
1467	loff_t endbyte;
1468	ssize_t err;
1469	unsigned int ilock_flags = 0;
1470	struct iomap_dio *dio;
1471
1472	if (iocb->ki_flags & IOCB_NOWAIT)
1473		ilock_flags |= BTRFS_ILOCK_TRY;
1474
1475	/*
1476	 * If the write DIO is within EOF, use a shared lock and also only if
1477	 * security bits will likely not be dropped by file_remove_privs() called
1478	 * from btrfs_write_check(). Either will need to be rechecked after the
1479	 * lock was acquired.
1480	 */
1481	if (iocb->ki_pos + iov_iter_count(from) <= i_size_read(inode) && IS_NOSEC(inode))
1482		ilock_flags |= BTRFS_ILOCK_SHARED;
1483
1484relock:
1485	err = btrfs_inode_lock(BTRFS_I(inode), ilock_flags);
1486	if (err < 0)
1487		return err;
1488
1489	/* Shared lock cannot be used with security bits set. */
1490	if ((ilock_flags & BTRFS_ILOCK_SHARED) && !IS_NOSEC(inode)) {
1491		btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1492		ilock_flags &= ~BTRFS_ILOCK_SHARED;
1493		goto relock;
1494	}
1495
1496	err = generic_write_checks(iocb, from);
1497	if (err <= 0) {
1498		btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1499		return err;
1500	}
1501
1502	err = btrfs_write_check(iocb, from, err);
1503	if (err < 0) {
1504		btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1505		goto out;
1506	}
1507
1508	pos = iocb->ki_pos;
1509	/*
1510	 * Re-check since file size may have changed just before taking the
1511	 * lock or pos may have changed because of O_APPEND in generic_write_check()
1512	 */
1513	if ((ilock_flags & BTRFS_ILOCK_SHARED) &&
1514	    pos + iov_iter_count(from) > i_size_read(inode)) {
1515		btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1516		ilock_flags &= ~BTRFS_ILOCK_SHARED;
1517		goto relock;
1518	}
1519
1520	if (check_direct_IO(fs_info, from, pos)) {
1521		btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1522		goto buffered;
1523	}
1524
1525	/*
1526	 * The iov_iter can be mapped to the same file range we are writing to.
1527	 * If that's the case, then we will deadlock in the iomap code, because
1528	 * it first calls our callback btrfs_dio_iomap_begin(), which will create
1529	 * an ordered extent, and after that it will fault in the pages that the
1530	 * iov_iter refers to. During the fault in we end up in the readahead
1531	 * pages code (starting at btrfs_readahead()), which will lock the range,
1532	 * find that ordered extent and then wait for it to complete (at
1533	 * btrfs_lock_and_flush_ordered_range()), resulting in a deadlock since
1534	 * obviously the ordered extent can never complete as we didn't submit
1535	 * yet the respective bio(s). This always happens when the buffer is
1536	 * memory mapped to the same file range, since the iomap DIO code always
1537	 * invalidates pages in the target file range (after starting and waiting
1538	 * for any writeback).
1539	 *
1540	 * So here we disable page faults in the iov_iter and then retry if we
1541	 * got -EFAULT, faulting in the pages before the retry.
1542	 */
1543	from->nofault = true;
1544	dio = btrfs_dio_write(iocb, from, written);
1545	from->nofault = false;
1546
1547	/*
1548	 * iomap_dio_complete() will call btrfs_sync_file() if we have a dsync
1549	 * iocb, and that needs to lock the inode. So unlock it before calling
1550	 * iomap_dio_complete() to avoid a deadlock.
1551	 */
1552	btrfs_inode_unlock(BTRFS_I(inode), ilock_flags);
1553
1554	if (IS_ERR_OR_NULL(dio))
1555		err = PTR_ERR_OR_ZERO(dio);
1556	else
1557		err = iomap_dio_complete(dio);
1558
1559	/* No increment (+=) because iomap returns a cumulative value. */
1560	if (err > 0)
1561		written = err;
1562
1563	if (iov_iter_count(from) > 0 && (err == -EFAULT || err > 0)) {
1564		const size_t left = iov_iter_count(from);
1565		/*
1566		 * We have more data left to write. Try to fault in as many as
1567		 * possible of the remainder pages and retry. We do this without
1568		 * releasing and locking again the inode, to prevent races with
1569		 * truncate.
1570		 *
1571		 * Also, in case the iov refers to pages in the file range of the
1572		 * file we want to write to (due to a mmap), we could enter an
1573		 * infinite loop if we retry after faulting the pages in, since
1574		 * iomap will invalidate any pages in the range early on, before
1575		 * it tries to fault in the pages of the iov. So we keep track of
1576		 * how much was left of iov in the previous EFAULT and fallback
1577		 * to buffered IO in case we haven't made any progress.
1578		 */
1579		if (left == prev_left) {
1580			err = -ENOTBLK;
1581		} else {
1582			fault_in_iov_iter_readable(from, left);
1583			prev_left = left;
1584			goto relock;
1585		}
1586	}
1587
1588	/*
1589	 * If 'err' is -ENOTBLK or we have not written all data, then it means
1590	 * we must fallback to buffered IO.
1591	 */
1592	if ((err < 0 && err != -ENOTBLK) || !iov_iter_count(from))
1593		goto out;
1594
1595buffered:
1596	/*
1597	 * If we are in a NOWAIT context, then return -EAGAIN to signal the caller
1598	 * it must retry the operation in a context where blocking is acceptable,
1599	 * because even if we end up not blocking during the buffered IO attempt
1600	 * below, we will block when flushing and waiting for the IO.
1601	 */
1602	if (iocb->ki_flags & IOCB_NOWAIT) {
1603		err = -EAGAIN;
1604		goto out;
1605	}
1606
1607	pos = iocb->ki_pos;
1608	written_buffered = btrfs_buffered_write(iocb, from);
1609	if (written_buffered < 0) {
1610		err = written_buffered;
1611		goto out;
1612	}
1613	/*
1614	 * Ensure all data is persisted. We want the next direct IO read to be
1615	 * able to read what was just written.
1616	 */
1617	endbyte = pos + written_buffered - 1;
1618	err = btrfs_fdatawrite_range(inode, pos, endbyte);
1619	if (err)
1620		goto out;
1621	err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte);
1622	if (err)
1623		goto out;
1624	written += written_buffered;
1625	iocb->ki_pos = pos + written_buffered;
1626	invalidate_mapping_pages(file->f_mapping, pos >> PAGE_SHIFT,
1627				 endbyte >> PAGE_SHIFT);
1628out:
1629	return err < 0 ? err : written;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1630}
1631
1632static ssize_t btrfs_encoded_write(struct kiocb *iocb, struct iov_iter *from,
1633			const struct btrfs_ioctl_encoded_io_args *encoded)
1634{
1635	struct file *file = iocb->ki_filp;
1636	struct inode *inode = file_inode(file);
1637	loff_t count;
1638	ssize_t ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1639
1640	btrfs_inode_lock(BTRFS_I(inode), 0);
1641	count = encoded->len;
1642	ret = generic_write_checks_count(iocb, &count);
1643	if (ret == 0 && count != encoded->len) {
 
 
 
 
 
1644		/*
1645		 * The write got truncated by generic_write_checks_count(). We
1646		 * can't do a partial encoded write.
1647		 */
1648		ret = -EFBIG;
 
 
 
 
 
1649	}
1650	if (ret || encoded->len == 0)
 
 
 
 
1651		goto out;
 
1652
1653	ret = btrfs_write_check(iocb, from, encoded->len);
1654	if (ret < 0)
 
 
 
 
 
 
 
1655		goto out;
1656
1657	ret = btrfs_do_encoded_write(iocb, from, encoded);
1658out:
1659	btrfs_inode_unlock(BTRFS_I(inode), 0);
1660	return ret;
1661}
1662
1663ssize_t btrfs_do_write_iter(struct kiocb *iocb, struct iov_iter *from,
1664			    const struct btrfs_ioctl_encoded_io_args *encoded)
1665{
1666	struct file *file = iocb->ki_filp;
1667	struct btrfs_inode *inode = BTRFS_I(file_inode(file));
1668	ssize_t num_written, num_sync;
1669
1670	/*
1671	 * If the fs flips readonly due to some impossible error, although we
1672	 * have opened a file as writable, we have to stop this write operation
1673	 * to ensure consistency.
 
1674	 */
1675	if (BTRFS_FS_ERROR(inode->root->fs_info))
1676		return -EROFS;
1677
1678	if (encoded && (iocb->ki_flags & IOCB_NOWAIT))
1679		return -EOPNOTSUPP;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1680
1681	if (encoded) {
1682		num_written = btrfs_encoded_write(iocb, from, encoded);
1683		num_sync = encoded->len;
1684	} else if (iocb->ki_flags & IOCB_DIRECT) {
1685		num_written = btrfs_direct_write(iocb, from);
1686		num_sync = num_written;
1687	} else {
1688		num_written = btrfs_buffered_write(iocb, from);
1689		num_sync = num_written;
 
 
 
 
1690	}
1691
1692	btrfs_set_inode_last_sub_trans(inode);
1693
1694	if (num_sync > 0) {
1695		num_sync = generic_write_sync(iocb, num_sync);
1696		if (num_sync < 0)
1697			num_written = num_sync;
1698	}
 
 
 
 
 
1699
1700	return num_written;
1701}
1702
1703static ssize_t btrfs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1704{
1705	return btrfs_do_write_iter(iocb, from, NULL);
1706}
1707
1708int btrfs_release_file(struct inode *inode, struct file *filp)
1709{
1710	struct btrfs_file_private *private = filp->private_data;
1711
1712	if (private) {
1713		kfree(private->filldir_buf);
1714		free_extent_state(private->llseek_cached_state);
1715		kfree(private);
1716		filp->private_data = NULL;
1717	}
1718
1719	/*
1720	 * Set by setattr when we are about to truncate a file from a non-zero
1721	 * size to a zero size.  This tries to flush down new bytes that may
1722	 * have been written if the application were using truncate to replace
1723	 * a file in place.
1724	 */
1725	if (test_and_clear_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
1726			       &BTRFS_I(inode)->runtime_flags))
1727			filemap_flush(inode->i_mapping);
1728	return 0;
1729}
1730
1731static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end)
1732{
1733	int ret;
1734	struct blk_plug plug;
1735
1736	/*
1737	 * This is only called in fsync, which would do synchronous writes, so
1738	 * a plug can merge adjacent IOs as much as possible.  Esp. in case of
1739	 * multiple disks using raid profile, a large IO can be split to
1740	 * several segments of stripe length (currently 64K).
1741	 */
1742	blk_start_plug(&plug);
 
1743	ret = btrfs_fdatawrite_range(inode, start, end);
 
