Linux Audio

Check our new training course

Loading...
v3.15
 
   1/*
   2 * Copyright (C) 2007 Oracle.  All rights reserved.
   3 *
   4 * This program is free software; you can redistribute it and/or
   5 * modify it under the terms of the GNU General Public
   6 * License v2 as published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11 * General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public
  14 * License along with this program; if not, write to the
  15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16 * Boston, MA 021110-1307, USA.
  17 */
  18
  19#include <linux/slab.h>
  20#include <linux/blkdev.h>
  21#include <linux/writeback.h>
  22#include <linux/pagevec.h>
 
 
  23#include "ctree.h"
  24#include "transaction.h"
  25#include "btrfs_inode.h"
  26#include "extent_io.h"
  27#include "disk-io.h"
 
 
 
 
 
  28
  29static struct kmem_cache *btrfs_ordered_extent_cache;
  30
  31static u64 entry_end(struct btrfs_ordered_extent *entry)
  32{
  33	if (entry->file_offset + entry->len < entry->file_offset)
  34		return (u64)-1;
  35	return entry->file_offset + entry->len;
  36}
  37
  38/* returns NULL if the insertion worked, or it returns the node it did find
  39 * in the tree
  40 */
  41static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
  42				   struct rb_node *node)
  43{
  44	struct rb_node **p = &root->rb_node;
  45	struct rb_node *parent = NULL;
  46	struct btrfs_ordered_extent *entry;
  47
  48	while (*p) {
  49		parent = *p;
  50		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
  51
  52		if (file_offset < entry->file_offset)
  53			p = &(*p)->rb_left;
  54		else if (file_offset >= entry_end(entry))
  55			p = &(*p)->rb_right;
  56		else
  57			return parent;
  58	}
  59
  60	rb_link_node(node, parent, p);
  61	rb_insert_color(node, root);
  62	return NULL;
  63}
  64
  65static void ordered_data_tree_panic(struct inode *inode, int errno,
  66					       u64 offset)
  67{
  68	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
  69	btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset "
  70		    "%llu\n", offset);
  71}
  72
  73/*
  74 * look for a given offset in the tree, and if it can't be found return the
  75 * first lesser offset
  76 */
  77static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
  78				     struct rb_node **prev_ret)
  79{
  80	struct rb_node *n = root->rb_node;
  81	struct rb_node *prev = NULL;
  82	struct rb_node *test;
  83	struct btrfs_ordered_extent *entry;
  84	struct btrfs_ordered_extent *prev_entry = NULL;
  85
  86	while (n) {
  87		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
  88		prev = n;
  89		prev_entry = entry;
  90
  91		if (file_offset < entry->file_offset)
  92			n = n->rb_left;
  93		else if (file_offset >= entry_end(entry))
  94			n = n->rb_right;
  95		else
  96			return n;
  97	}
  98	if (!prev_ret)
  99		return NULL;
 100
 101	while (prev && file_offset >= entry_end(prev_entry)) {
 102		test = rb_next(prev);
 103		if (!test)
 104			break;
 105		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 106				      rb_node);
 107		if (file_offset < entry_end(prev_entry))
 108			break;
 109
 110		prev = test;
 111	}
 112	if (prev)
 113		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
 114				      rb_node);
 115	while (prev && file_offset < entry_end(prev_entry)) {
 116		test = rb_prev(prev);
 117		if (!test)
 118			break;
 119		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 120				      rb_node);
 121		prev = test;
 122	}
 123	*prev_ret = prev;
 124	return NULL;
 125}
 126
 127/*
 128 * helper to check if a given offset is inside a given entry
 129 */
 130static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset)
 131{
 132	if (file_offset < entry->file_offset ||
 133	    entry->file_offset + entry->len <= file_offset)
 134		return 0;
 135	return 1;
 136}
 137
 138static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
 139			  u64 len)
 140{
 141	if (file_offset + len <= entry->file_offset ||
 142	    entry->file_offset + entry->len <= file_offset)
 143		return 0;
 144	return 1;
 145}
 146
 147/*
 148 * look find the first ordered struct that has this offset, otherwise
 149 * the first one less than this offset
 150 */
 151static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree,
 152					  u64 file_offset)
 153{
 154	struct rb_root *root = &tree->tree;
 155	struct rb_node *prev = NULL;
 156	struct rb_node *ret;
 157	struct btrfs_ordered_extent *entry;
 158
 159	if (tree->last) {
 160		entry = rb_entry(tree->last, struct btrfs_ordered_extent,
 161				 rb_node);
 162		if (offset_in_entry(entry, file_offset))
 163			return tree->last;
 164	}
 165	ret = __tree_search(root, file_offset, &prev);
 166	if (!ret)
 167		ret = prev;
 168	if (ret)
 169		tree->last = ret;
 170	return ret;
 171}
 172
 173/* allocate and add a new ordered_extent into the per-inode tree.
 174 * file_offset is the logical offset in the file
 175 *
 176 * start is the disk block number of an extent already reserved in the
 177 * extent allocation tree
 178 *
 179 * len is the length of the extent
 180 *
 181 * The tree is given a single reference on the ordered extent that was
 182 * inserted.
 183 */
 184static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 185				      u64 start, u64 len, u64 disk_len,
 186				      int type, int dio, int compress_type)
 187{
 188	struct btrfs_root *root = BTRFS_I(inode)->root;
 189	struct btrfs_ordered_inode_tree *tree;
 190	struct rb_node *node;
 191	struct btrfs_ordered_extent *entry;
 
 
 192
 193	tree = &BTRFS_I(inode)->ordered_tree;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 194	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
 195	if (!entry)
 196		return -ENOMEM;
 197
 198	entry->file_offset = file_offset;
 199	entry->start = start;
 200	entry->len = len;
 201	if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) &&
 202	    !(type == BTRFS_ORDERED_NOCOW))
 203		entry->csum_bytes_left = disk_len;
 204	entry->disk_len = disk_len;
 205	entry->bytes_left = len;
 206	entry->inode = igrab(inode);
 207	entry->compress_type = compress_type;
 208	entry->truncated_len = (u64)-1;
 209	if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE)
 210		set_bit(type, &entry->flags);
 211
 212	if (dio)
 213		set_bit(BTRFS_ORDERED_DIRECT, &entry->flags);
 214
 215	/* one ref for the tree */
 216	atomic_set(&entry->refs, 1);
 217	init_waitqueue_head(&entry->wait);
 218	INIT_LIST_HEAD(&entry->list);
 
 219	INIT_LIST_HEAD(&entry->root_extent_list);
 220	INIT_LIST_HEAD(&entry->work_list);
 
 221	init_completion(&entry->completion);
 222	INIT_LIST_HEAD(&entry->log_list);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 223
 224	trace_btrfs_ordered_extent_add(inode, entry);
 225
 226	spin_lock_irq(&tree->lock);
 227	node = tree_insert(&tree->tree, file_offset,
 
 
 
 
 
 
 228			   &entry->rb_node);
 229	if (node)
 230		ordered_data_tree_panic(inode, -EEXIST, file_offset);
 231	spin_unlock_irq(&tree->lock);
 
 
 232
 233	spin_lock(&root->ordered_extent_lock);
 234	list_add_tail(&entry->root_extent_list,
 235		      &root->ordered_extents);
 236	root->nr_ordered_extents++;
 237	if (root->nr_ordered_extents == 1) {
 238		spin_lock(&root->fs_info->ordered_root_lock);
 239		BUG_ON(!list_empty(&root->ordered_root));
 240		list_add_tail(&root->ordered_root,
 241			      &root->fs_info->ordered_roots);
 242		spin_unlock(&root->fs_info->ordered_root_lock);
 243	}
 244	spin_unlock(&root->ordered_extent_lock);
 245
 246	return 0;
 247}
 248
 249int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 250			     u64 start, u64 len, u64 disk_len, int type)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 251{
 252	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 253					  disk_len, type, 0,
 254					  BTRFS_COMPRESS_NONE);
 255}
 256
 257int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset,
 258				 u64 start, u64 len, u64 disk_len, int type)
 259{
 260	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 261					  disk_len, type, 1,
 262					  BTRFS_COMPRESS_NONE);
 263}
 264
 265int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset,
 266				      u64 start, u64 len, u64 disk_len,
 267				      int type, int compress_type)
 268{
 269	return __btrfs_add_ordered_extent(inode, file_offset, start, len,
 270					  disk_len, type, 0,
 271					  compress_type);
 272}
 273
 274/*
 275 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 276 * when an ordered extent is finished.  If the list covers more than one
 277 * ordered extent, it is split across multiples.
 278 */
 279void btrfs_add_ordered_sum(struct inode *inode,
 280			   struct btrfs_ordered_extent *entry,
 281			   struct btrfs_ordered_sum *sum)
 282{
 283	struct btrfs_ordered_inode_tree *tree;
 284
 285	tree = &BTRFS_I(inode)->ordered_tree;
 286	spin_lock_irq(&tree->lock);
 287	list_add_tail(&sum->list, &entry->list);
 288	WARN_ON(entry->csum_bytes_left < sum->len);
 289	entry->csum_bytes_left -= sum->len;
 290	if (entry->csum_bytes_left == 0)
 291		wake_up(&entry->wait);
 292	spin_unlock_irq(&tree->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 293}
 294
 295/*
 296 * this is used to account for finished IO across a given range
 297 * of the file.  The IO may span ordered extents.  If
 298 * a given ordered_extent is completely done, 1 is returned, otherwise
 299 * 0.
 300 *
 301 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 302 * to make sure this function only returns 1 once for a given ordered extent.
 
 
 
 
 303 *
 304 * file_offset is updated to one byte past the range that is recorded as
 305 * complete.  This allows you to walk forward in the file.
 306 */
 307int btrfs_dec_test_first_ordered_pending(struct inode *inode,
 308				   struct btrfs_ordered_extent **cached,
 309				   u64 *file_offset, u64 io_size, int uptodate)
 310{
 311	struct btrfs_ordered_inode_tree *tree;
 312	struct rb_node *node;
 313	struct btrfs_ordered_extent *entry = NULL;
 314	int ret;
 315	unsigned long flags;
 316	u64 dec_end;
 317	u64 dec_start;
 318	u64 to_dec;
 319
 320	tree = &BTRFS_I(inode)->ordered_tree;
 321	spin_lock_irqsave(&tree->lock, flags);
 322	node = tree_search(tree, *file_offset);
 323	if (!node) {
 324		ret = 1;
 325		goto out;
 326	}
 327
 328	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 329	if (!offset_in_entry(entry, *file_offset)) {
 330		ret = 1;
 331		goto out;
 332	}
 
 
 
 
 333
 334	dec_start = max(*file_offset, entry->file_offset);
 335	dec_end = min(*file_offset + io_size, entry->file_offset +
 336		      entry->len);
 337	*file_offset = dec_end;
 338	if (dec_start > dec_end) {
 339		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 340			"bad ordering dec_start %llu end %llu", dec_start, dec_end);
 341	}
 342	to_dec = dec_end - dec_start;
 343	if (to_dec > entry->bytes_left) {
 344		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 345			"bad ordered accounting left %llu size %llu",
 346			entry->bytes_left, to_dec);
 347	}
 348	entry->bytes_left -= to_dec;
 349	if (!uptodate)
 350		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
 351
 352	if (entry->bytes_left == 0) {
 353		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 354		if (waitqueue_active(&entry->wait))
 355			wake_up(&entry->wait);
 356	} else {
 357		ret = 1;
 358	}
 359out:
 360	if (!ret && cached && entry) {
 361		*cached = entry;
 362		atomic_inc(&entry->refs);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 363	}
 364	spin_unlock_irqrestore(&tree->lock, flags);
 365	return ret == 0;
 366}
 367
 368/*
 369 * this is used to account for finished IO across a given range
 370 * of the file.  The IO should not span ordered extents.  If
 371 * a given ordered_extent is completely done, 1 is returned, otherwise
 372 * 0.
 