1744	blk_finish_plug(&plug);
1745
1746	return ret;
1747}
1748
1749static inline bool skip_inode_logging(const struct btrfs_log_ctx *ctx)
1750{
1751	struct btrfs_inode *inode = BTRFS_I(ctx->inode);
1752	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1753
1754	if (btrfs_inode_in_log(inode, btrfs_get_fs_generation(fs_info)) &&
1755	    list_empty(&ctx->ordered_extents))
1756		return true;
1757
1758	/*
1759	 * If we are doing a fast fsync we can not bail out if the inode's
1760	 * last_trans is <= then the last committed transaction, because we only
1761	 * update the last_trans of the inode during ordered extent completion,
1762	 * and for a fast fsync we don't wait for that, we only wait for the
1763	 * writeback to complete.
1764	 */
1765	if (inode->last_trans <= btrfs_get_last_trans_committed(fs_info) &&
1766	    (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) ||
1767	     list_empty(&ctx->ordered_extents)))
1768		return true;
1769
1770	return false;
1771}
1772
1773/*
1774 * fsync call for both files and directories.  This logs the inode into
1775 * the tree log instead of forcing full commits whenever possible.
1776 *
1777 * It needs to call filemap_fdatawait so that all ordered extent updates are
1778 * in the metadata btree are up to date for copying to the log.
1779 *
1780 * It drops the inode mutex before doing the tree log commit.  This is an
1781 * important optimization for directories because holding the mutex prevents
1782 * new operations on the dir while we write to disk.
1783 */
1784int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1785{
1786	struct dentry *dentry = file_dentry(file);
1787	struct inode *inode = d_inode(dentry);
1788	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
1789	struct btrfs_root *root = BTRFS_I(inode)->root;
1790	struct btrfs_trans_handle *trans;
1791	struct btrfs_log_ctx ctx;
1792	int ret = 0, err;
1793	u64 len;
1794	bool full_sync;
1795
1796	trace_btrfs_sync_file(file, datasync);
1797
1798	btrfs_init_log_ctx(&ctx, inode);
1799
1800	/*
1801	 * Always set the range to a full range, otherwise we can get into
1802	 * several problems, from missing file extent items to represent holes
1803	 * when not using the NO_HOLES feature, to log tree corruption due to
1804	 * races between hole detection during logging and completion of ordered
1805	 * extents outside the range, to missing checksums due to ordered extents
1806	 * for which we flushed only a subset of their pages.
1807	 */
1808	start = 0;
1809	end = LLONG_MAX;
1810	len = (u64)LLONG_MAX + 1;
1811
1812	/*
1813	 * We write the dirty pages in the range and wait until they complete
1814	 * out of the ->i_mutex. If so, we can flush the dirty pages by
1815	 * multi-task, and make the performance up.  See
1816	 * btrfs_wait_ordered_range for an explanation of the ASYNC check.
1817	 */
1818	ret = start_ordered_ops(inode, start, end);
1819	if (ret)
1820		goto out;
1821
1822	btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
 
 
 
 
 
 
 
1823
1824	atomic_inc(&root->log_batch);
1825
1826	/*
1827	 * Before we acquired the inode's lock and the mmap lock, someone may
1828	 * have dirtied more pages in the target range. We need to make sure
1829	 * that writeback for any such pages does not start while we are logging
1830	 * the inode, because if it does, any of the following might happen when
1831	 * we are not doing a full inode sync:
 
 
 
 
 
 
 
 
 
 
 
 
 
1832	 *
1833	 * 1) We log an extent after its writeback finishes but before its
1834	 *    checksums are added to the csum tree, leading to -EIO errors
1835	 *    when attempting to read the extent after a log replay.
1836	 *
1837	 * 2) We can end up logging an extent before its writeback finishes.
1838	 *    Therefore after the log replay we will have a file extent item
1839	 *    pointing to an unwritten extent (and no data checksums as well).
1840	 *
1841	 * So trigger writeback for any eventual new dirty pages and then we
1842	 * wait for all ordered extents to complete below.
1843	 */
1844	ret = start_ordered_ops(inode, start, end);
1845	if (ret) {
1846		btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1847		goto out;
1848	}
1849
1850	/*
1851	 * Always check for the full sync flag while holding the inode's lock,
1852	 * to avoid races with other tasks. The flag must be either set all the
1853	 * time during logging or always off all the time while logging.
1854	 * We check the flag here after starting delalloc above, because when
1855	 * running delalloc the full sync flag may be set if we need to drop
1856	 * extra extent map ranges due to temporary memory allocation failures.
1857	 */
1858	full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1859			     &BTRFS_I(inode)->runtime_flags);
1860
1861	/*
1862	 * We have to do this here to avoid the priority inversion of waiting on
1863	 * IO of a lower priority task while holding a transaction open.
1864	 *
1865	 * For a full fsync we wait for the ordered extents to complete while
1866	 * for a fast fsync we wait just for writeback to complete, and then
1867	 * attach the ordered extents to the transaction so that a transaction
1868	 * commit waits for their completion, to avoid data loss if we fsync,
1869	 * the current transaction commits before the ordered extents complete
1870	 * and a power failure happens right after that.
1871	 *
1872	 * For zoned filesystem, if a write IO uses a ZONE_APPEND command, the
1873	 * logical address recorded in the ordered extent may change. We need
1874	 * to wait for the IO to stabilize the logical address.
1875	 */
1876	if (full_sync || btrfs_is_zoned(fs_info)) {
1877		ret = btrfs_wait_ordered_range(inode, start, len);
1878	} else {
1879		/*
1880		 * Get our ordered extents as soon as possible to avoid doing
1881		 * checksum lookups in the csum tree, and use instead the
1882		 * checksums attached to the ordered extents.
1883		 */
1884		btrfs_get_ordered_extents_for_logging(BTRFS_I(inode),
1885						      &ctx.ordered_extents);
1886		ret = filemap_fdatawait_range(inode->i_mapping, start, end);
1887	}
1888
1889	if (ret)
1890		goto out_release_extents;
1891
1892	atomic_inc(&root->log_batch);
1893
1894	if (skip_inode_logging(&ctx)) {
 
 
1895		/*
1896		 * We've had everything committed since the last time we were
1897		 * modified so clear this flag in case it was set for whatever
1898		 * reason, it's no longer relevant.
1899		 */
1900		clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
1901			  &BTRFS_I(inode)->runtime_flags);
1902		/*
1903		 * An ordered extent might have started before and completed
1904		 * already with io errors, in which case the inode was not
1905		 * updated and we end up here. So check the inode's mapping
1906		 * for any errors that might have happened since we last
1907		 * checked called fsync.
1908		 */
1909		ret = filemap_check_wb_err(inode->i_mapping, file->f_wb_err);
1910		goto out_release_extents;
 
 
1911	}
1912
1913	btrfs_init_log_ctx_scratch_eb(&ctx);
1914
1915	/*
1916	 * We use start here because we will need to wait on the IO to complete
1917	 * in btrfs_sync_log, which could require joining a transaction (for
1918	 * example checking cross references in the nocow path).  If we use join
1919	 * here we could get into a situation where we're waiting on IO to
1920	 * happen that is blocked on a transaction trying to commit.  With start
1921	 * we inc the extwriter counter, so we wait for all extwriters to exit
1922	 * before we start blocking joiners.  This comment is to keep somebody
1923	 * from thinking they are super smart and changing this to
1924	 * btrfs_join_transaction *cough*Josef*cough*.
1925	 */
1926	trans = btrfs_start_transaction(root, 0);
1927	if (IS_ERR(trans)) {
1928		ret = PTR_ERR(trans);
1929		goto out_release_extents;
 
 
1930	}
1931	trans->in_fsync = true;
1932
1933	ret = btrfs_log_dentry_safe(trans, dentry, &ctx);
1934	/*
1935	 * Scratch eb no longer needed, release before syncing log or commit
1936	 * transaction, to avoid holding unnecessary memory during such long
1937	 * operations.
1938	 */
1939	if (ctx.scratch_eb) {
1940		free_extent_buffer(ctx.scratch_eb);
1941		ctx.scratch_eb = NULL;
1942	}
1943	btrfs_release_log_ctx_extents(&ctx);
1944	if (ret < 0) {
1945		/* Fallthrough and commit/free transaction. */
1946		ret = BTRFS_LOG_FORCE_COMMIT;
1947	}
1948
1949	/* we've logged all the items and now have a consistent
1950	 * version of the file in the log.  It is possible that
1951	 * someone will come in and modify the file, but that's
1952	 * fine because the log is consistent on disk, and we
1953	 * have references to all of the file's extents
1954	 *
1955	 * It is possible that someone will come in and log the
1956	 * file again, but that will end up using the synchronization
1957	 * inside btrfs_sync_log to keep things safe.
1958	 */
1959	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
1960
1961	if (ret == BTRFS_NO_LOG_SYNC) {
1962		ret = btrfs_end_transaction(trans);
1963		goto out;
1964	}
1965
1966	/* We successfully logged the inode, attempt to sync the log. */
1967	if (!ret) {
1968		ret = btrfs_sync_log(trans, root, &ctx);
1969		if (!ret) {
1970			ret = btrfs_end_transaction(trans);
1971			goto out;
 
 
 
1972		}
1973	}
1974
1975	/*
1976	 * At this point we need to commit the transaction because we had
1977	 * btrfs_need_log_full_commit() or some other error.
1978	 *
1979	 * If we didn't do a full sync we have to stop the trans handle, wait on
1980	 * the ordered extents, start it again and commit the transaction.  If
1981	 * we attempt to wait on the ordered extents here we could deadlock with
1982	 * something like fallocate() that is holding the extent lock trying to
1983	 * start a transaction while some other thread is trying to commit the
1984	 * transaction while we (fsync) are currently holding the transaction
1985	 * open.
1986	 */
1987	if (!full_sync) {
1988		ret = btrfs_end_transaction(trans);
1989		if (ret)
1990			goto out;
1991		ret = btrfs_wait_ordered_range(inode, start, len);
1992		if (ret)
1993			goto out;
1994
1995		/*
1996		 * This is safe to use here because we're only interested in
1997		 * making sure the transaction that had the ordered extents is
1998		 * committed.  We aren't waiting on anything past this point,
1999		 * we're purely getting the transaction and committing it.
2000		 */
2001		trans = btrfs_attach_transaction_barrier(root);
2002		if (IS_ERR(trans)) {
2003			ret = PTR_ERR(trans);
2004
2005			/*
2006			 * We committed the transaction and there's no currently
2007			 * running transaction, this means everything we care
2008			 * about made it to disk and we are done.
2009			 */
2010			if (ret == -ENOENT)
2011				ret = 0;
2012			goto out;
2013		}
2014	}
2015
2016	ret = btrfs_commit_transaction(trans);
2017out:
2018	free_extent_buffer(ctx.scratch_eb);
2019	ASSERT(list_empty(&ctx.list));
2020	ASSERT(list_empty(&ctx.conflict_inodes));
2021	err = file_check_and_advance_wb_err(file);
2022	if (!ret)
2023		ret = err;
2024	return ret > 0 ? -EIO : ret;
2025
2026out_release_extents:
2027	btrfs_release_log_ctx_extents(&ctx);
2028	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2029	goto out;
2030}
2031
2032static const struct vm_operations_struct btrfs_file_vm_ops = {
2033	.fault		= filemap_fault,
2034	.map_pages	= filemap_map_pages,
2035	.page_mkwrite	= btrfs_page_mkwrite,
2036};
2037
2038static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
2039{
2040	struct address_space *mapping = filp->f_mapping;
2041
2042	if (!mapping->a_ops->read_folio)
2043		return -ENOEXEC;
2044
2045	file_accessed(filp);
2046	vma->vm_ops = &btrfs_file_vm_ops;
2047
2048	return 0;
2049}
2050
2051static int hole_mergeable(struct btrfs_inode *inode, struct extent_buffer *leaf,
2052			  int slot, u64 start, u64 end)
2053{
2054	struct btrfs_file_extent_item *fi;
2055	struct btrfs_key key;
2056
2057	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
2058		return 0;
2059
2060	btrfs_item_key_to_cpu(leaf, &key, slot);
2061	if (key.objectid != btrfs_ino(inode) ||
2062	    key.type != BTRFS_EXTENT_DATA_KEY)
2063		return 0;
2064
2065	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2066
2067	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2068		return 0;
2069
2070	if (btrfs_file_extent_disk_bytenr(leaf, fi))
2071		return 0;
2072
2073	if (key.offset == end)
2074		return 1;
2075	if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start)
2076		return 1;
2077	return 0;
2078}
2079
2080static int fill_holes(struct btrfs_trans_handle *trans,
2081		struct btrfs_inode *inode,
2082		struct btrfs_path *path, u64 offset, u64 end)
2083{
2084	struct btrfs_fs_info *fs_info = trans->fs_info;
2085	struct btrfs_root *root = inode->root;
2086	struct extent_buffer *leaf;
2087	struct btrfs_file_extent_item *fi;
2088	struct extent_map *hole_em;
 