 
 
 
 
 
 
 
 373 *
 374 * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used
 375 * to make sure this function only returns 1 once for a given ordered extent.
 376 */
 377int btrfs_dec_test_ordered_pending(struct inode *inode,
 378				   struct btrfs_ordered_extent **cached,
 379				   u64 file_offset, u64 io_size, int uptodate)
 380{
 381	struct btrfs_ordered_inode_tree *tree;
 382	struct rb_node *node;
 383	struct btrfs_ordered_extent *entry = NULL;
 384	unsigned long flags;
 385	int ret;
 386
 387	tree = &BTRFS_I(inode)->ordered_tree;
 388	spin_lock_irqsave(&tree->lock, flags);
 389	if (cached && *cached) {
 390		entry = *cached;
 391		goto have_entry;
 392	}
 393
 394	node = tree_search(tree, file_offset);
 395	if (!node) {
 396		ret = 1;
 397		goto out;
 398	}
 399
 400	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 401have_entry:
 402	if (!offset_in_entry(entry, file_offset)) {
 403		ret = 1;
 404		goto out;
 405	}
 406
 407	if (io_size > entry->bytes_left) {
 408		btrfs_crit(BTRFS_I(inode)->root->fs_info,
 409			   "bad ordered accounting left %llu size %llu",
 410		       entry->bytes_left, io_size);
 411	}
 412	entry->bytes_left -= io_size;
 413	if (!uptodate)
 414		set_bit(BTRFS_ORDERED_IOERR, &entry->flags);
 415
 416	if (entry->bytes_left == 0) {
 417		ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 418		if (waitqueue_active(&entry->wait))
 419			wake_up(&entry->wait);
 420	} else {
 421		ret = 1;
 
 
 422	}
 423out:
 424	if (!ret && cached && entry) {
 425		*cached = entry;
 426		atomic_inc(&entry->refs);
 427	}
 428	spin_unlock_irqrestore(&tree->lock, flags);
 429	return ret == 0;
 430}
 431
 432/* Needs to either be called under a log transaction or the log_mutex */
 433void btrfs_get_logged_extents(struct inode *inode,
 434			      struct list_head *logged_list)
 435{
 436	struct btrfs_ordered_inode_tree *tree;
 437	struct btrfs_ordered_extent *ordered;
 438	struct rb_node *n;
 439
 440	tree = &BTRFS_I(inode)->ordered_tree;
 441	spin_lock_irq(&tree->lock);
 442	for (n = rb_first(&tree->tree); n; n = rb_next(n)) {
 443		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
 444		if (!list_empty(&ordered->log_list))
 445			continue;
 446		list_add_tail(&ordered->log_list, logged_list);
 447		atomic_inc(&ordered->refs);
 448	}
 449	spin_unlock_irq(&tree->lock);
 450}
 451
 452void btrfs_put_logged_extents(struct list_head *logged_list)
 453{
 454	struct btrfs_ordered_extent *ordered;
 455
 456	while (!list_empty(logged_list)) {
 457		ordered = list_first_entry(logged_list,
 458					   struct btrfs_ordered_extent,
 459					   log_list);
 460		list_del_init(&ordered->log_list);
 461		btrfs_put_ordered_extent(ordered);
 462	}
 463}
 464
 465void btrfs_submit_logged_extents(struct list_head *logged_list,
 466				 struct btrfs_root *log)
 467{
 468	int index = log->log_transid % 2;
 469
 470	spin_lock_irq(&log->log_extents_lock[index]);
 471	list_splice_tail(logged_list, &log->logged_list[index]);
 472	spin_unlock_irq(&log->log_extents_lock[index]);
 473}
 474
 475void btrfs_wait_logged_extents(struct btrfs_root *log, u64 transid)
 476{
 477	struct btrfs_ordered_extent *ordered;
 478	int index = transid % 2;
 479
 480	spin_lock_irq(&log->log_extents_lock[index]);
 481	while (!list_empty(&log->logged_list[index])) {
 482		ordered = list_first_entry(&log->logged_list[index],
 483					   struct btrfs_ordered_extent,
 484					   log_list);
 485		list_del_init(&ordered->log_list);
 486		spin_unlock_irq(&log->log_extents_lock[index]);
 487		wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE,
 488						   &ordered->flags));
 489		btrfs_put_ordered_extent(ordered);
 490		spin_lock_irq(&log->log_extents_lock[index]);
 491	}
 492	spin_unlock_irq(&log->log_extents_lock[index]);
 493}
 494
 495void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid)
 496{
 497	struct btrfs_ordered_extent *ordered;
 498	int index = transid % 2;
 499
 500	spin_lock_irq(&log->log_extents_lock[index]);
 501	while (!list_empty(&log->logged_list[index])) {
 502		ordered = list_first_entry(&log->logged_list[index],
 503					   struct btrfs_ordered_extent,
 504					   log_list);
 505		list_del_init(&ordered->log_list);
 506		spin_unlock_irq(&log->log_extents_lock[index]);
 507		btrfs_put_ordered_extent(ordered);
 508		spin_lock_irq(&log->log_extents_lock[index]);
 509	}
 510	spin_unlock_irq(&log->log_extents_lock[index]);
 
 511}
 512
 513/*
 514 * used to drop a reference on an ordered extent.  This will free
 515 * the extent if the last reference is dropped
 516 */
 517void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 518{
 519	struct list_head *cur;
 520	struct btrfs_ordered_sum *sum;
 521
 522	trace_btrfs_ordered_extent_put(entry->inode, entry);
 523
 524	if (atomic_dec_and_test(&entry->refs)) {
 
 
 
 525		if (entry->inode)
 526			btrfs_add_delayed_iput(entry->inode);
 527		while (!list_empty(&entry->list)) {
 528			cur = entry->list.next;
 529			sum = list_entry(cur, struct btrfs_ordered_sum, list);
 530			list_del(&sum->list);
 531			kfree(sum);
 532		}
 533		kmem_cache_free(btrfs_ordered_extent_cache, entry);
 534	}
 535}
 536
 537/*
 538 * remove an ordered extent from the tree.  No references are dropped
 539 * and waiters are woken up.
 540 */
 541void btrfs_remove_ordered_extent(struct inode *inode,
 542				 struct btrfs_ordered_extent *entry)
 543{
 544	struct btrfs_ordered_inode_tree *tree;
 545	struct btrfs_root *root = BTRFS_I(inode)->root;
 546	struct rb_node *node;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 547
 548	tree = &BTRFS_I(inode)->ordered_tree;
 549	spin_lock_irq(&tree->lock);
 
 
 
 
 
 
 
 
 
 
 
 550	node = &entry->rb_node;
 551	rb_erase(node, &tree->tree);
 552	if (tree->last == node)
 553		tree->last = NULL;
 
 554	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 555	spin_unlock_irq(&tree->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 556
 557	spin_lock(&root->ordered_extent_lock);
 558	list_del_init(&entry->root_extent_list);
 559	root->nr_ordered_extents--;
 560
 561	trace_btrfs_ordered_extent_remove(inode, entry);
 562
 563	/*
 564	 * we have no more ordered extents for this inode and
 565	 * no dirty pages.  We can safely remove it from the
 566	 * list of ordered extents
 567	 */
 568	if (RB_EMPTY_ROOT(&tree->tree) &&
 569	    !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
 570		spin_lock(&root->fs_info->ordered_root_lock);
 571		list_del_init(&BTRFS_I(inode)->ordered_operations);
 572		spin_unlock(&root->fs_info->ordered_root_lock);
 573	}
 574
 575	if (!root->nr_ordered_extents) {
 576		spin_lock(&root->fs_info->ordered_root_lock);
 577		BUG_ON(list_empty(&root->ordered_root));
 578		list_del_init(&root->ordered_root);
 579		spin_unlock(&root->fs_info->ordered_root_lock);
 580	}
 581	spin_unlock(&root->ordered_extent_lock);
 582	wake_up(&entry->wait);
 
 
 583}
 584
 585static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
 586{
 587	struct btrfs_ordered_extent *ordered;
 588
 589	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
 590	btrfs_start_ordered_extent(ordered->inode, ordered, 1);
 591	complete(&ordered->completion);
 592}
 593
 594/*
 595 * wait for all the ordered extents in a root.  This is done when balancing
 596 * space between drives.
 597 */
 598int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr)
 
 599{
 600	struct list_head splice, works;
 
 
 
 601	struct btrfs_ordered_extent *ordered, *next;
 602	int count = 0;
 603
 604	INIT_LIST_HEAD(&splice);
 605	INIT_LIST_HEAD(&works);
 606
 607	mutex_lock(&root->ordered_extent_mutex);
 608	spin_lock(&root->ordered_extent_lock);
 609	list_splice_init(&root->ordered_extents, &splice);
 610	while (!list_empty(&splice) && nr) {
 611		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
 612					   root_extent_list);
 
 
 
 
 
 
 
 
 613		list_move_tail(&ordered->root_extent_list,
 614			       &root->ordered_extents);
 615		atomic_inc(&ordered->refs);
 616		spin_unlock(&root->ordered_extent_lock);
 617
 618		btrfs_init_work(&ordered->flush_work,
 619				btrfs_run_ordered_extent_work, NULL, NULL);
 620		list_add_tail(&ordered->work_list, &works);
 621		btrfs_queue_work(root->fs_info->flush_workers,
 622				 &ordered->flush_work);
 623
 624		cond_resched();
 625		spin_lock(&root->ordered_extent_lock);
 626		if (nr != -1)
 627			nr--;
 628		count++;
 629	}
 
 630	list_splice_tail(&splice, &root->ordered_extents);
 631	spin_unlock(&root->ordered_extent_lock);
 632
 633	list_for_each_entry_safe(ordered, next, &works, work_list) {
 634		list_del_init(&ordered->work_list);
 635		wait_for_completion(&ordered->completion);
 636		btrfs_put_ordered_extent(ordered);
 637		cond_resched();
 638	}
 639	mutex_unlock(&root->ordered_extent_mutex);
 640
 641	return count;
 642}
 643
 644void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr)
 