2089	struct btrfs_key key;
2090	int ret;
2091
2092	if (btrfs_fs_incompat(fs_info, NO_HOLES))
2093		goto out;
2094
2095	key.objectid = btrfs_ino(inode);
2096	key.type = BTRFS_EXTENT_DATA_KEY;
2097	key.offset = offset;
2098
2099	ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2100	if (ret <= 0) {
2101		/*
2102		 * We should have dropped this offset, so if we find it then
2103		 * something has gone horribly wrong.
2104		 */
2105		if (ret == 0)
2106			ret = -EINVAL;
2107		return ret;
2108	}
2109
2110	leaf = path->nodes[0];
2111	if (hole_mergeable(inode, leaf, path->slots[0] - 1, offset, end)) {
2112		u64 num_bytes;
2113
2114		path->slots[0]--;
2115		fi = btrfs_item_ptr(leaf, path->slots[0],
2116				    struct btrfs_file_extent_item);
2117		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) +
2118			end - offset;
2119		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2120		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2121		btrfs_set_file_extent_offset(leaf, fi, 0);
2122		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123		btrfs_mark_buffer_dirty(trans, leaf);
2124		goto out;
2125	}
2126
2127	if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) {
2128		u64 num_bytes;
2129
2130		key.offset = offset;
2131		btrfs_set_item_key_safe(trans, path, &key);
2132		fi = btrfs_item_ptr(leaf, path->slots[0],
2133				    struct btrfs_file_extent_item);
2134		num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end -
2135			offset;
2136		btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2137		btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes);
2138		btrfs_set_file_extent_offset(leaf, fi, 0);
2139		btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2140		btrfs_mark_buffer_dirty(trans, leaf);
2141		goto out;
2142	}
2143	btrfs_release_path(path);
2144
2145	ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset,
2146				       end - offset);
2147	if (ret)
2148		return ret;
2149
2150out:
2151	btrfs_release_path(path);
2152
2153	hole_em = alloc_extent_map();
2154	if (!hole_em) {
2155		btrfs_drop_extent_map_range(inode, offset, end - 1, false);
2156		btrfs_set_inode_full_sync(inode);
2157	} else {
2158		hole_em->start = offset;
2159		hole_em->len = end - offset;
2160		hole_em->ram_bytes = hole_em->len;
2161		hole_em->orig_start = offset;
2162
2163		hole_em->block_start = EXTENT_MAP_HOLE;
2164		hole_em->block_len = 0;
2165		hole_em->orig_block_len = 0;
 
 
2166		hole_em->generation = trans->transid;
2167
2168		ret = btrfs_replace_extent_map_range(inode, hole_em, true);
 
 
 
 
 
2169		free_extent_map(hole_em);
2170		if (ret)
2171			btrfs_set_inode_full_sync(inode);
 
2172	}
2173
2174	return 0;
2175}
2176
2177/*
2178 * Find a hole extent on given inode and change start/len to the end of hole
2179 * extent.(hole/vacuum extent whose em->start <= start &&
2180 *	   em->start + em->len > start)
2181 * When a hole extent is found, return 1 and modify start/len.
2182 */
2183static int find_first_non_hole(struct btrfs_inode *inode, u64 *start, u64 *len)
2184{
2185	struct btrfs_fs_info *fs_info = inode->root->fs_info;
2186	struct extent_map *em;
2187	int ret = 0;
2188
2189	em = btrfs_get_extent(inode, NULL,
2190			      round_down(*start, fs_info->sectorsize),
2191			      round_up(*len, fs_info->sectorsize));
2192	if (IS_ERR(em))
2193		return PTR_ERR(em);
2194
2195	/* Hole or vacuum extent(only exists in no-hole mode) */
2196	if (em->block_start == EXTENT_MAP_HOLE) {
2197		ret = 1;
2198		*len = em->start + em->len > *start + *len ?
2199		       0 : *start + *len - em->start - em->len;
2200		*start = em->start + em->len;
2201	}
2202	free_extent_map(em);
2203	return ret;
2204}
2205
2206static void btrfs_punch_hole_lock_range(struct inode *inode,
2207					const u64 lockstart,
2208					const u64 lockend,
2209					struct extent_state **cached_state)
2210{
2211	/*
2212	 * For subpage case, if the range is not at page boundary, we could
2213	 * have pages at the leading/tailing part of the range.
2214	 * This could lead to dead loop since filemap_range_has_page()
2215	 * will always return true.
2216	 * So here we need to do extra page alignment for
2217	 * filemap_range_has_page().
2218	 */
2219	const u64 page_lockstart = round_up(lockstart, PAGE_SIZE);
2220	const u64 page_lockend = round_down(lockend + 1, PAGE_SIZE) - 1;
2221
2222	while (1) {
2223		truncate_pagecache_range(inode, lockstart, lockend);
2224
2225		lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2226			    cached_state);
2227		/*
2228		 * We can't have ordered extents in the range, nor dirty/writeback
2229		 * pages, because we have locked the inode's VFS lock in exclusive
2230		 * mode, we have locked the inode's i_mmap_lock in exclusive mode,
2231		 * we have flushed all delalloc in the range and we have waited
2232		 * for any ordered extents in the range to complete.
2233		 * We can race with anyone reading pages from this range, so after
2234		 * locking the range check if we have pages in the range, and if
2235		 * we do, unlock the range and retry.
2236		 */
2237		if (!filemap_range_has_page(inode->i_mapping, page_lockstart,
2238					    page_lockend))
 
 
2239			break;
2240
2241		unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2242			      cached_state);
 
 
 
 
 
 
2243	}
2244
2245	btrfs_assert_inode_range_clean(BTRFS_I(inode), lockstart, lockend);
2246}
2247
2248static int btrfs_insert_replace_extent(struct btrfs_trans_handle *trans,
2249				     struct btrfs_inode *inode,
2250				     struct btrfs_path *path,
2251				     struct btrfs_replace_extent_info *extent_info,
2252				     const u64 replace_len,
2253				     const u64 bytes_to_drop)
2254{
2255	struct btrfs_fs_info *fs_info = trans->fs_info;
2256	struct btrfs_root *root = inode->root;
2257	struct btrfs_file_extent_item *extent;
2258	struct extent_buffer *leaf;
2259	struct btrfs_key key;
2260	int slot;
2261	struct btrfs_ref ref = { 0 };
 
2262	int ret;
2263
2264	if (replace_len == 0)
2265		return 0;
2266
2267	if (extent_info->disk_offset == 0 &&
2268	    btrfs_fs_incompat(fs_info, NO_HOLES)) {
2269		btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2270		return 0;
2271	}
2272
2273	key.objectid = btrfs_ino(inode);
2274	key.type = BTRFS_EXTENT_DATA_KEY;
2275	key.offset = extent_info->file_offset;
2276	ret = btrfs_insert_empty_item(trans, root, path, &key,
2277				      sizeof(struct btrfs_file_extent_item));
2278	if (ret)
2279		return ret;
2280	leaf = path->nodes[0];
2281	slot = path->slots[0];
2282	write_extent_buffer(leaf, extent_info->extent_buf,
2283			    btrfs_item_ptr_offset(leaf, slot),
2284			    sizeof(struct btrfs_file_extent_item));
2285	extent = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
2286	ASSERT(btrfs_file_extent_type(leaf, extent) != BTRFS_FILE_EXTENT_INLINE);
2287	btrfs_set_file_extent_offset(leaf, extent, extent_info->data_offset);
2288	btrfs_set_file_extent_num_bytes(leaf, extent, replace_len);
2289	if (extent_info->is_new_extent)
2290		btrfs_set_file_extent_generation(leaf, extent, trans->transid);
2291	btrfs_mark_buffer_dirty(trans, leaf);
2292	btrfs_release_path(path);
2293
2294	ret = btrfs_inode_set_file_extent_range(inode, extent_info->file_offset,
2295						replace_len);
2296	if (ret)
2297		return ret;
2298
2299	/* If it's a hole, nothing more needs to be done. */
2300	if (extent_info->disk_offset == 0) {
2301		btrfs_update_inode_bytes(inode, 0, bytes_to_drop);
2302		return 0;
2303	}
2304
2305	btrfs_update_inode_bytes(inode, replace_len, bytes_to_drop);
2306
2307	if (extent_info->is_new_extent && extent_info->insertions == 0) {
2308		key.objectid = extent_info->disk_offset;
2309		key.type = BTRFS_EXTENT_ITEM_KEY;
2310		key.offset = extent_info->disk_len;
2311		ret = btrfs_alloc_reserved_file_extent(trans, root,
2312						       btrfs_ino(inode),
2313						       extent_info->file_offset,
2314						       extent_info->qgroup_reserved,
2315						       &key);
2316	} else {
2317		u64 ref_offset;
2318
2319		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2320				       extent_info->disk_offset,
2321				       extent_info->disk_len, 0,
2322				       root->root_key.objectid);
2323		ref_offset = extent_info->file_offset - extent_info->data_offset;
2324		btrfs_init_data_ref(&ref, root->root_key.objectid,
2325				    btrfs_ino(inode), ref_offset, 0, false);
2326		ret = btrfs_inc_extent_ref(trans, &ref);
2327	}
2328
2329	extent_info->insertions++;
 
 
 
 
 
 
 
2330
2331	return ret;
2332}
2333
2334/*
2335 * The respective range must have been previously locked, as well as the inode.
2336 * The end offset is inclusive (last byte of the range).
2337 * @extent_info is NULL for fallocate's hole punching and non-NULL when replacing
2338 * the file range with an extent.
2339 * When not punching a hole, we don't want to end up in a state where we dropped
2340 * extents without inserting a new one, so we must abort the transaction to avoid
2341 * a corruption.
2342 */
2343int btrfs_replace_file_extents(struct btrfs_inode *inode,
2344			       struct btrfs_path *path, const u64 start,
2345			       const u64 end,
2346			       struct btrfs_replace_extent_info *extent_info,
2347			       struct btrfs_trans_handle **trans_out)
2348{
2349	struct btrfs_drop_extents_args drop_args = { 0 };
2350	struct btrfs_root *root = inode->root;
2351	struct btrfs_fs_info *fs_info = root->fs_info;
2352	u64 min_size = btrfs_calc_insert_metadata_size(fs_info, 1);
2353	u64 ino_size = round_up(inode->vfs_inode.i_size, fs_info->sectorsize);
 
2354	struct btrfs_trans_handle *trans = NULL;
2355	struct btrfs_block_rsv *rsv;
2356	unsigned int rsv_count;
2357	u64 cur_offset;
 
2358	u64 len = end - start;
2359	int ret = 0;
2360
2361	if (end <= start)
2362		return -EINVAL;
2363
2364	rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
2365	if (!rsv) {
2366		ret = -ENOMEM;
2367		goto out;
2368	}
2369	rsv->size = btrfs_calc_insert_metadata_size(fs_info, 1);
2370	rsv->failfast = true;
2371
2372	/*
2373	 * 1 - update the inode
2374	 * 1 - removing the extents in the range
2375	 * 1 - adding the hole extent if no_holes isn't set or if we are
2376	 *     replacing the range with a new extent
2377	 */
2378	if (!btrfs_fs_incompat(fs_info, NO_HOLES) || extent_info)
2379		rsv_count = 3;
2380	else
2381		rsv_count = 2;
2382
2383	trans = btrfs_start_transaction(root, rsv_count);
2384	if (IS_ERR(trans)) {
2385		ret = PTR_ERR(trans);
2386		trans = NULL;
2387		goto out_free;
2388	}
2389
2390	ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
2391				      min_size, false);
2392	if (WARN_ON(ret))
2393		goto out_trans;
2394	trans->block_rsv = rsv;
2395
2396	cur_offset = start;
2397	drop_args.path = path;
2398	drop_args.end = end + 1;
2399	drop_args.drop_cache = true;
2400	while (cur_offset < end) {
2401		drop_args.start = cur_offset;
2402		ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2403		/* If we are punching a hole decrement the inode's byte count */
2404		if (!extent_info)
2405			btrfs_update_inode_bytes(inode, 0,
2406						 drop_args.bytes_found);
2407		if (ret != -ENOSPC) {
2408			/*
2409			 * The only time we don't want to abort is if we are
2410			 * attempting to clone a partial inline extent, in which
2411			 * case we'll get EOPNOTSUPP.  However if we aren't
2412			 * clone we need to abort no matter what, because if we
2413			 * got EOPNOTSUPP via prealloc then we messed up and
2414			 * need to abort.
2415			 */
2416			if (ret &&
2417			    (ret != -EOPNOTSUPP ||
2418			     (extent_info && extent_info->is_new_extent)))
2419				btrfs_abort_transaction(trans, ret);
2420			break;
2421		}
2422
2423		trans->block_rsv = &fs_info->trans_block_rsv;
2424
2425		if (!extent_info && cur_offset < drop_args.drop_end &&
2426		    cur_offset < ino_size) {
2427			ret = fill_holes(trans, inode, path, cur_offset,
2428					 drop_args.drop_end);
2429			if (ret) {
2430				/*
2431				 * If we failed then we didn't insert our hole
2432				 * entries for the area we dropped, so now the
2433				 * fs is corrupted, so we must abort the
2434				 * transaction.
2435				 */
2436				btrfs_abort_transaction(trans, ret);
2437				break;
2438			}
2439		} else if (!extent_info && cur_offset < drop_args.drop_end) {
2440			/*
2441			 * We are past the i_size here, but since we didn't
2442			 * insert holes we need to clear the mapped area so we
2443			 * know to not set disk_i_size in this area until a new
2444			 * file extent is inserted here.
2445			 */
2446			ret = btrfs_inode_clear_file_extent_range(inode,
2447					cur_offset,
2448					drop_args.drop_end - cur_offset);
2449			if (ret) {
2450				/*
2451				 * We couldn't clear our area, so we could
2452				 * presumably adjust up and corrupt the fs, so
2453				 * we need to abort.
2454				 */
2455				btrfs_abort_transaction(trans, ret);
2456				break;
2457			}
2458		}
2459
2460		if (extent_info &&
2461		    drop_args.drop_end > extent_info->file_offset) {
2462			u64 replace_len = drop_args.drop_end -
2463					  extent_info->file_offset;
2464
2465			ret = btrfs_insert_replace_extent(trans, inode,	path,
2466					extent_info, replace_len,
2467					drop_args.bytes_found);
2468			if (ret) {
2469				btrfs_abort_transaction(trans, ret);
2470				break;
2471			}
2472			extent_info->data_len -= replace_len;
2473			extent_info->data_offset += replace_len;
2474			extent_info->file_offset += replace_len;
2475		}
2476
2477		/*
2478		 * We are releasing our handle on the transaction, balance the
2479		 * dirty pages of the btree inode and flush delayed items, and
2480		 * then get a new transaction handle, which may now point to a
2481		 * new transaction in case someone else may have committed the
2482		 * transaction we used to replace/drop file extent items. So
2483		 * bump the inode's iversion and update mtime and ctime except
2484		 * if we are called from a dedupe context. This is because a
2485		 * power failure/crash may happen after the transaction is
2486		 * committed and before we finish replacing/dropping all the
2487		 * file extent items we need.
2488		 */
2489		inode_inc_iversion(&inode->vfs_inode);
2490
2491		if (!extent_info || extent_info->update_times)
2492			inode_set_mtime_to_ts(&inode->vfs_inode,
2493					      inode_set_ctime_current(&inode->vfs_inode));
2494
2495		ret = btrfs_update_inode(trans, inode);
2496		if (ret)
2497			break;
2498
2499		btrfs_end_transaction(trans);
2500		btrfs_btree_balance_dirty(fs_info);
2501
2502		trans = btrfs_start_transaction(root, rsv_count);
2503		if (IS_ERR(trans)) {
2504			ret = PTR_ERR(trans);
2505			trans = NULL;
2506			break;
2507		}
2508
2509		ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
2510					      rsv, min_size, false);
2511		if (WARN_ON(ret))
2512			break;
2513		trans->block_rsv = rsv;
2514
2515		cur_offset = drop_args.drop_end;
2516		len = end - cur_offset;
2517		if (!extent_info && len) {
2518			ret = find_first_non_hole(inode, &cur_offset, &len);
2519			if (unlikely(ret < 0))
2520				break;
2521			if (ret && !len) {
2522				ret = 0;
2523				break;
2524			}
2525		}
2526	}
2527
2528	/*
2529	 * If we were cloning, force the next fsync to be a full one since we
2530	 * we replaced (or just dropped in the case of cloning holes when
2531	 * NO_HOLES is enabled) file extent items and did not setup new extent
2532	 * maps for the replacement extents (or holes).
2533	 */
2534	if (extent_info && !extent_info->is_new_extent)
2535		btrfs_set_inode_full_sync(inode);
 
 
 
2536
2537	if (ret)
2538		goto out_trans;
2539
2540	trans->block_rsv = &fs_info->trans_block_rsv;
2541	/*
2542	 * If we are using the NO_HOLES feature we might have had already an
2543	 * hole that overlaps a part of the region [lockstart, lockend] and
2544	 * ends at (or beyond) lockend. Since we have no file extent items to
2545	 * represent holes, drop_end can be less than lockend and so we must
2546	 * make sure we have an extent map representing the existing hole (the
2547	 * call to __btrfs_drop_extents() might have dropped the existing extent
2548	 * map representing the existing hole), otherwise the fast fsync path
2549	 * will not record the existence of the hole region
2550	 * [existing_hole_start, lockend].
2551	 */
2552	if (drop_args.drop_end <= end)
2553		drop_args.drop_end = end + 1;
2554	/*
2555	 * Don't insert file hole extent item if it's for a range beyond eof
2556	 * (because it's useless) or if it represents a 0 bytes range (when
2557	 * cur_offset == drop_end).
2558	 */
2559	if (!extent_info && cur_offset < ino_size &&
2560	    cur_offset < drop_args.drop_end) {
2561		ret = fill_holes(trans, inode, path, cur_offset,
2562				 drop_args.drop_end);
2563		if (ret) {
2564			/* Same comment as above. */
2565			btrfs_abort_transaction(trans, ret);
2566			goto out_trans;
2567		}
2568	} else if (!extent_info && cur_offset < drop_args.drop_end) {
2569		/* See the comment in the loop above for the reasoning here. */
2570		ret = btrfs_inode_clear_file_extent_range(inode, cur_offset,
2571					drop_args.drop_end - cur_offset);
2572		if (ret) {
2573			btrfs_abort_transaction(trans, ret);
2574			goto out_trans;
2575		}
2576
2577	}
2578	if (extent_info) {
2579		ret = btrfs_insert_replace_extent(trans, inode, path,
2580				extent_info, extent_info->data_len,
2581				drop_args.bytes_found);
2582		if (ret) {
2583			btrfs_abort_transaction(trans, ret);
2584			goto out_trans;
2585		}
2586	}
2587
2588out_trans:
2589	if (!trans)
2590		goto out_free;
2591
2592	trans->block_rsv = &fs_info->trans_block_rsv;
2593	if (ret)
2594		btrfs_end_transaction(trans);
2595	else
2596		*trans_out = trans;
2597out_free:
2598	btrfs_free_block_rsv(fs_info, rsv);
2599out:
2600	return ret;
2601}
2602
2603static int btrfs_punch_hole(struct file *file, loff_t offset, loff_t len)
2604{
2605	struct inode *inode = file_inode(file);
2606	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2607	struct btrfs_root *root = BTRFS_I(inode)->root;
2608	struct extent_state *cached_state = NULL;
2609	struct btrfs_path *path;
2610	struct btrfs_trans_handle *trans = NULL;
2611	u64 lockstart;
2612	u64 lockend;
2613	u64 tail_start;
2614	u64 tail_len;
2615	u64 orig_start = offset;
2616	int ret = 0;
2617	bool same_block;
2618	u64 ino_size;
2619	bool truncated_block = false;
2620	bool updated_inode = false;
2621
2622	btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2623
2624	ret = btrfs_wait_ordered_range(inode, offset, len);
2625	if (ret)
2626		goto out_only_mutex;
2627
 