 645{
 646	struct btrfs_root *root;
 647	struct list_head splice;
 648	int done;
 649
 650	INIT_LIST_HEAD(&splice);
 651
 652	mutex_lock(&fs_info->ordered_operations_mutex);
 653	spin_lock(&fs_info->ordered_root_lock);
 654	list_splice_init(&fs_info->ordered_roots, &splice);
 655	while (!list_empty(&splice) && nr) {
 656		root = list_first_entry(&splice, struct btrfs_root,
 657					ordered_root);
 658		root = btrfs_grab_fs_root(root);
 659		BUG_ON(!root);
 660		list_move_tail(&root->ordered_root,
 661			       &fs_info->ordered_roots);
 662		spin_unlock(&fs_info->ordered_root_lock);
 663
 664		done = btrfs_wait_ordered_extents(root, nr);
 665		btrfs_put_fs_root(root);
 
 666
 667		spin_lock(&fs_info->ordered_root_lock);
 668		if (nr != -1) {
 669			nr -= done;
 670			WARN_ON(nr < 0);
 671		}
 672	}
 673	list_splice_tail(&splice, &fs_info->ordered_roots);
 674	spin_unlock(&fs_info->ordered_root_lock);
 675	mutex_unlock(&fs_info->ordered_operations_mutex);
 676}
 677
 678/*
 679 * this is used during transaction commit to write all the inodes
 680 * added to the ordered operation list.  These files must be fully on
 681 * disk before the transaction commits.
 682 *
 683 * we have two modes here, one is to just start the IO via filemap_flush
 684 * and the other is to wait for all the io.  When we wait, we have an
 685 * extra check to make sure the ordered operation list really is empty
 686 * before we return
 687 */
 688int btrfs_run_ordered_operations(struct btrfs_trans_handle *trans,
 689				 struct btrfs_root *root, int wait)
 690{
 691	struct btrfs_inode *btrfs_inode;
 692	struct inode *inode;
 693	struct btrfs_transaction *cur_trans = trans->transaction;
 694	struct list_head splice;
 695	struct list_head works;
 696	struct btrfs_delalloc_work *work, *next;
 697	int ret = 0;
 698
 699	INIT_LIST_HEAD(&splice);
 700	INIT_LIST_HEAD(&works);
 701
 702	mutex_lock(&root->fs_info->ordered_extent_flush_mutex);
 703	spin_lock(&root->fs_info->ordered_root_lock);
 704	list_splice_init(&cur_trans->ordered_operations, &splice);
 705	while (!list_empty(&splice)) {
 706		btrfs_inode = list_entry(splice.next, struct btrfs_inode,
 707				   ordered_operations);
 708		inode = &btrfs_inode->vfs_inode;
 709
 710		list_del_init(&btrfs_inode->ordered_operations);
 711
 712		/*
 713		 * the inode may be getting freed (in sys_unlink path).
 714		 */
 715		inode = igrab(inode);
 716		if (!inode)
 717			continue;
 718
 719		if (!wait)
 720			list_add_tail(&BTRFS_I(inode)->ordered_operations,
 721				      &cur_trans->ordered_operations);
 722		spin_unlock(&root->fs_info->ordered_root_lock);
 723
 724		work = btrfs_alloc_delalloc_work(inode, wait, 1);
 725		if (!work) {
 726			spin_lock(&root->fs_info->ordered_root_lock);
 727			if (list_empty(&BTRFS_I(inode)->ordered_operations))
 728				list_add_tail(&btrfs_inode->ordered_operations,
 729					      &splice);
 730			list_splice_tail(&splice,
 731					 &cur_trans->ordered_operations);
 732			spin_unlock(&root->fs_info->ordered_root_lock);
 733			ret = -ENOMEM;
 734			goto out;
 735		}
 736		list_add_tail(&work->list, &works);
 737		btrfs_queue_work(root->fs_info->flush_workers,
 738				 &work->work);
 739
 740		cond_resched();
 741		spin_lock(&root->fs_info->ordered_root_lock);
 742	}
 743	spin_unlock(&root->fs_info->ordered_root_lock);
 744out:
 745	list_for_each_entry_safe(work, next, &works, list) {
 746		list_del_init(&work->list);
 747		btrfs_wait_and_free_delalloc_work(work);
 748	}
 749	mutex_unlock(&root->fs_info->ordered_extent_flush_mutex);
 750	return ret;
 751}
 752
 753/*
 754 * Used to start IO or wait for a given ordered extent to finish.
 755 *
 756 * If wait is one, this effectively waits on page writeback for all the pages
 757 * in the extent, and it waits on the io completion code to insert
 758 * metadata into the btree corresponding to the extent
 759 */
 760void btrfs_start_ordered_extent(struct inode *inode,
 761				       struct btrfs_ordered_extent *entry,
 762				       int wait)
 763{
 764	u64 start = entry->file_offset;
 765	u64 end = start + entry->len - 1;
 
 
 766
 767	trace_btrfs_ordered_extent_start(inode, entry);
 768
 769	/*
 
 
 
 
 
 
 770	 * pages in the range can be dirty, clean or writeback.  We
 771	 * start IO on any dirty ones so the wait doesn't stall waiting
 772	 * for the flusher thread to find them
 773	 */
 774	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
 775		filemap_fdatawrite_range(inode->i_mapping, start, end);
 776	if (wait) {
 777		wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE,
 778						 &entry->flags));
 779	}
 780}
 781
 782/*
 783 * Used to wait on ordered extents across a large range of bytes.
 784 */
 785int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 786{
 787	int ret = 0;
 
 788	u64 end;
 789	u64 orig_end;
 790	struct btrfs_ordered_extent *ordered;
 791
 792	if (start + len < start) {
 793		orig_end = INT_LIMIT(loff_t);
 794	} else {
 795		orig_end = start + len - 1;
 796		if (orig_end > INT_LIMIT(loff_t))
 797			orig_end = INT_LIMIT(loff_t);
 798	}
 799
 800	/* start IO across the range first to instantiate any delalloc
 801	 * extents
 802	 */
 803	ret = filemap_fdatawrite_range(inode->i_mapping, start, orig_end);
 804	if (ret)
 805		return ret;
 
 806	/*
 807	 * So with compression we will find and lock a dirty page and clear the
 808	 * first one as dirty, setup an async extent, and immediately return
 809	 * with the entire range locked but with nobody actually marked with
 810	 * writeback.  So we can't just filemap_write_and_wait_range() and
 811	 * expect it to work since it will just kick off a thread to do the
 812	 * actual work.  So we need to call filemap_fdatawrite_range _again_
 813	 * since it will wait on the page lock, which won't be unlocked until
 814	 * after the pages have been marked as writeback and so we're good to go
 815	 * from there.  We have to do this otherwise we'll miss the ordered
 816	 * extents and that results in badness.  Please Josef, do not think you
 817	 * know better and pull this out at some point in the future, it is
 818	 * right and you are wrong.
 819	 */
 820	if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
 821		     &BTRFS_I(inode)->runtime_flags)) {
 822		ret = filemap_fdatawrite_range(inode->i_mapping, start,
 823					       orig_end);
 824		if (ret)
 825			return ret;
 826	}
 827	ret = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 828	if (ret)
 829		return ret;
 830
 831	end = orig_end;
 832	while (1) {
 833		ordered = btrfs_lookup_first_ordered_extent(inode, end);
 834		if (!ordered)
 835			break;
 836		if (ordered->file_offset > orig_end) {
 837			btrfs_put_ordered_extent(ordered);
 838			break;
 839		}
 840		if (ordered->file_offset + ordered->len <= start) {
 841			btrfs_put_ordered_extent(ordered);
 842			break;
 843		}
 844		btrfs_start_ordered_extent(inode, ordered, 1);
 845		end = ordered->file_offset;
 
 
 
 
 
 846		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
 847			ret = -EIO;
 848		btrfs_put_ordered_extent(ordered);
 849		if (ret || end == 0 || end == start)
 850			break;
 851		end--;
 852	}
 853	return ret;
 854}
 855
 856/*
 857 * find an ordered extent corresponding to file_offset.  return NULL if
 858 * nothing is found, otherwise take a reference on the extent and return it
 859 */
 860struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode,
 861							 u64 file_offset)
 862{
 863	struct btrfs_ordered_inode_tree *tree;
 864	struct rb_node *node;
 865	struct btrfs_ordered_extent *entry = NULL;
 
 866
 867	tree = &BTRFS_I(inode)->ordered_tree;
 868	spin_lock_irq(&tree->lock);
 869	node = tree_search(tree, file_offset);
 870	if (!node)
 871		goto out;
 872
 873	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 874	if (!offset_in_entry(entry, file_offset))
 875		entry = NULL;
 876	if (entry)
 877		atomic_inc(&entry->refs);
 
 
 878out:
 879	spin_unlock_irq(&tree->lock);
 880	return entry;
 881}
 882
 883/* Since the DIO code tries to lock a wide area we need to look for any ordered
 884 * extents that exist in the range, rather than just the start of the range.
 885 */
 886struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode,
 887							u64 file_offset,
 888							u64 len)
 889{
 890	struct btrfs_ordered_inode_tree *tree;
 891	struct rb_node *node;
 892	struct btrfs_ordered_extent *entry = NULL;
 893
 894	tree = &BTRFS_I(inode)->ordered_tree;
 895	spin_lock_irq(&tree->lock);
 896	node = tree_search(tree, file_offset);
 897	if (!node) {
 898		node = tree_search(tree, file_offset + len);
 899		if (!node)
 900			goto out;
 901	}
 902
 903	while (1) {
 904		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 905		if (range_overlaps(entry, file_offset, len))
 906			break;
 907
 908		if (entry->file_offset >= file_offset + len) {
 909			entry = NULL;
 910			break;
 911		}
 912		entry = NULL;
 913		node = rb_next(node);
 914		if (!node)
 915			break;
 916	}
 917out:
 918	if (entry)
 919		atomic_inc(&entry->refs);
 920	spin_unlock_irq(&tree->lock);
 
 
 921	return entry;
 922}
 923
 924/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 925 * lookup and return any extent before 'file_offset'.  NULL is returned
 926 * if none is found
 927 */
 928struct btrfs_ordered_extent *
 929btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset)
 930{
 931	struct btrfs_ordered_inode_tree *tree;
 932	struct rb_node *node;
 933	struct btrfs_ordered_extent *entry = NULL;
 934
 935	tree = &BTRFS_I(inode)->ordered_tree;
 936	spin_lock_irq(&tree->lock);
 937	node = tree_search(tree, file_offset);
 938	if (!node)
 939		goto out;
 940
 941	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 942	atomic_inc(&entry->refs);
 
 943out:
 944	spin_unlock_irq(&tree->lock);
 945	return entry;
 946}
 947
 948/*
 949 * After an extent is done, call this to conditionally update the on disk
 950 * i_size.  i_size is updated to cover any fully written part of the file.
 