2628	ino_size = round_up(inode->i_size, fs_info->sectorsize);
2629	ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2630	if (ret < 0)
2631		goto out_only_mutex;
2632	if (ret && !len) {
2633		/* Already in a large hole */
2634		ret = 0;
2635		goto out_only_mutex;
2636	}
2637
2638	ret = file_modified(file);
2639	if (ret)
2640		goto out_only_mutex;
2641
2642	lockstart = round_up(offset, fs_info->sectorsize);
2643	lockend = round_down(offset + len, fs_info->sectorsize) - 1;
2644	same_block = (BTRFS_BYTES_TO_BLKS(fs_info, offset))
2645		== (BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1));
2646	/*
2647	 * We needn't truncate any block which is beyond the end of the file
2648	 * because we are sure there is no data there.
2649	 */
2650	/*
2651	 * Only do this if we are in the same block and we aren't doing the
2652	 * entire block.
2653	 */
2654	if (same_block && len < fs_info->sectorsize) {
2655		if (offset < ino_size) {
2656			truncated_block = true;
2657			ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2658						   0);
2659		} else {
2660			ret = 0;
2661		}
2662		goto out_only_mutex;
2663	}
2664
2665	/* zero back part of the first block */
2666	if (offset < ino_size) {
2667		truncated_block = true;
2668		ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2669		if (ret) {
2670			btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2671			return ret;
2672		}
2673	}
2674
2675	/* Check the aligned pages after the first unaligned page,
2676	 * if offset != orig_start, which means the first unaligned page
2677	 * including several following pages are already in holes,
2678	 * the extra check can be skipped */
2679	if (offset == orig_start) {
2680		/* after truncate page, check hole again */
2681		len = offset + len - lockstart;
2682		offset = lockstart;
2683		ret = find_first_non_hole(BTRFS_I(inode), &offset, &len);
2684		if (ret < 0)
2685			goto out_only_mutex;
2686		if (ret && !len) {
2687			ret = 0;
2688			goto out_only_mutex;
2689		}
2690		lockstart = offset;
2691	}
2692
2693	/* Check the tail unaligned part is in a hole */
2694	tail_start = lockend + 1;
2695	tail_len = offset + len - tail_start;
2696	if (tail_len) {
2697		ret = find_first_non_hole(BTRFS_I(inode), &tail_start, &tail_len);
2698		if (unlikely(ret < 0))
2699			goto out_only_mutex;
2700		if (!ret) {
2701			/* zero the front end of the last page */
2702			if (tail_start + tail_len < ino_size) {
2703				truncated_block = true;
2704				ret = btrfs_truncate_block(BTRFS_I(inode),
2705							tail_start + tail_len,
2706							0, 1);
2707				if (ret)
2708					goto out_only_mutex;
2709			}
2710		}
2711	}
2712
2713	if (lockend < lockstart) {
2714		ret = 0;
2715		goto out_only_mutex;
2716	}
2717
2718	btrfs_punch_hole_lock_range(inode, lockstart, lockend, &cached_state);
 
 
 
2719
2720	path = btrfs_alloc_path();
2721	if (!path) {
2722		ret = -ENOMEM;
2723		goto out;
2724	}
2725
2726	ret = btrfs_replace_file_extents(BTRFS_I(inode), path, lockstart,
2727					 lockend, NULL, &trans);
2728	btrfs_free_path(path);
2729	if (ret)
2730		goto out;
2731
2732	ASSERT(trans != NULL);
2733	inode_inc_iversion(inode);
2734	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
2735	ret = btrfs_update_inode(trans, BTRFS_I(inode));
2736	updated_inode = true;
2737	btrfs_end_transaction(trans);
2738	btrfs_btree_balance_dirty(fs_info);
2739out:
2740	unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
2741		      &cached_state);
2742out_only_mutex:
2743	if (!updated_inode && truncated_block && !ret) {
2744		/*
2745		 * If we only end up zeroing part of a page, we still need to
2746		 * update the inode item, so that all the time fields are
2747		 * updated as well as the necessary btrfs inode in memory fields
2748		 * for detecting, at fsync time, if the inode isn't yet in the
2749		 * log tree or it's there but not up to date.
2750		 */
2751		struct timespec64 now = inode_set_ctime_current(inode);
2752
2753		inode_inc_iversion(inode);
2754		inode_set_mtime_to_ts(inode, now);
 
2755		trans = btrfs_start_transaction(root, 1);
2756		if (IS_ERR(trans)) {
2757			ret = PTR_ERR(trans);
2758		} else {
2759			int ret2;
2760
2761			ret = btrfs_update_inode(trans, BTRFS_I(inode));
2762			ret2 = btrfs_end_transaction(trans);
2763			if (!ret)
2764				ret = ret2;
2765		}
2766	}
2767	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
2768	return ret;
2769}
2770
2771/* Helper structure to record which range is already reserved */
2772struct falloc_range {
2773	struct list_head list;
2774	u64 start;
2775	u64 len;
2776};
2777
2778/*
2779 * Helper function to add falloc range
2780 *
2781 * Caller should have locked the larger range of extent containing
2782 * [start, len)
2783 */
2784static int add_falloc_range(struct list_head *head, u64 start, u64 len)
2785{
 
2786	struct falloc_range *range = NULL;
2787
2788	if (!list_empty(head)) {
2789		/*
2790		 * As fallocate iterates by bytenr order, we only need to check
2791		 * the last range.
2792		 */
2793		range = list_last_entry(head, struct falloc_range, list);
2794		if (range->start + range->len == start) {
2795			range->len += len;
2796			return 0;
2797		}
 
2798	}
2799
2800	range = kmalloc(sizeof(*range), GFP_KERNEL);
2801	if (!range)
2802		return -ENOMEM;
2803	range->start = start;
2804	range->len = len;
2805	list_add_tail(&range->list, head);
2806	return 0;
2807}
2808
2809static int btrfs_fallocate_update_isize(struct inode *inode,
2810					const u64 end,
2811					const int mode)
2812{
2813	struct btrfs_trans_handle *trans;
2814	struct btrfs_root *root = BTRFS_I(inode)->root;
2815	int ret;
2816	int ret2;
2817
2818	if (mode & FALLOC_FL_KEEP_SIZE || end <= i_size_read(inode))
2819		return 0;
2820
2821	trans = btrfs_start_transaction(root, 1);
2822	if (IS_ERR(trans))
2823		return PTR_ERR(trans);
2824
2825	inode_set_ctime_current(inode);
2826	i_size_write(inode, end);
2827	btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2828	ret = btrfs_update_inode(trans, BTRFS_I(inode));
2829	ret2 = btrfs_end_transaction(trans);
2830
2831	return ret ? ret : ret2;
2832}
2833
2834enum {
2835	RANGE_BOUNDARY_WRITTEN_EXTENT,
2836	RANGE_BOUNDARY_PREALLOC_EXTENT,
2837	RANGE_BOUNDARY_HOLE,
2838};
2839
2840static int btrfs_zero_range_check_range_boundary(struct btrfs_inode *inode,
2841						 u64 offset)
2842{
2843	const u64 sectorsize = inode->root->fs_info->sectorsize;
2844	struct extent_map *em;
2845	int ret;
2846
2847	offset = round_down(offset, sectorsize);
2848	em = btrfs_get_extent(inode, NULL, offset, sectorsize);
2849	if (IS_ERR(em))
2850		return PTR_ERR(em);
2851
2852	if (em->block_start == EXTENT_MAP_HOLE)
2853		ret = RANGE_BOUNDARY_HOLE;
2854	else if (em->flags & EXTENT_FLAG_PREALLOC)
2855		ret = RANGE_BOUNDARY_PREALLOC_EXTENT;
2856	else
2857		ret = RANGE_BOUNDARY_WRITTEN_EXTENT;
2858
2859	free_extent_map(em);
2860	return ret;
2861}
2862
2863static int btrfs_zero_range(struct inode *inode,
2864			    loff_t offset,
2865			    loff_t len,
2866			    const int mode)
2867{
2868	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2869	struct extent_map *em;
2870	struct extent_changeset *data_reserved = NULL;
2871	int ret;
2872	u64 alloc_hint = 0;
2873	const u64 sectorsize = fs_info->sectorsize;
2874	u64 alloc_start = round_down(offset, sectorsize);
2875	u64 alloc_end = round_up(offset + len, sectorsize);
2876	u64 bytes_to_reserve = 0;
2877	bool space_reserved = false;
2878
2879	em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start,
2880			      alloc_end - alloc_start);
 
 
2881	if (IS_ERR(em)) {
2882		ret = PTR_ERR(em);
2883		goto out;
2884	}
2885
2886	/*
2887	 * Avoid hole punching and extent allocation for some cases. More cases
2888	 * could be considered, but these are unlikely common and we keep things
2889	 * as simple as possible for now. Also, intentionally, if the target
2890	 * range contains one or more prealloc extents together with regular
2891	 * extents and holes, we drop all the existing extents and allocate a
2892	 * new prealloc extent, so that we get a larger contiguous disk extent.
2893	 */
2894	if (em->start <= alloc_start && (em->flags & EXTENT_FLAG_PREALLOC)) {
 
2895		const u64 em_end = em->start + em->len;
2896
2897		if (em_end >= offset + len) {
2898			/*
2899			 * The whole range is already a prealloc extent,
2900			 * do nothing except updating the inode's i_size if
2901			 * needed.
2902			 */
2903			free_extent_map(em);
2904			ret = btrfs_fallocate_update_isize(inode, offset + len,
2905							   mode);
2906			goto out;
2907		}
2908		/*
2909		 * Part of the range is already a prealloc extent, so operate
2910		 * only on the remaining part of the range.
2911		 */
2912		alloc_start = em_end;
2913		ASSERT(IS_ALIGNED(alloc_start, sectorsize));
2914		len = offset + len - alloc_start;
2915		offset = alloc_start;
2916		alloc_hint = em->block_start + em->len;
2917	}
2918	free_extent_map(em);
2919
2920	if (BTRFS_BYTES_TO_BLKS(fs_info, offset) ==
2921	    BTRFS_BYTES_TO_BLKS(fs_info, offset + len - 1)) {
2922		em = btrfs_get_extent(BTRFS_I(inode), NULL, alloc_start, sectorsize);
 
2923		if (IS_ERR(em)) {
2924			ret = PTR_ERR(em);
2925			goto out;
2926		}
2927
2928		if (em->flags & EXTENT_FLAG_PREALLOC) {
2929			free_extent_map(em);
2930			ret = btrfs_fallocate_update_isize(inode, offset + len,
2931							   mode);
2932			goto out;
2933		}
2934		if (len < sectorsize && em->block_start != EXTENT_MAP_HOLE) {
2935			free_extent_map(em);
2936			ret = btrfs_truncate_block(BTRFS_I(inode), offset, len,
2937						   0);
2938			if (!ret)
2939				ret = btrfs_fallocate_update_isize(inode,
2940								   offset + len,
2941								   mode);
2942			return ret;
2943		}
2944		free_extent_map(em);
2945		alloc_start = round_down(offset, sectorsize);
2946		alloc_end = alloc_start + sectorsize;
2947		goto reserve_space;
2948	}
2949
2950	alloc_start = round_up(offset, sectorsize);
2951	alloc_end = round_down(offset + len, sectorsize);
2952
2953	/*
2954	 * For unaligned ranges, check the pages at the boundaries, they might
2955	 * map to an extent, in which case we need to partially zero them, or
2956	 * they might map to a hole, in which case we need our allocation range
2957	 * to cover them.
2958	 */
2959	if (!IS_ALIGNED(offset, sectorsize)) {
2960		ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2961							    offset);
2962		if (ret < 0)
2963			goto out;
2964		if (ret == RANGE_BOUNDARY_HOLE) {
2965			alloc_start = round_down(offset, sectorsize);
2966			ret = 0;
2967		} else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2968			ret = btrfs_truncate_block(BTRFS_I(inode), offset, 0, 0);
2969			if (ret)
2970				goto out;
2971		} else {
2972			ret = 0;
2973		}
2974	}
2975
2976	if (!IS_ALIGNED(offset + len, sectorsize)) {
2977		ret = btrfs_zero_range_check_range_boundary(BTRFS_I(inode),
2978							    offset + len);
2979		if (ret < 0)
2980			goto out;
2981		if (ret == RANGE_BOUNDARY_HOLE) {
2982			alloc_end = round_up(offset + len, sectorsize);
2983			ret = 0;
2984		} else if (ret == RANGE_BOUNDARY_WRITTEN_EXTENT) {
2985			ret = btrfs_truncate_block(BTRFS_I(inode), offset + len,
2986						   0, 1);
2987			if (ret)
2988				goto out;
2989		} else {
2990			ret = 0;
2991		}
2992	}
2993
2994reserve_space:
2995	if (alloc_start < alloc_end) {
2996		struct extent_state *cached_state = NULL;
2997		const u64 lockstart = alloc_start;
2998		const u64 lockend = alloc_end - 1;
2999
3000		bytes_to_reserve = alloc_end - alloc_start;
3001		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3002						      bytes_to_reserve);
3003		if (ret < 0)
3004			goto out;
3005		space_reserved = true;
3006		btrfs_punch_hole_lock_range(inode, lockstart, lockend,
3007					    &cached_state);
3008		ret = btrfs_qgroup_reserve_data(BTRFS_I(inode), &data_reserved,
3009						alloc_start, bytes_to_reserve);
3010		if (ret) {
3011			unlock_extent(&BTRFS_I(inode)->io_tree, lockstart,
3012				      lockend, &cached_state);
 
 
3013			goto out;
3014		}
3015		ret = btrfs_prealloc_file_range(inode, mode, alloc_start,
3016						alloc_end - alloc_start,
3017						fs_info->sectorsize,
3018						offset + len, &alloc_hint);
3019		unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
3020			      &cached_state);
3021		/* btrfs_prealloc_file_range releases reserved space on error */
3022		if (ret) {
3023			space_reserved = false;
3024			goto out;
3025		}
3026	}
3027	ret = btrfs_fallocate_update_isize(inode, offset + len, mode);
3028 out:
3029	if (ret && space_reserved)
3030		btrfs_free_reserved_data_space(BTRFS_I(inode), data_reserved,
3031					       alloc_start, bytes_to_reserve);
3032	extent_changeset_free(data_reserved);
3033
3034	return ret;
3035}
3036
3037static long btrfs_fallocate(struct file *file, int mode,
3038			    loff_t offset, loff_t len)
3039{
3040	struct inode *inode = file_inode(file);
3041	struct extent_state *cached_state = NULL;
3042	struct extent_changeset *data_reserved = NULL;
3043	struct falloc_range *range;
3044	struct falloc_range *tmp;
3045	LIST_HEAD(reserve_list);
3046	u64 cur_offset;
3047	u64 last_byte;
3048	u64 alloc_start;
3049	u64 alloc_end;
3050	u64 alloc_hint = 0;
3051	u64 locked_end;
3052	u64 actual_end = 0;
3053	u64 data_space_needed = 0;
3054	u64 data_space_reserved = 0;
3055	u64 qgroup_reserved = 0;
3056	struct extent_map *em;
3057	int blocksize = BTRFS_I(inode)->root->fs_info->sectorsize;
3058	int ret;
3059
3060	/* Do not allow fallocate in ZONED mode */
3061	if (btrfs_is_zoned(inode_to_fs_info(inode)))
3062		return -EOPNOTSUPP;
3063
3064	alloc_start = round_down(offset, blocksize);
3065	alloc_end = round_up(offset + len, blocksize);
3066	cur_offset = alloc_start;
3067
3068	/* Make sure we aren't being give some crap mode */
3069	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
3070		     FALLOC_FL_ZERO_RANGE))
3071		return -EOPNOTSUPP;
3072
3073	if (mode & FALLOC_FL_PUNCH_HOLE)
3074		return btrfs_punch_hole(file, offset, len);
3075
3076	btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
 
 
 
 
 
 
 
 
 
 
 
 
3077
3078	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) {
3079		ret = inode_newsize_ok(inode, offset + len);
3080		if (ret)
3081			goto out;
3082	}
3083
3084	ret = file_modified(file);
3085	if (ret)
3086		goto out;
3087
3088	/*
3089	 * TODO: Move these two operations after we have checked
3090	 * accurate reserved space, or fallocate can still fail but
3091	 * with page truncated or size expanded.
3092	 *
3093	 * But that's a minor problem and won't do much harm BTW.
3094	 */
3095	if (alloc_start > inode->i_size) {
3096		ret = btrfs_cont_expand(BTRFS_I(inode), i_size_read(inode),
3097					alloc_start);
3098		if (ret)
3099			goto out;
3100	} else if (offset + len > inode->i_size) {
3101		/*
3102		 * If we are fallocating from the end of the file onward we
3103		 * need to zero out the end of the block if i_size lands in the
3104		 * middle of a block.
3105		 */
3106		ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
3107		if (ret)
3108			goto out;
3109	}
3110
3111	/*
3112	 * We have locked the inode at the VFS level (in exclusive mode) and we
3113	 * have locked the i_mmap_lock lock (in exclusive mode). Now before
3114	 * locking the file range, flush all dealloc in the range and wait for
3115	 * all ordered extents in the range to complete. After this we can lock
3116	 * the file range and, due to the previous locking we did, we know there
3117	 * can't be more delalloc or ordered extents in the range.
3118	 */
3119	ret = btrfs_wait_ordered_range(inode, alloc_start,
3120				       alloc_end - alloc_start);
3121	if (ret)
3122		goto out;
3123
3124	if (mode & FALLOC_FL_ZERO_RANGE) {
3125		ret = btrfs_zero_range(inode, offset, len, mode);
3126		btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
3127		return ret;
3128	}
3129
3130	locked_end = alloc_end - 1;
3131	lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3132		    &cached_state);
3133
3134	btrfs_assert_inode_range_clean(BTRFS_I(inode), alloc_start, locked_end);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3135
3136	/* First, check if we exceed the qgroup limit */
 
3137	while (cur_offset < alloc_end) {
3138		em = btrfs_get_extent(BTRFS_I(inode), NULL, cur_offset,
3139				      alloc_end - cur_offset);
3140		if (IS_ERR(em)) {
3141			ret = PTR_ERR(em);
3142			break;
3143		}
3144		last_byte = min(extent_map_end(em), alloc_end);
3145		actual_end = min_t(u64, extent_map_end(em), offset + len);
3146		last_byte = ALIGN(last_byte, blocksize);
3147		if (em->block_start == EXTENT_MAP_HOLE ||
3148		    (cur_offset >= inode->i_size &&
3149		     !(em->flags & EXTENT_FLAG_PREALLOC))) {
3150			const u64 range_len = last_byte - cur_offset;
3151
3152			ret = add_falloc_range(&reserve_list, cur_offset, range_len);
3153			if (ret < 0) {
3154				free_extent_map(em);
3155				break;
3156			}
3157			ret = btrfs_qgroup_reserve_data(BTRFS_I(inode),
3158					&data_reserved, cur_offset, range_len);
3159			if (ret < 0) {
 
3160				free_extent_map(em);
3161				break;
3162			}
3163			qgroup_reserved += range_len;
3164			data_space_needed += range_len;
 
 
 
 
 
 
3165		}
3166		free_extent_map(em);
3167		cur_offset = last_byte;
3168	}
3169
3170	if (!ret && data_space_needed > 0) {
3171		/*
3172		 * We are safe to reserve space here as we can't have delalloc
3173		 * in the range, see above.
3174		 */
3175		ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode),
3176						      data_space_needed);
3177		if (!ret)
3178			data_space_reserved = data_space_needed;
3179	}
3180
3181	/*
3182	 * If ret is still 0, means we're OK to fallocate.
3183	 * Or just cleanup the list and exit.
3184	 */
3185	list_for_each_entry_safe(range, tmp, &reserve_list, list) {
3186		if (!ret) {
3187			ret = btrfs_prealloc_file_range(inode, mode,
3188					range->start,
3189					range->len, blocksize,
3190					offset + len, &alloc_hint);
3191			/*
3192			 * btrfs_prealloc_file_range() releases space even
3193			 * if it returns an error.
3194			 */
3195			data_space_reserved -= range->len;
3196			qgroup_reserved -= range->len;
3197		} else if (data_space_reserved > 0) {
3198			btrfs_free_reserved_data_space(BTRFS_I(inode),
3199					       data_reserved, range->start,
3200					       range->len);
3201			data_space_reserved -= range->len;
3202			qgroup_reserved -= range->len;
3203		} else if (qgroup_reserved > 0) {
3204			btrfs_qgroup_free_data(BTRFS_I(inode), data_reserved,
3205					       range->start, range->len, NULL);
3206			qgroup_reserved -= range->len;
3207		}
3208		list_del(&range->list);
3209		kfree(range);
3210	}
3211	if (ret < 0)
3212		goto out_unlock;
3213
3214	/*
3215	 * We didn't need to allocate any more space, but we still extended the
3216	 * size of the file so we need to update i_size and the inode item.
3217	 */
3218	ret = btrfs_fallocate_update_isize(inode, actual_end, mode);
3219out_unlock:
3220	unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
3221		      &cached_state);
3222out:
3223	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_MMAP);
 
 
 