 
 
 
 
 951 */
 952int btrfs_ordered_update_i_size(struct inode *inode, u64 offset,
 953				struct btrfs_ordered_extent *ordered)
 954{
 955	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
 956	u64 disk_i_size;
 957	u64 new_i_size;
 958	u64 i_size = i_size_read(inode);
 959	struct rb_node *node;
 960	struct rb_node *prev = NULL;
 961	struct btrfs_ordered_extent *test;
 962	int ret = 1;
 963
 964	spin_lock_irq(&tree->lock);
 965	if (ordered) {
 966		offset = entry_end(ordered);
 967		if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags))
 968			offset = min(offset,
 969				     ordered->file_offset +
 970				     ordered->truncated_len);
 971	} else {
 972		offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize);
 973	}
 974	disk_i_size = BTRFS_I(inode)->disk_i_size;
 975
 976	/* truncate file */
 977	if (disk_i_size > i_size) {
 978		BTRFS_I(inode)->disk_i_size = i_size;
 979		ret = 0;
 980		goto out;
 981	}
 982
 983	/*
 984	 * if the disk i_size is already at the inode->i_size, or
 985	 * this ordered extent is inside the disk i_size, we're done
 986	 */
 987	if (disk_i_size == i_size)
 988		goto out;
 989
 990	/*
 991	 * We still need to update disk_i_size if outstanding_isize is greater
 992	 * than disk_i_size.
 993	 */
 994	if (offset <= disk_i_size &&
 995	    (!ordered || ordered->outstanding_isize <= disk_i_size))
 996		goto out;
 997
 
 
 998	/*
 999	 * walk backward from this ordered extent to disk_i_size.
1000	 * if we find an ordered extent then we can't update disk i_size
1001	 * yet
 
1002	 */
1003	if (ordered) {
1004		node = rb_prev(&ordered->rb_node);
1005	} else {
1006		prev = tree_search(tree, offset);
1007		/*
1008		 * we insert file extents without involving ordered struct,
1009		 * so there should be no ordered struct cover this offset
1010		 */
1011		if (prev) {
1012			test = rb_entry(prev, struct btrfs_ordered_extent,
1013					rb_node);
1014			BUG_ON(offset_in_entry(test, offset));
1015		}
1016		node = prev;
1017	}
1018	for (; node; node = rb_prev(node)) {
1019		test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
1020
1021		/* We treat this entry as if it doesnt exist */
1022		if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags))
1023			continue;
1024		if (test->file_offset + test->len <= disk_i_size)
1025			break;
1026		if (test->file_offset >= i_size)
1027			break;
1028		if (entry_end(test) > disk_i_size) {
1029			/*
1030			 * we don't update disk_i_size now, so record this
1031			 * undealt i_size. Or we will not know the real
1032			 * i_size.
1033			 */
1034			if (test->outstanding_isize < offset)
1035				test->outstanding_isize = offset;
1036			if (ordered &&
1037			    ordered->outstanding_isize >
1038			    test->outstanding_isize)
1039				test->outstanding_isize =
1040						ordered->outstanding_isize;
1041			goto out;
1042		}
1043	}
1044	new_i_size = min_t(u64, offset, i_size);
 
 
 
1045
1046	/*
1047	 * Some ordered extents may completed before the current one, and
1048	 * we hold the real i_size in ->outstanding_isize.
1049	 */
1050	if (ordered && ordered->outstanding_isize > new_i_size)
1051		new_i_size = min_t(u64, ordered->outstanding_isize, i_size);
1052	BTRFS_I(inode)->disk_i_size = new_i_size;
1053	ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
1054out:
1055	/*
1056	 * We need to do this because we can't remove ordered extents until
1057	 * after the i_disk_size has been updated and then the inode has been
1058	 * updated to reflect the change, so we need to tell anybody who finds
1059	 * this ordered extent that we've already done all the real work, we
1060	 * just haven't completed all the other work.
1061	 */
1062	if (ordered)
1063		set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags);
1064	spin_unlock_irq(&tree->lock);
1065	return ret;
1066}
1067
1068/*
1069 * search the ordered extents for one corresponding to 'offset' and
1070 * try to find a checksum.  This is used because we allow pages to
1071 * be reclaimed before their checksum is actually put into the btree
 
 
 
 
 
 
 
 
 
1072 */
1073int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr,
1074			   u32 *sum, int len)
 
1075{
1076	struct btrfs_ordered_sum *ordered_sum;
1077	struct btrfs_ordered_extent *ordered;
1078	struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
1079	unsigned long num_sectors;
1080	unsigned long i;
1081	u32 sectorsize = BTRFS_I(inode)->root->sectorsize;
1082	int index = 0;
1083
1084	ordered = btrfs_lookup_ordered_extent(inode, offset);
1085	if (!ordered)
1086		return 0;
1087
1088	spin_lock_irq(&tree->lock);
1089	list_for_each_entry_reverse(ordered_sum, &ordered->list, list) {
1090		if (disk_bytenr >= ordered_sum->bytenr &&
1091		    disk_bytenr < ordered_sum->bytenr + ordered_sum->len) {
1092			i = (disk_bytenr - ordered_sum->bytenr) >>
1093			    inode->i_sb->s_blocksize_bits;
1094			num_sectors = ordered_sum->len >>
1095				      inode->i_sb->s_blocksize_bits;
1096			num_sectors = min_t(int, len - index, num_sectors - i);
1097			memcpy(sum + index, ordered_sum->sums + i,
1098			       num_sectors);
1099
1100			index += (int)num_sectors;
1101			if (index == len)
1102				goto out;
1103			disk_bytenr += num_sectors * sectorsize;
1104		}
 
 
 
1105	}
1106out:
1107	spin_unlock_irq(&tree->lock);
1108	btrfs_put_ordered_extent(ordered);
1109	return index;
1110}
1111
1112
1113/*
1114 * add a given inode to the list of inodes that must be fully on
1115 * disk before a transaction commit finishes.
1116 *
1117 * This basically gives us the ext3 style data=ordered mode, and it is mostly
1118 * used to make sure renamed files are fully on disk.
1119 *
1120 * It is a noop if the inode is already fully on disk.
1121 *
1122 * If trans is not null, we'll do a friendly check for a transaction that
1123 * is already flushing things and force the IO down ourselves.
1124 */
1125void btrfs_add_ordered_operation(struct btrfs_trans_handle *trans,
1126				 struct btrfs_root *root, struct inode *inode)
1127{
1128	struct btrfs_transaction *cur_trans = trans->transaction;
1129	u64 last_mod;
1130
1131	last_mod = max(BTRFS_I(inode)->generation, BTRFS_I(inode)->last_trans);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1132
1133	/*
1134	 * if this file hasn't been changed since the last transaction
1135	 * commit, we can safely return without doing anything
1136	 */
1137	if (last_mod <= root->fs_info->last_trans_committed)
1138		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1139
1140	spin_lock(&root->fs_info->ordered_root_lock);
1141	if (list_empty(&BTRFS_I(inode)->ordered_operations)) {
1142		list_add_tail(&BTRFS_I(inode)->ordered_operations,
1143			      &cur_trans->ordered_operations);
 
 
 
1144	}
1145	spin_unlock(&root->fs_info->ordered_root_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1146}
1147
1148int __init ordered_data_init(void)
1149{
1150	btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent",
1151				     sizeof(struct btrfs_ordered_extent), 0,
1152				     SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
1153				     NULL);
1154	if (!btrfs_ordered_extent_cache)
1155		return -ENOMEM;
1156
1157	return 0;
1158}
1159
1160void ordered_data_exit(void)
1161{
1162	if (btrfs_ordered_extent_cache)
1163		kmem_cache_destroy(btrfs_ordered_extent_cache);
1164}
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2007 Oracle.  All rights reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   4 */
   5
   6#include <linux/slab.h>
   7#include <linux/blkdev.h>
   8#include <linux/writeback.h>
   9#include <linux/sched/mm.h>
  10#include "messages.h"
  11#include "misc.h"
  12#include "ctree.h"
  13#include "transaction.h"
  14#include "btrfs_inode.h"
  15#include "extent_io.h"
  16#include "disk-io.h"
  17#include "compression.h"
  18#include "delalloc-space.h"
  19#include "qgroup.h"
  20#include "subpage.h"
  21#include "file.h"
  22
  23static struct kmem_cache *btrfs_ordered_extent_cache;
  24
  25static u64 entry_end(struct btrfs_ordered_extent *entry)
  26{
  27	if (entry->file_offset + entry->num_bytes < entry->file_offset)
  28		return (u64)-1;
  29	return entry->file_offset + entry->num_bytes;
  30}
  31
  32/* returns NULL if the insertion worked, or it returns the node it did find
  33 * in the tree
  34 */
  35static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset,
  36				   struct rb_node *node)
  37{
  38	struct rb_node **p = &root->rb_node;
  39	struct rb_node *parent = NULL;
  40	struct btrfs_ordered_extent *entry;
  41
  42	while (*p) {
  43		parent = *p;
  44		entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node);
  45
  46		if (file_offset < entry->file_offset)
  47			p = &(*p)->rb_left;
  48		else if (file_offset >= entry_end(entry))
  49			p = &(*p)->rb_right;
  50		else
  51			return parent;
  52	}
  53
  54	rb_link_node(node, parent, p);
  55	rb_insert_color(node, root);
  56	return NULL;
  57}
  58
 
 
 
 
 
 
 
 
  59/*
  60 * look for a given offset in the tree, and if it can't be found return the
  61 * first lesser offset
  62 */
  63static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset,
  64				     struct rb_node **prev_ret)
  65{
  66	struct rb_node *n = root->rb_node;
  67	struct rb_node *prev = NULL;
  68	struct rb_node *test;
  69	struct btrfs_ordered_extent *entry;
  70	struct btrfs_ordered_extent *prev_entry = NULL;
  71
  72	while (n) {
  73		entry = rb_entry(n, struct btrfs_ordered_extent, rb_node);
  74		prev = n;
  75		prev_entry = entry;
  76
  77		if (file_offset < entry->file_offset)
  78			n = n->rb_left;
  79		else if (file_offset >= entry_end(entry))
  80			n = n->rb_right;
  81		else
  82			return n;
  83	}
  84	if (!prev_ret)
  85		return NULL;
  86
  87	while (prev && file_offset >= entry_end(prev_entry)) {
  88		test = rb_next(prev);
  89		if (!test)
  90			break;
  91		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
  92				      rb_node);
  93		if (file_offset < entry_end(prev_entry))
  94			break;
  95
  96		prev = test;
  97	}
  98	if (prev)
  99		prev_entry = rb_entry(prev, struct btrfs_ordered_extent,
 100				      rb_node);
 101	while (prev && file_offset < entry_end(prev_entry)) {
 102		test = rb_prev(prev);
 103		if (!test)
 104			break;
 105		prev_entry = rb_entry(test, struct btrfs_ordered_extent,
 106				      rb_node);
 107		prev = test;
 108	}
 109	*prev_ret = prev;
 110	return NULL;
 111}
 112
 