 
3224	extent_changeset_free(data_reserved);
3225	return ret;
3226}
3227
3228/*
3229 * Helper for btrfs_find_delalloc_in_range(). Find a subrange in a given range
3230 * that has unflushed and/or flushing delalloc. There might be other adjacent
3231 * subranges after the one it found, so btrfs_find_delalloc_in_range() keeps
3232 * looping while it gets adjacent subranges, and merging them together.
3233 */
3234static bool find_delalloc_subrange(struct btrfs_inode *inode, u64 start, u64 end,
3235				   struct extent_state **cached_state,
3236				   bool *search_io_tree,
3237				   u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3238{
3239	u64 len = end + 1 - start;
3240	u64 delalloc_len = 0;
3241	struct btrfs_ordered_extent *oe;
3242	u64 oe_start;
3243	u64 oe_end;
3244
3245	/*
3246	 * Search the io tree first for EXTENT_DELALLOC. If we find any, it
3247	 * means we have delalloc (dirty pages) for which writeback has not
3248	 * started yet.
3249	 */
3250	if (*search_io_tree) {
3251		spin_lock(&inode->lock);
3252		if (inode->delalloc_bytes > 0) {
3253			spin_unlock(&inode->lock);
3254			*delalloc_start_ret = start;
3255			delalloc_len = count_range_bits(&inode->io_tree,
3256							delalloc_start_ret, end,
3257							len, EXTENT_DELALLOC, 1,
3258							cached_state);
3259		} else {
3260			spin_unlock(&inode->lock);
3261		}
3262	}
3263
3264	if (delalloc_len > 0) {
3265		/*
3266		 * If delalloc was found then *delalloc_start_ret has a sector size
3267		 * aligned value (rounded down).
3268		 */
3269		*delalloc_end_ret = *delalloc_start_ret + delalloc_len - 1;
3270
3271		if (*delalloc_start_ret == start) {
3272			/* Delalloc for the whole range, nothing more to do. */
3273			if (*delalloc_end_ret == end)
3274				return true;
3275			/* Else trim our search range for ordered extents. */
3276			start = *delalloc_end_ret + 1;
3277			len = end + 1 - start;
3278		}
3279	} else {
3280		/* No delalloc, future calls don't need to search again. */
3281		*search_io_tree = false;
3282	}
3283
3284	/*
3285	 * Now also check if there's any ordered extent in the range.
3286	 * We do this because:
3287	 *
3288	 * 1) When delalloc is flushed, the file range is locked, we clear the
3289	 *    EXTENT_DELALLOC bit from the io tree and create an extent map and
3290	 *    an ordered extent for the write. So we might just have been called
3291	 *    after delalloc is flushed and before the ordered extent completes
3292	 *    and inserts the new file extent item in the subvolume's btree;
3293	 *
3294	 * 2) We may have an ordered extent created by flushing delalloc for a
3295	 *    subrange that starts before the subrange we found marked with
3296	 *    EXTENT_DELALLOC in the io tree.
3297	 *
3298	 * We could also use the extent map tree to find such delalloc that is
3299	 * being flushed, but using the ordered extents tree is more efficient
3300	 * because it's usually much smaller as ordered extents are removed from
3301	 * the tree once they complete. With the extent maps, we mau have them
3302	 * in the extent map tree for a very long time, and they were either
3303	 * created by previous writes or loaded by read operations.
3304	 */
3305	oe = btrfs_lookup_first_ordered_range(inode, start, len);
3306	if (!oe)
3307		return (delalloc_len > 0);
3308
3309	/* The ordered extent may span beyond our search range. */
3310	oe_start = max(oe->file_offset, start);
3311	oe_end = min(oe->file_offset + oe->num_bytes - 1, end);
3312
3313	btrfs_put_ordered_extent(oe);
3314
3315	/* Don't have unflushed delalloc, return the ordered extent range. */
3316	if (delalloc_len == 0) {
3317		*delalloc_start_ret = oe_start;
3318		*delalloc_end_ret = oe_end;
3319		return true;
3320	}
3321
3322	/*
3323	 * We have both unflushed delalloc (io_tree) and an ordered extent.
3324	 * If the ranges are adjacent returned a combined range, otherwise
3325	 * return the leftmost range.
3326	 */
3327	if (oe_start < *delalloc_start_ret) {
3328		if (oe_end < *delalloc_start_ret)
3329			*delalloc_end_ret = oe_end;
3330		*delalloc_start_ret = oe_start;
3331	} else if (*delalloc_end_ret + 1 == oe_start) {
3332		*delalloc_end_ret = oe_end;
3333	}
3334
3335	return true;
3336}
3337
3338/*
3339 * Check if there's delalloc in a given range.
3340 *
3341 * @inode:               The inode.
3342 * @start:               The start offset of the range. It does not need to be
3343 *                       sector size aligned.
3344 * @end:                 The end offset (inclusive value) of the search range.
3345 *                       It does not need to be sector size aligned.
3346 * @cached_state:        Extent state record used for speeding up delalloc
3347 *                       searches in the inode's io_tree. Can be NULL.
3348 * @delalloc_start_ret:  Output argument, set to the start offset of the
3349 *                       subrange found with delalloc (may not be sector size
3350 *                       aligned).
3351 * @delalloc_end_ret:    Output argument, set to he end offset (inclusive value)
3352 *                       of the subrange found with delalloc.
3353 *
3354 * Returns true if a subrange with delalloc is found within the given range, and
3355 * if so it sets @delalloc_start_ret and @delalloc_end_ret with the start and
3356 * end offsets of the subrange.
3357 */
3358bool btrfs_find_delalloc_in_range(struct btrfs_inode *inode, u64 start, u64 end,
3359				  struct extent_state **cached_state,
3360				  u64 *delalloc_start_ret, u64 *delalloc_end_ret)
3361{
3362	u64 cur_offset = round_down(start, inode->root->fs_info->sectorsize);
3363	u64 prev_delalloc_end = 0;
3364	bool search_io_tree = true;
3365	bool ret = false;
3366
3367	while (cur_offset <= end) {
3368		u64 delalloc_start;
3369		u64 delalloc_end;
3370		bool delalloc;
3371
3372		delalloc = find_delalloc_subrange(inode, cur_offset, end,
3373						  cached_state, &search_io_tree,
3374						  &delalloc_start,
3375						  &delalloc_end);
3376		if (!delalloc)
3377			break;
3378
3379		if (prev_delalloc_end == 0) {
3380			/* First subrange found. */
3381			*delalloc_start_ret = max(delalloc_start, start);
3382			*delalloc_end_ret = delalloc_end;
3383			ret = true;
3384		} else if (delalloc_start == prev_delalloc_end + 1) {
3385			/* Subrange adjacent to the previous one, merge them. */
3386			*delalloc_end_ret = delalloc_end;
3387		} else {
3388			/* Subrange not adjacent to the previous one, exit. */
3389			break;
3390		}
3391
3392		prev_delalloc_end = delalloc_end;
3393		cur_offset = delalloc_end + 1;
3394		cond_resched();
3395	}
3396
3397	return ret;
3398}
3399
3400/*
3401 * Check if there's a hole or delalloc range in a range representing a hole (or
3402 * prealloc extent) found in the inode's subvolume btree.
3403 *
3404 * @inode:      The inode.
3405 * @whence:     Seek mode (SEEK_DATA or SEEK_HOLE).
3406 * @start:      Start offset of the hole region. It does not need to be sector
3407 *              size aligned.
3408 * @end:        End offset (inclusive value) of the hole region. It does not
3409 *              need to be sector size aligned.
3410 * @start_ret:  Return parameter, used to set the start of the subrange in the
3411 *              hole that matches the search criteria (seek mode), if such
3412 *              subrange is found (return value of the function is true).
3413 *              The value returned here may not be sector size aligned.
3414 *
3415 * Returns true if a subrange matching the given seek mode is found, and if one
3416 * is found, it updates @start_ret with the start of the subrange.
3417 */
3418static bool find_desired_extent_in_hole(struct btrfs_inode *inode, int whence,
3419					struct extent_state **cached_state,
3420					u64 start, u64 end, u64 *start_ret)
3421{
3422	u64 delalloc_start;
3423	u64 delalloc_end;
3424	bool delalloc;
3425
3426	delalloc = btrfs_find_delalloc_in_range(inode, start, end, cached_state,
3427						&delalloc_start, &delalloc_end);
3428	if (delalloc && whence == SEEK_DATA) {
3429		*start_ret = delalloc_start;
3430		return true;
3431	}
3432
3433	if (delalloc && whence == SEEK_HOLE) {
3434		/*
3435		 * We found delalloc but it starts after out start offset. So we
3436		 * have a hole between our start offset and the delalloc start.
3437		 */
3438		if (start < delalloc_start) {
3439			*start_ret = start;
3440			return true;
3441		}
3442		/*
3443		 * Delalloc range starts at our start offset.
3444		 * If the delalloc range's length is smaller than our range,
3445		 * then it means we have a hole that starts where the delalloc
3446		 * subrange ends.
3447		 */
3448		if (delalloc_end < end) {
3449			*start_ret = delalloc_end + 1;
3450			return true;
3451		}
3452
3453		/* There's delalloc for the whole range. */
3454		return false;
3455	}
3456
3457	if (!delalloc && whence == SEEK_HOLE) {
3458		*start_ret = start;
3459		return true;
3460	}
3461
3462	/*
3463	 * No delalloc in the range and we are seeking for data. The caller has
3464	 * to iterate to the next extent item in the subvolume btree.
3465	 */
3466	return false;
3467}
3468
3469static loff_t find_desired_extent(struct file *file, loff_t offset, int whence)
3470{
3471	struct btrfs_inode *inode = BTRFS_I(file->f_mapping->host);
3472	struct btrfs_file_private *private = file->private_data;
3473	struct btrfs_fs_info *fs_info = inode->root->fs_info;
3474	struct extent_state *cached_state = NULL;
3475	struct extent_state **delalloc_cached_state;
3476	const loff_t i_size = i_size_read(&inode->vfs_inode);
3477	const u64 ino = btrfs_ino(inode);
3478	struct btrfs_root *root = inode->root;
3479	struct btrfs_path *path;
3480	struct btrfs_key key;
3481	u64 last_extent_end;
3482	u64 lockstart;
3483	u64 lockend;
3484	u64 start;
3485	int ret;
3486	bool found = false;
3487
3488	if (i_size == 0 || offset >= i_size)
3489		return -ENXIO;
3490
3491	/*
3492	 * Quick path. If the inode has no prealloc extents and its number of
3493	 * bytes used matches its i_size, then it can not have holes.
3494	 */
3495	if (whence == SEEK_HOLE &&
3496	    !(inode->flags & BTRFS_INODE_PREALLOC) &&
3497	    inode_get_bytes(&inode->vfs_inode) == i_size)
3498		return i_size;
3499
3500	if (!private) {
3501		private = kzalloc(sizeof(*private), GFP_KERNEL);
3502		/*
3503		 * No worries if memory allocation failed.
3504		 * The private structure is used only for speeding up multiple
3505		 * lseek SEEK_HOLE/DATA calls to a file when there's delalloc,
3506		 * so everything will still be correct.
3507		 */
3508		file->private_data = private;
3509	}
3510
3511	if (private)
3512		delalloc_cached_state = &private->llseek_cached_state;
3513	else
3514		delalloc_cached_state = NULL;
3515
3516	/*
3517	 * offset can be negative, in this case we start finding DATA/HOLE from
3518	 * the very start of the file.
3519	 */
3520	start = max_t(loff_t, 0, offset);
3521
3522	lockstart = round_down(start, fs_info->sectorsize);
3523	lockend = round_up(i_size, fs_info->sectorsize);
 
3524	if (lockend <= lockstart)
3525		lockend = lockstart + fs_info->sectorsize;
3526	lockend--;
 