 
 
 
 
 
 
 
 
 
 
 113static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset,
 114			  u64 len)
 115{
 116	if (file_offset + len <= entry->file_offset ||
 117	    entry->file_offset + entry->num_bytes <= file_offset)
 118		return 0;
 119	return 1;
 120}
 121
 122/*
 123 * look find the first ordered struct that has this offset, otherwise
 124 * the first one less than this offset
 125 */
 126static inline struct rb_node *ordered_tree_search(struct btrfs_inode *inode,
 127						  u64 file_offset)
 128{
 
 129	struct rb_node *prev = NULL;
 130	struct rb_node *ret;
 131	struct btrfs_ordered_extent *entry;
 132
 133	if (inode->ordered_tree_last) {
 134		entry = rb_entry(inode->ordered_tree_last, struct btrfs_ordered_extent,
 135				 rb_node);
 136		if (in_range(file_offset, entry->file_offset, entry->num_bytes))
 137			return inode->ordered_tree_last;
 138	}
 139	ret = __tree_search(&inode->ordered_tree, file_offset, &prev);
 140	if (!ret)
 141		ret = prev;
 142	if (ret)
 143		inode->ordered_tree_last = ret;
 144	return ret;
 145}
 146
 147static struct btrfs_ordered_extent *alloc_ordered_extent(
 148			struct btrfs_inode *inode, u64 file_offset, u64 num_bytes,
 149			u64 ram_bytes, u64 disk_bytenr, u64 disk_num_bytes,
 150			u64 offset, unsigned long flags, int compress_type)
 
 
 
 
 
 
 
 
 
 
 151{
 
 
 
 152	struct btrfs_ordered_extent *entry;
 153	int ret;
 154	u64 qgroup_rsv = 0;
 155
 156	if (flags &
 157	    ((1 << BTRFS_ORDERED_NOCOW) | (1 << BTRFS_ORDERED_PREALLOC))) {
 158		/* For nocow write, we can release the qgroup rsv right now */
 159		ret = btrfs_qgroup_free_data(inode, NULL, file_offset, num_bytes, &qgroup_rsv);
 160		if (ret < 0)
 161			return ERR_PTR(ret);
 162	} else {
 163		/*
 164		 * The ordered extent has reserved qgroup space, release now
 165		 * and pass the reserved number for qgroup_record to free.
 166		 */
 167		ret = btrfs_qgroup_release_data(inode, file_offset, num_bytes, &qgroup_rsv);
 168		if (ret < 0)
 169			return ERR_PTR(ret);
 170	}
 171	entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS);
 172	if (!entry)
 173		return ERR_PTR(-ENOMEM);
 174
 175	entry->file_offset = file_offset;
 176	entry->num_bytes = num_bytes;
 177	entry->ram_bytes = ram_bytes;
 178	entry->disk_bytenr = disk_bytenr;
 179	entry->disk_num_bytes = disk_num_bytes;
 180	entry->offset = offset;
 181	entry->bytes_left = num_bytes;
 182	entry->inode = igrab(&inode->vfs_inode);
 
 183	entry->compress_type = compress_type;
 184	entry->truncated_len = (u64)-1;
 185	entry->qgroup_rsv = qgroup_rsv;
 186	entry->flags = flags;
 187	refcount_set(&entry->refs, 1);
 
 
 
 
 
 188	init_waitqueue_head(&entry->wait);
 189	INIT_LIST_HEAD(&entry->list);
 190	INIT_LIST_HEAD(&entry->log_list);
 191	INIT_LIST_HEAD(&entry->root_extent_list);
 192	INIT_LIST_HEAD(&entry->work_list);
 193	INIT_LIST_HEAD(&entry->bioc_list);
 194	init_completion(&entry->completion);
 195
 196	/*
 197	 * We don't need the count_max_extents here, we can assume that all of
 198	 * that work has been done at higher layers, so this is truly the
 199	 * smallest the extent is going to get.
 200	 */
 201	spin_lock(&inode->lock);
 202	btrfs_mod_outstanding_extents(inode, 1);
 203	spin_unlock(&inode->lock);
 204
 205	return entry;
 206}
 207
 208static void insert_ordered_extent(struct btrfs_ordered_extent *entry)
 209{
 210	struct btrfs_inode *inode = BTRFS_I(entry->inode);
 211	struct btrfs_root *root = inode->root;
 212	struct btrfs_fs_info *fs_info = root->fs_info;
 213	struct rb_node *node;
 214
 215	trace_btrfs_ordered_extent_add(inode, entry);
 216
 217	percpu_counter_add_batch(&fs_info->ordered_bytes, entry->num_bytes,
 218				 fs_info->delalloc_batch);
 219
 220	/* One ref for the tree. */
 221	refcount_inc(&entry->refs);
 222
 223	spin_lock_irq(&inode->ordered_tree_lock);
 224	node = tree_insert(&inode->ordered_tree, entry->file_offset,
 225			   &entry->rb_node);
 226	if (node)
 227		btrfs_panic(fs_info, -EEXIST,
 228				"inconsistency in ordered tree at offset %llu",
 229				entry->file_offset);
 230	spin_unlock_irq(&inode->ordered_tree_lock);
 231
 232	spin_lock(&root->ordered_extent_lock);
 233	list_add_tail(&entry->root_extent_list,
 234		      &root->ordered_extents);
 235	root->nr_ordered_extents++;
 236	if (root->nr_ordered_extents == 1) {
 237		spin_lock(&fs_info->ordered_root_lock);
 238		BUG_ON(!list_empty(&root->ordered_root));
 239		list_add_tail(&root->ordered_root, &fs_info->ordered_roots);
 240		spin_unlock(&fs_info->ordered_root_lock);
 
 241	}
 242	spin_unlock(&root->ordered_extent_lock);
 
 
 243}
 244
 245/*
 246 * Add an ordered extent to the per-inode tree.
 247 *
 248 * @inode:           Inode that this extent is for.
 249 * @file_offset:     Logical offset in file where the extent starts.
 250 * @num_bytes:       Logical length of extent in file.
 251 * @ram_bytes:       Full length of unencoded data.
 252 * @disk_bytenr:     Offset of extent on disk.
 253 * @disk_num_bytes:  Size of extent on disk.
 254 * @offset:          Offset into unencoded data where file data starts.
 255 * @flags:           Flags specifying type of extent (1 << BTRFS_ORDERED_*).
 256 * @compress_type:   Compression algorithm used for data.
 257 *
 258 * Most of these parameters correspond to &struct btrfs_file_extent_item. The
 259 * tree is given a single reference on the ordered extent that was inserted, and
 260 * the returned pointer is given a second reference.
 261 *
 262 * Return: the new ordered extent or error pointer.
 263 */
 264struct btrfs_ordered_extent *btrfs_alloc_ordered_extent(
 265			struct btrfs_inode *inode, u64 file_offset,
 266			u64 num_bytes, u64 ram_bytes, u64 disk_bytenr,
 267			u64 disk_num_bytes, u64 offset, unsigned long flags,
 268			int compress_type)
 269{
 270	struct btrfs_ordered_extent *entry;
 
 
 
 271
 272	ASSERT((flags & ~BTRFS_ORDERED_TYPE_FLAGS) == 0);
 
 
 
 
 
 
 273
 274	entry = alloc_ordered_extent(inode, file_offset, num_bytes, ram_bytes,
 275				     disk_bytenr, disk_num_bytes, offset, flags,
 276				     compress_type);
 277	if (!IS_ERR(entry))
 278		insert_ordered_extent(entry);
 279	return entry;
 
 280}
 281
 282/*
 283 * Add a struct btrfs_ordered_sum into the list of checksums to be inserted
 284 * when an ordered extent is finished.  If the list covers more than one
 285 * ordered extent, it is split across multiples.
 286 */
 287void btrfs_add_ordered_sum(struct btrfs_ordered_extent *entry,
 
 288			   struct btrfs_ordered_sum *sum)
 289{
 290	struct btrfs_inode *inode = BTRFS_I(entry->inode);
 291
 292	spin_lock_irq(&inode->ordered_tree_lock);
 
 293	list_add_tail(&sum->list, &entry->list);
 294	spin_unlock_irq(&inode->ordered_tree_lock);
 295}
 296
 297static void finish_ordered_fn(struct btrfs_work *work)
 298{
 299	struct btrfs_ordered_extent *ordered_extent;
 300
 301	ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
 302	btrfs_finish_ordered_io(ordered_extent);
 303}
 304
 305static bool can_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
 306				      struct page *page, u64 file_offset,
 307				      u64 len, bool uptodate)
 308{
 309	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
 310	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 311
 312	lockdep_assert_held(&inode->ordered_tree_lock);
 313
 314	if (page) {
 315		ASSERT(page->mapping);
 316		ASSERT(page_offset(page) <= file_offset);
 317		ASSERT(file_offset + len <= page_offset(page) + PAGE_SIZE);
 318
 319		/*
 320		 * Ordered (Private2) bit indicates whether we still have
 321		 * pending io unfinished for the ordered extent.
 322		 *
 323		 * If there's no such bit, we need to skip to next range.
 324		 */
 325		if (!btrfs_folio_test_ordered(fs_info, page_folio(page),
 326					      file_offset, len))
 327			return false;
 328		btrfs_folio_clear_ordered(fs_info, page_folio(page), file_offset, len);
 329	}
 330
 331	/* Now we're fine to update the accounting. */
 332	if (WARN_ON_ONCE(len > ordered->bytes_left)) {
 333		btrfs_crit(fs_info,
 334"bad ordered extent accounting, root=%llu ino=%llu OE offset=%llu OE len=%llu to_dec=%llu left=%llu",
 335			   inode->root->root_key.objectid, btrfs_ino(inode),
 336			   ordered->file_offset, ordered->num_bytes,
 337			   len, ordered->bytes_left);
 338		ordered->bytes_left = 0;
 339	} else {
 340		ordered->bytes_left -= len;
 341	}
 342
 343	if (!uptodate)
 344		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
 345
 346	if (ordered->bytes_left)
 347		return false;
 348
 349	/*
 350	 * All the IO of the ordered extent is finished, we need to queue
 351	 * the finish_func to be executed.
 352	 */
 353	set_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags);
 354	cond_wake_up(&ordered->wait);
 355	refcount_inc(&ordered->refs);
 356	trace_btrfs_ordered_extent_mark_finished(inode, ordered);
 357	return true;
 358}
 359
 360static void btrfs_queue_ordered_fn(struct btrfs_ordered_extent *ordered)
 361{
 362	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
 363	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 364	struct btrfs_workqueue *wq = btrfs_is_free_space_inode(inode) ?
 365		fs_info->endio_freespace_worker : fs_info->endio_write_workers;
 366
 367	btrfs_init_work(&ordered->work, finish_ordered_fn, NULL);
 368	btrfs_queue_work(wq, &ordered->work);
 369}
 370
 371bool btrfs_finish_ordered_extent(struct btrfs_ordered_extent *ordered,
 372				 struct page *page, u64 file_offset, u64 len,
 373				 bool uptodate)
 374{
 375	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
 376	unsigned long flags;
 377	bool ret;
 378
 379	trace_btrfs_finish_ordered_extent(inode, file_offset, len, uptodate);
 380
 381	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
 382	ret = can_finish_ordered_extent(ordered, page, file_offset, len, uptodate);
 383	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
 384
 385	if (ret)
 386		btrfs_queue_ordered_fn(ordered);
 387	return ret;
 388}
 389
 390/*
 391 * Mark all ordered extents io inside the specified range finished.
 
 
 
 392 *
 393 * @page:	 The involved page for the operation.
 394 *		 For uncompressed buffered IO, the page status also needs to be
 395 *		 updated to indicate whether the pending ordered io is finished.
 396 *		 Can be NULL for direct IO and compressed write.
 397 *		 For these cases, callers are ensured they won't execute the
 398 *		 endio function twice.
 399 *
 400 * This function is called for endio, thus the range must have ordered
 401 * extent(s) covering it.
 402 */
 403void btrfs_mark_ordered_io_finished(struct btrfs_inode *inode,
 404				    struct page *page, u64 file_offset,
 405				    u64 num_bytes, bool uptodate)
 406{
 
 407	struct rb_node *node;
 408	struct btrfs_ordered_extent *entry = NULL;
 
 409	unsigned long flags;
 410	u64 cur = file_offset;
 
 
 
 
 
 
 
 
 
 
 411
 412	trace_btrfs_writepage_end_io_hook(inode, file_offset,
 413					  file_offset + num_bytes - 1,
 414					  uptodate);
 415
 416	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
 417	while (cur < file_offset + num_bytes) {
 418		u64 entry_end;
 419		u64 end;
 420		u32 len;
 421
 422		node = ordered_tree_search(inode, cur);
 423		/* No ordered extents at all */
 424		if (!node)
 425			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 426
 427		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 428		entry_end = entry->file_offset + entry->num_bytes;
 429		/*
 430		 * |<-- OE --->|  |
 431		 *		  cur
 432		 * Go to next OE.
 433		 */
 434		if (cur >= entry_end) {
 435			node = rb_next(node);
 436			/* No more ordered extents, exit */
 437			if (!node)
 438				break;
 439			entry = rb_entry(node, struct btrfs_ordered_extent,
 440					 rb_node);
 441
 442			/* Go to next ordered extent and continue */
 443			cur = entry->file_offset;
 444			continue;
 445		}
 446		/*
 447		 * |	|<--- OE --->|
 448		 * cur
 449		 * Go to the start of OE.
 450		 */
 451		if (cur < entry->file_offset) {
 452			cur = entry->file_offset;
 453			continue;
 454		}
 455
 456		/*
 457		 * Now we are definitely inside one ordered extent.
 458		 *
 459		 * |<--- OE --->|
 460		 *	|
 461		 *	cur
 462		 */
 463		end = min(entry->file_offset + entry->num_bytes,
 464			  file_offset + num_bytes) - 1;
 465		ASSERT(end + 1 - cur < U32_MAX);
 466		len = end + 1 - cur;
 467
 468		if (can_finish_ordered_extent(entry, page, cur, len, uptodate)) {
 469			spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
 470			btrfs_queue_ordered_fn(entry);
 471			spin_lock_irqsave(&inode->ordered_tree_lock, flags);
 472		}
 473		cur += len;
 474	}
 475	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
 
 476}
 477
 478/*
 479 * Finish IO for one ordered extent across a given range.  The range can only
 480 * contain one ordered extent.
 481 *
 482 * @cached:	 The cached ordered extent. If not NULL, we can skip the tree
 483 *               search and use the ordered extent directly.
 484 * 		 Will be also used to store the finished ordered extent.
 485 * @file_offset: File offset for the finished IO
 486 * @io_size:	 Length of the finish IO range
 487 *
 488 * Return true if the ordered extent is finished in the range, and update
 489 * @cached.
 490 * Return false otherwise.
 491 *
 492 * NOTE: The range can NOT cross multiple ordered extents.
 493 * Thus caller should ensure the range doesn't cross ordered extents.
 494 */
 495bool btrfs_dec_test_ordered_pending(struct btrfs_inode *inode,
 496				    struct btrfs_ordered_extent **cached,
 497				    u64 file_offset, u64 io_size)
 498{
 
 499	struct rb_node *node;
 500	struct btrfs_ordered_extent *entry = NULL;
 501	unsigned long flags;
 502	bool finished = false;
 503
 504	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
 
 505	if (cached && *cached) {
 506		entry = *cached;
 507		goto have_entry;
 508	}
 509
 510	node = ordered_tree_search(inode, file_offset);
 511	if (!node)
 
 512		goto out;
 
 513
 514	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 515have_entry:
 516	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
 
 517		goto out;
 
 518
 519	if (io_size > entry->bytes_left)
 520		btrfs_crit(inode->root->fs_info,
 521			   "bad ordered accounting left %llu size %llu",
 522		       entry->bytes_left, io_size);
 523
 524	entry->bytes_left -= io_size;
 
 
 525
 526	if (entry->bytes_left == 0) {
 527		/*
 528		 * Ensure only one caller can set the flag and finished_ret
 529		 * accordingly
 530		 */
 531		finished = !test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags);
 532		/* test_and_set_bit implies a barrier */
 533		cond_wake_up_nomb(&entry->wait);
 534	}
 535out:
 536	if (finished && cached && entry) {
 537		*cached = entry;
 538		refcount_inc(&entry->refs);
 539		trace_btrfs_ordered_extent_dec_test_pending(inode, entry);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 540	}
 541	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
 542	return finished;
 543}
 544
 545/*
 546 * used to drop a reference on an ordered extent.  This will free
 547 * the extent if the last reference is dropped
 548 */
 549void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry)
 550{
 551	struct list_head *cur;
 552	struct btrfs_ordered_sum *sum;
 553
 554	trace_btrfs_ordered_extent_put(BTRFS_I(entry->inode), entry);
 555
 556	if (refcount_dec_and_test(&entry->refs)) {
 557		ASSERT(list_empty(&entry->root_extent_list));
 558		ASSERT(list_empty(&entry->log_list));
 559		ASSERT(RB_EMPTY_NODE(&entry->rb_node));
 560		if (entry->inode)
 561			btrfs_add_delayed_iput(BTRFS_I(entry->inode));
 562		while (!list_empty(&entry->list)) {
 563			cur = entry->list.next;
 564			sum = list_entry(cur, struct btrfs_ordered_sum, list);
 565			list_del(&sum->list);
 566			kvfree(sum);
 567		}
 568		kmem_cache_free(btrfs_ordered_extent_cache, entry);
 569	}
 570}
 571
 572/*
 573 * remove an ordered extent from the tree.  No references are dropped
 574 * and waiters are woken up.
 575 */
 576void btrfs_remove_ordered_extent(struct btrfs_inode *btrfs_inode,
 577				 struct btrfs_ordered_extent *entry)
 578{
 579	struct btrfs_root *root = btrfs_inode->root;
 580	struct btrfs_fs_info *fs_info = root->fs_info;
 581	struct rb_node *node;
 582	bool pending;
 583	bool freespace_inode;
 584
 585	/*
 586	 * If this is a free space inode the thread has not acquired the ordered
 587	 * extents lockdep map.
 588	 */
 589	freespace_inode = btrfs_is_free_space_inode(btrfs_inode);
 590
 591	btrfs_lockdep_acquire(fs_info, btrfs_trans_pending_ordered);
 592	/* This is paired with btrfs_alloc_ordered_extent. */
 593	spin_lock(&btrfs_inode->lock);
 594	btrfs_mod_outstanding_extents(btrfs_inode, -1);
 595	spin_unlock(&btrfs_inode->lock);
 596	if (root != fs_info->tree_root) {
 597		u64 release;
 598
 599		if (test_bit(BTRFS_ORDERED_ENCODED, &entry->flags))
 600			release = entry->disk_num_bytes;
 601		else
 602			release = entry->num_bytes;
 603		btrfs_delalloc_release_metadata(btrfs_inode, release,
 604						test_bit(BTRFS_ORDERED_IOERR,
 605							 &entry->flags));
 606	}
 607
 608	percpu_counter_add_batch(&fs_info->ordered_bytes, -entry->num_bytes,
 609				 fs_info->delalloc_batch);
 610
 611	spin_lock_irq(&btrfs_inode->ordered_tree_lock);
 612	node = &entry->rb_node;
 613	rb_erase(node, &btrfs_inode->ordered_tree);
 614	RB_CLEAR_NODE(node);
 615	if (btrfs_inode->ordered_tree_last == node)
 616		btrfs_inode->ordered_tree_last = NULL;
 617	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
 618	pending = test_and_clear_bit(BTRFS_ORDERED_PENDING, &entry->flags);
 619	spin_unlock_irq(&btrfs_inode->ordered_tree_lock);
 620
 621	/*
 622	 * The current running transaction is waiting on us, we need to let it
 623	 * know that we're complete and wake it up.
 624	 */
 625	if (pending) {
 626		struct btrfs_transaction *trans;
 627
 628		/*
 629		 * The checks for trans are just a formality, it should be set,
 630		 * but if it isn't we don't want to deref/assert under the spin
 631		 * lock, so be nice and check if trans is set, but ASSERT() so
 632		 * if it isn't set a developer will notice.
 633		 */
 634		spin_lock(&fs_info->trans_lock);
 635		trans = fs_info->running_transaction;
 636		if (trans)
 637			refcount_inc(&trans->use_count);
 638		spin_unlock(&fs_info->trans_lock);
 639
 640		ASSERT(trans || BTRFS_FS_ERROR(fs_info));
 641		if (trans) {
 642			if (atomic_dec_and_test(&trans->pending_ordered))
 643				wake_up(&trans->pending_wait);
 644			btrfs_put_transaction(trans);
 645		}
 646	}
 647
 648	btrfs_lockdep_release(fs_info, btrfs_trans_pending_ordered);
 649
 650	spin_lock(&root->ordered_extent_lock);
 651	list_del_init(&entry->root_extent_list);
 652	root->nr_ordered_extents--;
 653
 654	trace_btrfs_ordered_extent_remove(btrfs_inode, entry);
 
 
 
 
 
 
 
 
 
 
 