3527
3528	path = btrfs_alloc_path();
3529	if (!path)
3530		return -ENOMEM;
3531	path->reada = READA_FORWARD;
3532
3533	key.objectid = ino;
3534	key.type = BTRFS_EXTENT_DATA_KEY;
3535	key.offset = start;
3536
3537	last_extent_end = lockstart;
3538
3539	lock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3540
3541	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3542	if (ret < 0) {
3543		goto out;
3544	} else if (ret > 0 && path->slots[0] > 0) {
3545		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1);
3546		if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY)
3547			path->slots[0]--;
3548	}
3549
3550	while (start < i_size) {
3551		struct extent_buffer *leaf = path->nodes[0];
3552		struct btrfs_file_extent_item *extent;
3553		u64 extent_end;
3554		u8 type;
3555
3556		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3557			ret = btrfs_next_leaf(root, path);
3558			if (ret < 0)
3559				goto out;
3560			else if (ret > 0)
3561				break;
3562
3563			leaf = path->nodes[0];
3564		}
3565
3566		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3567		if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
 
 
 
 
 
3568			break;
3569
3570		extent_end = btrfs_file_extent_end(path);
3571
3572		/*
3573		 * In the first iteration we may have a slot that points to an
3574		 * extent that ends before our start offset, so skip it.
3575		 */
3576		if (extent_end <= start) {
3577			path->slots[0]++;
3578			continue;
3579		}
3580
3581		/* We have an implicit hole, NO_HOLES feature is likely set. */
3582		if (last_extent_end < key.offset) {
3583			u64 search_start = last_extent_end;
3584			u64 found_start;
3585
3586			/*
3587			 * First iteration, @start matches @offset and it's
3588			 * within the hole.
3589			 */
3590			if (start == offset)
3591				search_start = offset;
3592
3593			found = find_desired_extent_in_hole(inode, whence,
3594							    delalloc_cached_state,
3595							    search_start,
3596							    key.offset - 1,
3597							    &found_start);
3598			if (found) {
3599				start = found_start;
3600				break;
3601			}
3602			/*
3603			 * Didn't find data or a hole (due to delalloc) in the
3604			 * implicit hole range, so need to analyze the extent.
3605			 */
3606		}
3607
3608		extent = btrfs_item_ptr(leaf, path->slots[0],
3609					struct btrfs_file_extent_item);
3610		type = btrfs_file_extent_type(leaf, extent);
3611
3612		/*
3613		 * Can't access the extent's disk_bytenr field if this is an
3614		 * inline extent, since at that offset, it's where the extent
3615		 * data starts.
3616		 */
3617		if (type == BTRFS_FILE_EXTENT_PREALLOC ||
3618		    (type == BTRFS_FILE_EXTENT_REG &&
3619		     btrfs_file_extent_disk_bytenr(leaf, extent) == 0)) {
3620			/*
3621			 * Explicit hole or prealloc extent, search for delalloc.
3622			 * A prealloc extent is treated like a hole.
3623			 */
3624			u64 search_start = key.offset;
3625			u64 found_start;
3626
3627			/*
3628			 * First iteration, @start matches @offset and it's
3629			 * within the hole.
3630			 */
3631			if (start == offset)
3632				search_start = offset;
3633
3634			found = find_desired_extent_in_hole(inode, whence,
3635							    delalloc_cached_state,
3636							    search_start,
3637							    extent_end - 1,
3638							    &found_start);
3639			if (found) {
3640				start = found_start;
3641				break;
3642			}
3643			/*
3644			 * Didn't find data or a hole (due to delalloc) in the
3645			 * implicit hole range, so need to analyze the next
3646			 * extent item.
3647			 */
3648		} else {
3649			/*
3650			 * Found a regular or inline extent.
3651			 * If we are seeking for data, adjust the start offset
3652			 * and stop, we're done.
3653			 */
3654			if (whence == SEEK_DATA) {
3655				start = max_t(u64, key.offset, offset);
3656				found = true;
3657				break;
3658			}
3659			/*
3660			 * Else, we are seeking for a hole, check the next file
3661			 * extent item.
3662			 */
3663		}
3664
3665		start = extent_end;
3666		last_extent_end = extent_end;
3667		path->slots[0]++;
3668		if (fatal_signal_pending(current)) {
3669			ret = -EINTR;
3670			goto out;
3671		}
3672		cond_resched();
3673	}
3674
3675	/* We have an implicit hole from the last extent found up to i_size. */
3676	if (!found && start < i_size) {
3677		found = find_desired_extent_in_hole(inode, whence,
3678						    delalloc_cached_state, start,
3679						    i_size - 1, &start);
3680		if (!found)
3681			start = i_size;
3682	}
3683
3684out:
3685	unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state);
3686	btrfs_free_path(path);
3687
3688	if (ret < 0)
3689		return ret;
3690
3691	if (whence == SEEK_DATA && start >= i_size)
3692		return -ENXIO;
3693
3694	return min_t(loff_t, start, i_size);
3695}
3696
3697static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence)
3698{
3699	struct inode *inode = file->f_mapping->host;
 
3700
 
3701	switch (whence) {
3702	default:
3703		return generic_file_llseek(file, offset, whence);
 
 
3704	case SEEK_DATA:
3705	case SEEK_HOLE:
3706		btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3707		offset = find_desired_extent(file, offset, whence);
3708		btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3709		break;
 
 
 
 
 
 
3710	}
3711
3712	if (offset < 0)
3713		return offset;
3714
3715	return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
3716}
3717
3718static int btrfs_file_open(struct inode *inode, struct file *filp)
3719{
3720	int ret;
3721
3722	filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC | FMODE_BUF_WASYNC |
3723		        FMODE_CAN_ODIRECT;
3724
3725	ret = fsverity_file_open(inode, filp);
3726	if (ret)
3727		return ret;
3728	return generic_file_open(inode, filp);
3729}
3730
3731static int check_direct_read(struct btrfs_fs_info *fs_info,
3732			     const struct iov_iter *iter, loff_t offset)
3733{
3734	int ret;
3735	int i, seg;
3736
3737	ret = check_direct_IO(fs_info, iter, offset);
3738	if (ret < 0)
3739		return ret;
3740
3741	if (!iter_is_iovec(iter))
3742		return 0;
3743
3744	for (seg = 0; seg < iter->nr_segs; seg++) {
3745		for (i = seg + 1; i < iter->nr_segs; i++) {
3746			const struct iovec *iov1 = iter_iov(iter) + seg;
3747			const struct iovec *iov2 = iter_iov(iter) + i;
3748
3749			if (iov1->iov_base == iov2->iov_base)
3750				return -EINVAL;
3751		}
3752	}
3753	return 0;
3754}
3755
3756static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
3757{
3758	struct inode *inode = file_inode(iocb->ki_filp);
3759	size_t prev_left = 0;
3760	ssize_t read = 0;
3761	ssize_t ret;
3762
3763	if (fsverity_active(inode))
3764		return 0;
3765
3766	if (check_direct_read(inode_to_fs_info(inode), to, iocb->ki_pos))
3767		return 0;
3768
3769	btrfs_inode_lock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3770again:
3771	/*
3772	 * This is similar to what we do for direct IO writes, see the comment
3773	 * at btrfs_direct_write(), but we also disable page faults in addition
3774	 * to disabling them only at the iov_iter level. This is because when
3775	 * reading from a hole or prealloc extent, iomap calls iov_iter_zero(),
3776	 * which can still trigger page fault ins despite having set ->nofault
3777	 * to true of our 'to' iov_iter.
3778	 *
3779	 * The difference to direct IO writes is that we deadlock when trying
3780	 * to lock the extent range in the inode's tree during he page reads
3781	 * triggered by the fault in (while for writes it is due to waiting for
3782	 * our own ordered extent). This is because for direct IO reads,
3783	 * btrfs_dio_iomap_begin() returns with the extent range locked, which
3784	 * is only unlocked in the endio callback (end_bio_extent_readpage()).
3785	 */
3786	pagefault_disable();
3787	to->nofault = true;
3788	ret = btrfs_dio_read(iocb, to, read);
3789	to->nofault = false;
3790	pagefault_enable();
3791
3792	/* No increment (+=) because iomap returns a cumulative value. */
3793	if (ret > 0)
3794		read = ret;
3795
3796	if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
3797		const size_t left = iov_iter_count(to);
3798
3799		if (left == prev_left) {
3800			/*
3801			 * We didn't make any progress since the last attempt,
3802			 * fallback to a buffered read for the remainder of the
3803			 * range. This is just to avoid any possibility of looping
3804			 * for too long.
3805			 */
3806			ret = read;
3807		} else {
3808			/*
3809			 * We made some progress since the last retry or this is
3810			 * the first time we are retrying. Fault in as many pages
3811			 * as possible and retry.
3812			 */
3813			fault_in_iov_iter_writeable(to, left);
3814			prev_left = left;
3815			goto again;
3816		}
3817	}
3818	btrfs_inode_unlock(BTRFS_I(inode), BTRFS_ILOCK_SHARED);
3819	return ret < 0 ? ret : read;
3820}
3821
3822static ssize_t btrfs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
3823{
3824	ssize_t ret = 0;
3825
3826	if (iocb->ki_flags & IOCB_DIRECT) {
3827		ret = btrfs_direct_read(iocb, to);
3828		if (ret < 0 || !iov_iter_count(to) ||
3829		    iocb->ki_pos >= i_size_read(file_inode(iocb->ki_filp)))
3830			return ret;
3831	}
3832
3833	return filemap_read(iocb, to, ret);
3834}
3835
3836const struct file_operations btrfs_file_operations = {
3837	.llseek		= btrfs_file_llseek,
3838	.read_iter      = btrfs_file_read_iter,
3839	.splice_read	= filemap_splice_read,
3840	.write_iter	= btrfs_file_write_iter,
3841	.splice_write	= iter_file_splice_write,
3842	.mmap		= btrfs_file_mmap,
3843	.open		= btrfs_file_open,
3844	.release	= btrfs_release_file,
3845	.get_unmapped_area = thp_get_unmapped_area,
3846	.fsync		= btrfs_sync_file,
3847	.fallocate	= btrfs_fallocate,
3848	.unlocked_ioctl	= btrfs_ioctl,
3849#ifdef CONFIG_COMPAT
3850	.compat_ioctl	= btrfs_compat_ioctl,
3851#endif
3852	.remap_file_range = btrfs_remap_file_range,
3853};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3854
3855int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end)
3856{
3857	int ret;
3858
3859	/*
3860	 * So with compression we will find and lock a dirty page and clear the
3861	 * first one as dirty, setup an async extent, and immediately return
3862	 * with the entire range locked but with nobody actually marked with
3863	 * writeback.  So we can't just filemap_write_and_wait_range() and
3864	 * expect it to work since it will just kick off a thread to do the
3865	 * actual work.  So we need to call filemap_fdatawrite_range _again_
3866	 * since it will wait on the page lock, which won't be unlocked until
3867	 * after the pages have been marked as writeback and so we're good to go
3868	 * from there.  We have to do this otherwise we'll miss the ordered
3869	 * extents and that results in badness.  Please Josef, do not think you
3870	 * know better and pull this out at some point in the future, it is
3871	 * right and you are wrong.
3872	 */
3873	ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3874	if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
3875			     &BTRFS_I(inode)->runtime_flags))
3876		ret = filemap_fdatawrite_range(inode->i_mapping, start, end);
3877
3878	return ret;
3879}