 
 655
 656	if (!root->nr_ordered_extents) {
 657		spin_lock(&fs_info->ordered_root_lock);
 658		BUG_ON(list_empty(&root->ordered_root));
 659		list_del_init(&root->ordered_root);
 660		spin_unlock(&fs_info->ordered_root_lock);
 661	}
 662	spin_unlock(&root->ordered_extent_lock);
 663	wake_up(&entry->wait);
 664	if (!freespace_inode)
 665		btrfs_lockdep_release(fs_info, btrfs_ordered_extent);
 666}
 667
 668static void btrfs_run_ordered_extent_work(struct btrfs_work *work)
 669{
 670	struct btrfs_ordered_extent *ordered;
 671
 672	ordered = container_of(work, struct btrfs_ordered_extent, flush_work);
 673	btrfs_start_ordered_extent(ordered);
 674	complete(&ordered->completion);
 675}
 676
 677/*
 678 * wait for all the ordered extents in a root.  This is done when balancing
 679 * space between drives.
 680 */
 681u64 btrfs_wait_ordered_extents(struct btrfs_root *root, u64 nr,
 682			       const u64 range_start, const u64 range_len)
 683{
 684	struct btrfs_fs_info *fs_info = root->fs_info;
 685	LIST_HEAD(splice);
 686	LIST_HEAD(skipped);
 687	LIST_HEAD(works);
 688	struct btrfs_ordered_extent *ordered, *next;
 689	u64 count = 0;
 690	const u64 range_end = range_start + range_len;
 
 
 691
 692	mutex_lock(&root->ordered_extent_mutex);
 693	spin_lock(&root->ordered_extent_lock);
 694	list_splice_init(&root->ordered_extents, &splice);
 695	while (!list_empty(&splice) && nr) {
 696		ordered = list_first_entry(&splice, struct btrfs_ordered_extent,
 697					   root_extent_list);
 698
 699		if (range_end <= ordered->disk_bytenr ||
 700		    ordered->disk_bytenr + ordered->disk_num_bytes <= range_start) {
 701			list_move_tail(&ordered->root_extent_list, &skipped);
 702			cond_resched_lock(&root->ordered_extent_lock);
 703			continue;
 704		}
 705
 706		list_move_tail(&ordered->root_extent_list,
 707			       &root->ordered_extents);
 708		refcount_inc(&ordered->refs);
 709		spin_unlock(&root->ordered_extent_lock);
 710
 711		btrfs_init_work(&ordered->flush_work,
 712				btrfs_run_ordered_extent_work, NULL);
 713		list_add_tail(&ordered->work_list, &works);
 714		btrfs_queue_work(fs_info->flush_workers, &ordered->flush_work);
 
 715
 716		cond_resched();
 717		spin_lock(&root->ordered_extent_lock);
 718		if (nr != U64_MAX)
 719			nr--;
 720		count++;
 721	}
 722	list_splice_tail(&skipped, &root->ordered_extents);
 723	list_splice_tail(&splice, &root->ordered_extents);
 724	spin_unlock(&root->ordered_extent_lock);
 725
 726	list_for_each_entry_safe(ordered, next, &works, work_list) {
 727		list_del_init(&ordered->work_list);
 728		wait_for_completion(&ordered->completion);
 729		btrfs_put_ordered_extent(ordered);
 730		cond_resched();
 731	}
 732	mutex_unlock(&root->ordered_extent_mutex);
 733
 734	return count;
 735}
 736
 737void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, u64 nr,
 738			     const u64 range_start, const u64 range_len)
 739{
 740	struct btrfs_root *root;
 741	LIST_HEAD(splice);
 742	u64 done;
 
 
 743
 744	mutex_lock(&fs_info->ordered_operations_mutex);
 745	spin_lock(&fs_info->ordered_root_lock);
 746	list_splice_init(&fs_info->ordered_roots, &splice);
 747	while (!list_empty(&splice) && nr) {
 748		root = list_first_entry(&splice, struct btrfs_root,
 749					ordered_root);
 750		root = btrfs_grab_root(root);
 751		BUG_ON(!root);
 752		list_move_tail(&root->ordered_root,
 753			       &fs_info->ordered_roots);
 754		spin_unlock(&fs_info->ordered_root_lock);
 755
 756		done = btrfs_wait_ordered_extents(root, nr,
 757						  range_start, range_len);
 758		btrfs_put_root(root);
 759
 760		spin_lock(&fs_info->ordered_root_lock);
 761		if (nr != U64_MAX) {
 762			nr -= done;
 
 763		}
 764	}
 765	list_splice_tail(&splice, &fs_info->ordered_roots);
 766	spin_unlock(&fs_info->ordered_root_lock);
 767	mutex_unlock(&fs_info->ordered_operations_mutex);
 768}
 769
 770/*
 771 * Start IO and wait for a given ordered extent to finish.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 772 *
 773 * Wait on page writeback for all the pages in the extent and the IO completion
 774 * code to insert metadata into the btree corresponding to the extent.
 
 775 */
 776void btrfs_start_ordered_extent(struct btrfs_ordered_extent *entry)
 
 
 777{
 778	u64 start = entry->file_offset;
 779	u64 end = start + entry->num_bytes - 1;
 780	struct btrfs_inode *inode = BTRFS_I(entry->inode);
 781	bool freespace_inode;
 782
 783	trace_btrfs_ordered_extent_start(inode, entry);
 784
 785	/*
 786	 * If this is a free space inode do not take the ordered extents lockdep
 787	 * map.
 788	 */
 789	freespace_inode = btrfs_is_free_space_inode(inode);
 790
 791	/*
 792	 * pages in the range can be dirty, clean or writeback.  We
 793	 * start IO on any dirty ones so the wait doesn't stall waiting
 794	 * for the flusher thread to find them
 795	 */
 796	if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags))
 797		filemap_fdatawrite_range(inode->vfs_inode.i_mapping, start, end);
 798
 799	if (!freespace_inode)
 800		btrfs_might_wait_for_event(inode->root->fs_info, btrfs_ordered_extent);
 801	wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, &entry->flags));
 802}
 803
 804/*
 805 * Used to wait on ordered extents across a large range of bytes.
 806 */
 807int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len)
 808{
 809	int ret = 0;
 810	int ret_wb = 0;
 811	u64 end;
 812	u64 orig_end;
 813	struct btrfs_ordered_extent *ordered;
 814
 815	if (start + len < start) {
 816		orig_end = OFFSET_MAX;
 817	} else {
 818		orig_end = start + len - 1;
 819		if (orig_end > OFFSET_MAX)
 820			orig_end = OFFSET_MAX;
 821	}
 822
 823	/* start IO across the range first to instantiate any delalloc
 824	 * extents
 825	 */
 826	ret = btrfs_fdatawrite_range(inode, start, orig_end);
 827	if (ret)
 828		return ret;
 829
 830	/*
 831	 * If we have a writeback error don't return immediately. Wait first
 832	 * for any ordered extents that haven't completed yet. This is to make
 833	 * sure no one can dirty the same page ranges and call writepages()
 834	 * before the ordered extents complete - to avoid failures (-EEXIST)
 835	 * when adding the new ordered extents to the ordered tree.
 
 
 
 
 
 
 
 836	 */
 837	ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end);
 
 
 
 
 
 
 
 
 
 838
 839	end = orig_end;
 840	while (1) {
 841		ordered = btrfs_lookup_first_ordered_extent(BTRFS_I(inode), end);
 842		if (!ordered)
 843			break;
 844		if (ordered->file_offset > orig_end) {
 845			btrfs_put_ordered_extent(ordered);
 846			break;
 847		}
 848		if (ordered->file_offset + ordered->num_bytes <= start) {
 849			btrfs_put_ordered_extent(ordered);
 850			break;
 851		}
 852		btrfs_start_ordered_extent(ordered);
 853		end = ordered->file_offset;
 854		/*
 855		 * If the ordered extent had an error save the error but don't
 856		 * exit without waiting first for all other ordered extents in
 857		 * the range to complete.
 858		 */
 859		if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
 860			ret = -EIO;
 861		btrfs_put_ordered_extent(ordered);
 862		if (end == 0 || end == start)
 863			break;
 864		end--;
 865	}
 866	return ret_wb ? ret_wb : ret;
 867}
 868
 869/*
 870 * find an ordered extent corresponding to file_offset.  return NULL if
 871 * nothing is found, otherwise take a reference on the extent and return it
 872 */
 873struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct btrfs_inode *inode,
 874							 u64 file_offset)
 875{
 
 876	struct rb_node *node;
 877	struct btrfs_ordered_extent *entry = NULL;
 878	unsigned long flags;
 879
 880	spin_lock_irqsave(&inode->ordered_tree_lock, flags);
 881	node = ordered_tree_search(inode, file_offset);
 
 882	if (!node)
 883		goto out;
 884
 885	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 886	if (!in_range(file_offset, entry->file_offset, entry->num_bytes))
 887		entry = NULL;
 888	if (entry) {
 889		refcount_inc(&entry->refs);
 890		trace_btrfs_ordered_extent_lookup(inode, entry);
 891	}
 892out:
 893	spin_unlock_irqrestore(&inode->ordered_tree_lock, flags);
 894	return entry;
 895}
 896
 897/* Since the DIO code tries to lock a wide area we need to look for any ordered
 898 * extents that exist in the range, rather than just the start of the range.
 899 */
 900struct btrfs_ordered_extent *btrfs_lookup_ordered_range(
 901		struct btrfs_inode *inode, u64 file_offset, u64 len)
 
 902{
 
 903	struct rb_node *node;
 904	struct btrfs_ordered_extent *entry = NULL;
 905
 906	spin_lock_irq(&inode->ordered_tree_lock);
 907	node = ordered_tree_search(inode, file_offset);
 
 908	if (!node) {
 909		node = ordered_tree_search(inode, file_offset + len);
 910		if (!node)
 911			goto out;
 912	}
 913
 914	while (1) {
 915		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 916		if (range_overlaps(entry, file_offset, len))
 917			break;
 918
 919		if (entry->file_offset >= file_offset + len) {
 920			entry = NULL;
 921			break;
 922		}
 923		entry = NULL;
 924		node = rb_next(node);
 925		if (!node)
 926			break;
 927	}
 928out:
 929	if (entry) {
 930		refcount_inc(&entry->refs);
 931		trace_btrfs_ordered_extent_lookup_range(inode, entry);
 932	}
 933	spin_unlock_irq(&inode->ordered_tree_lock);
 934	return entry;
 935}
 936
 937/*
 938 * Adds all ordered extents to the given list. The list ends up sorted by the
 939 * file_offset of the ordered extents.
 940 */
 941void btrfs_get_ordered_extents_for_logging(struct btrfs_inode *inode,
 942					   struct list_head *list)
 943{
 944	struct rb_node *n;
 945
 946	ASSERT(inode_is_locked(&inode->vfs_inode));
 947
 948	spin_lock_irq(&inode->ordered_tree_lock);
 949	for (n = rb_first(&inode->ordered_tree); n; n = rb_next(n)) {
 950		struct btrfs_ordered_extent *ordered;
 951
 952		ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node);
 953
 954		if (test_bit(BTRFS_ORDERED_LOGGED, &ordered->flags))
 955			continue;
 956
 957		ASSERT(list_empty(&ordered->log_list));
 958		list_add_tail(&ordered->log_list, list);
 959		refcount_inc(&ordered->refs);
 960		trace_btrfs_ordered_extent_lookup_for_logging(inode, ordered);
 961	}
 962	spin_unlock_irq(&inode->ordered_tree_lock);
 963}
 964
 965/*
 966 * lookup and return any extent before 'file_offset'.  NULL is returned
 967 * if none is found
 968 */
 969struct btrfs_ordered_extent *
 970btrfs_lookup_first_ordered_extent(struct btrfs_inode *inode, u64 file_offset)
 971{
 
 972	struct rb_node *node;
 973	struct btrfs_ordered_extent *entry = NULL;
 974
 975	spin_lock_irq(&inode->ordered_tree_lock);
 976	node = ordered_tree_search(inode, file_offset);
 
 977	if (!node)
 978		goto out;
 979
 980	entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 981	refcount_inc(&entry->refs);
 982	trace_btrfs_ordered_extent_lookup_first(inode, entry);
 983out:
 984	spin_unlock_irq(&inode->ordered_tree_lock);
 985	return entry;
 986}
 987
 988/*
 989 * Lookup the first ordered extent that overlaps the range
 990 * [@file_offset, @file_offset + @len).
 991 *
 992 * The difference between this and btrfs_lookup_first_ordered_extent() is
 993 * that this one won't return any ordered extent that does not overlap the range.
 994 * And the difference against btrfs_lookup_ordered_extent() is, this function
 995 * ensures the first ordered extent gets returned.
 996 */
 997struct btrfs_ordered_extent *btrfs_lookup_first_ordered_range(
 998			struct btrfs_inode *inode, u64 file_offset, u64 len)
 999{
 
 
 
 
1000	struct rb_node *node;
1001	struct rb_node *cur;
1002	struct rb_node *prev;
1003	struct rb_node *next;
1004	struct btrfs_ordered_extent *entry = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1005
1006	spin_lock_irq(&inode->ordered_tree_lock);
1007	node = inode->ordered_tree.rb_node;
1008	/*
1009	 * Here we don't want to use tree_search() which will use tree->last
1010	 * and screw up the search order.
1011	 * And __tree_search() can't return the adjacent ordered extents
1012	 * either, thus here we do our own search.
1013	 */
1014	while (node) {
1015		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1016
1017		if (file_offset < entry->file_offset) {
1018			node = node->rb_left;
1019		} else if (file_offset >= entry_end(entry)) {
1020			node = node->rb_right;
1021		} else {
 
 
 
1022			/*
1023			 * Direct hit, got an ordered extent that starts at
1024			 * @file_offset
 
1025			 */
 
 
 
 
 
 
 
1026			goto out;
1027		}
1028	}
1029	if (!entry) {
1030		/* Empty tree */
1031		goto out;
1032	}
1033
1034	cur = &entry->rb_node;
1035	/* We got an entry around @file_offset, check adjacent entries */
1036	if (entry->file_offset < file_offset) {
1037		prev = cur;
1038		next = rb_next(cur);
1039	} else {
1040		prev = rb_prev(cur);
1041		next = cur;
1042	}
1043	if (prev) {
1044		entry = rb_entry(prev, struct btrfs_ordered_extent, rb_node);
1045		if (range_overlaps(entry, file_offset, len))
1046			goto out;
1047	}
1048	if (next) {
1049		entry = rb_entry(next, struct btrfs_ordered_extent, rb_node);
1050		if (range_overlaps(entry, file_offset, len))
1051			goto out;
1052	}
1053	/* No ordered extent in the range */
1054	entry = NULL;
1055out:
1056	if (entry) {
1057		refcount_inc(&entry->refs);
1058		trace_btrfs_ordered_extent_lookup_first_range(inode, entry);
1059	}
1060
1061	spin_unlock_irq(&inode->ordered_tree_lock);
1062	return entry;
 
 
 
 
1063}
1064
1065/*
1066 * Lock the passed range and ensures all pending ordered extents in it are run
1067 * to completion.
1068 *
1069 * @inode:        Inode whose ordered tree is to be searched
1070 * @start:        Beginning of range to flush
1071 * @end:          Last byte of range to lock
1072 * @cached_state: If passed, will return the extent state responsible for the
1073 *                locked range. It's the caller's responsibility to free the
1074 *                cached state.
1075 *
1076 * Always return with the given range locked, ensuring after it's called no
1077 * order extent can be pending.
1078 */
1079void btrfs_lock_and_flush_ordered_range(struct btrfs_inode *inode, u64 start,
1080					u64 end,
1081					struct extent_state **cached_state)
1082{
 
1083	struct btrfs_ordered_extent *ordered;
1084	struct extent_state *cache = NULL;
1085	struct extent_state **cachedp = &cache;
 
 
 
1086
1087	if (cached_state)
1088		cachedp = cached_state;
 
1089
1090	while (1) {
1091		lock_extent(&inode->io_tree, start, end, cachedp);
1092		ordered = btrfs_lookup_ordered_range(inode, start,
1093						     end - start + 1);
1094		if (!ordered) {
1095			/*
1096			 * If no external cached_state has been passed then
1097			 * decrement the extra ref taken for cachedp since we
1098			 * aren't exposing it outside of this function
1099			 */
1100			if (!cached_state)
1101				refcount_dec(&cache->refs);
1102			break;
 
 
 
1103		}
1104		unlock_extent(&inode->io_tree, start, end, cachedp);
1105		btrfs_start_ordered_extent(ordered);
1106		btrfs_put_ordered_extent(ordered);
1107	}
 
 
 
 
1108}
1109
 
1110/*
1111 * Lock the passed range and ensure all pending ordered extents in it are run
1112 * to completion in nowait mode.
1113 *
1114 * Return true if btrfs_lock_ordered_range does not return any extents,
1115 * otherwise false.
 
 
 
 
 
1116 */
1117bool btrfs_try_lock_ordered_range(struct btrfs_inode *inode, u64 start, u64 end,
1118				  struct extent_state **cached_state)
1119{
1120	struct btrfs_ordered_extent *ordered;
 
1121
1122	if (!try_lock_extent(&inode->io_tree, start, end, cached_state))
1123		return false;
1124
1125	ordered = btrfs_lookup_ordered_range(inode, start, end - start + 1);
1126	if (!ordered)
1127		return true;
1128
1129	btrfs_put_ordered_extent(ordered);
1130	unlock_extent(&inode->io_tree, start, end, cached_state);
1131
1132	return false;
1133}
1134
1135/* Split out a new ordered extent for this first @len bytes of @ordered. */
1136struct btrfs_ordered_extent *btrfs_split_ordered_extent(
1137			struct btrfs_ordered_extent *ordered, u64 len)
1138{
1139	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1140	struct btrfs_root *root = inode->root;
1141	struct btrfs_fs_info *fs_info = root->fs_info;
1142	u64 file_offset = ordered->file_offset;
1143	u64 disk_bytenr = ordered->disk_bytenr;
1144	unsigned long flags = ordered->flags;
1145	struct btrfs_ordered_sum *sum, *tmpsum;
1146	struct btrfs_ordered_extent *new;
1147	struct rb_node *node;
1148	u64 offset = 0;
1149
1150	trace_btrfs_ordered_extent_split(inode, ordered);
1151
1152	ASSERT(!(flags & (1U << BTRFS_ORDERED_COMPRESSED)));
1153
1154	/*
1155	 * The entire bio must be covered by the ordered extent, but we can't
1156	 * reduce the original extent to a zero length either.
1157	 */
1158	if (WARN_ON_ONCE(len >= ordered->num_bytes))
1159		return ERR_PTR(-EINVAL);
1160	/* We cannot split partially completed ordered extents. */
1161	if (ordered->bytes_left) {
1162		ASSERT(!(flags & ~BTRFS_ORDERED_TYPE_FLAGS));
1163		if (WARN_ON_ONCE(ordered->bytes_left != ordered->disk_num_bytes))
1164			return ERR_PTR(-EINVAL);
1165	}
1166	/* We cannot split a compressed ordered extent. */
1167	if (WARN_ON_ONCE(ordered->disk_num_bytes != ordered->num_bytes))
1168		return ERR_PTR(-EINVAL);
1169
1170	new = alloc_ordered_extent(inode, file_offset, len, len, disk_bytenr,
1171				   len, 0, flags, ordered->compress_type);
1172	if (IS_ERR(new))
1173		return new;
1174
1175	/* One ref for the tree. */
1176	refcount_inc(&new->refs);
1177
1178	spin_lock_irq(&root->ordered_extent_lock);
1179	spin_lock(&inode->ordered_tree_lock);
1180	/* Remove from tree once */
1181	node = &ordered->rb_node;
1182	rb_erase(node, &inode->ordered_tree);
1183	RB_CLEAR_NODE(node);
1184	if (inode->ordered_tree_last == node)
1185		inode->ordered_tree_last = NULL;
1186
1187	ordered->file_offset += len;
1188	ordered->disk_bytenr += len;
1189	ordered->num_bytes -= len;
1190	ordered->disk_num_bytes -= len;
1191	ordered->ram_bytes -= len;
1192
1193	if (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags)) {
1194		ASSERT(ordered->bytes_left == 0);
1195		new->bytes_left = 0;
1196	} else {
1197		ordered->bytes_left -= len;
1198	}
1199
1200	if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) {
1201		if (ordered->truncated_len > len) {
1202			ordered->truncated_len -= len;
1203		} else {
1204			new->truncated_len = ordered->truncated_len;
1205			ordered->truncated_len = 0;
1206		}
1207	}
1208
1209	list_for_each_entry_safe(sum, tmpsum, &ordered->list, list) {
1210		if (offset == len)
1211			break;
1212		list_move_tail(&sum->list, &new->list);
1213		offset += sum->len;
1214	}
1215
1216	/* Re-insert the node */
1217	node = tree_insert(&inode->ordered_tree, ordered->file_offset,
1218			   &ordered->rb_node);
1219	if (node)
1220		btrfs_panic(fs_info, -EEXIST,
1221			"zoned: inconsistency in ordered tree at offset %llu",
1222			ordered->file_offset);
1223
1224	node = tree_insert(&inode->ordered_tree, new->file_offset, &new->rb_node);
1225	if (node)
1226		btrfs_panic(fs_info, -EEXIST,
1227			"zoned: inconsistency in ordered tree at offset %llu",
1228			new->file_offset);
1229	spin_unlock(&inode->ordered_tree_lock);
1230
1231	list_add_tail(&new->root_extent_list, &root->ordered_extents);
1232	root->nr_ordered_extents++;
1233	spin_unlock_irq(&root->ordered_extent_lock);
1234	return new;
1235}
1236
1237int __init ordered_data_init(void)
1238{
1239	btrfs_ordered_extent_cache = KMEM_CACHE(btrfs_ordered_extent, 0);
 
 
 
1240	if (!btrfs_ordered_extent_cache)
1241		return -ENOMEM;
1242
1243	return 0;
1244}
1245
1246void __cold ordered_data_exit(void)
1247{
1248	kmem_cache_destroy(btrfs_ordered_extent_cache);
 
1249}