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/fs.h>
  20#include <linux/pagemap.h>
  21#include <linux/highmem.h>
  22#include <linux/time.h>
  23#include <linux/init.h>
  24#include <linux/string.h>
  25#include <linux/backing-dev.h>
  26#include <linux/mpage.h>
  27#include <linux/falloc.h>
  28#include <linux/swap.h>
  29#include <linux/writeback.h>
  30#include <linux/statfs.h>
  31#include <linux/compat.h>
  32#include <linux/slab.h>
  33#include "ctree.h"
  34#include "disk-io.h"
  35#include "transaction.h"
  36#include "btrfs_inode.h"
  37#include "ioctl.h"
  38#include "print-tree.h"
  39#include "tree-log.h"
  40#include "locking.h"
  41#include "compat.h"
  42
  43/*
  44 * when auto defrag is enabled we
  45 * queue up these defrag structs to remember which
  46 * inodes need defragging passes
  47 */
  48struct inode_defrag {
  49	struct rb_node rb_node;
  50	/* objectid */
  51	u64 ino;
  52	/*
  53	 * transid where the defrag was added, we search for
  54	 * extents newer than this
  55	 */
  56	u64 transid;
  57
  58	/* root objectid */
  59	u64 root;
  60
  61	/* last offset we were able to defrag */
  62	u64 last_offset;
  63
  64	/* if we've wrapped around back to zero once already */
  65	int cycled;
  66};
  67
  68/* pop a record for an inode into the defrag tree.  The lock
  69 * must be held already
  70 *
  71 * If you're inserting a record for an older transid than an
  72 * existing record, the transid already in the tree is lowered
  73 *
  74 * If an existing record is found the defrag item you
  75 * pass in is freed
  76 */
  77static void __btrfs_add_inode_defrag(struct inode *inode,
  78				    struct inode_defrag *defrag)
  79{
  80	struct btrfs_root *root = BTRFS_I(inode)->root;
  81	struct inode_defrag *entry;
  82	struct rb_node **p;
  83	struct rb_node *parent = NULL;
  84
  85	p = &root->fs_info->defrag_inodes.rb_node;
  86	while (*p) {
  87		parent = *p;
  88		entry = rb_entry(parent, struct inode_defrag, rb_node);
  89
  90		if (defrag->ino < entry->ino)
  91			p = &parent->rb_left;
  92		else if (defrag->ino > entry->ino)
  93			p = &parent->rb_right;
  94		else {
  95			/* if we're reinserting an entry for
  96			 * an old defrag run, make sure to
  97			 * lower the transid of our existing record
  98			 */
  99			if (defrag->transid < entry->transid)
 100				entry->transid = defrag->transid;
 101			if (defrag->last_offset > entry->last_offset)
 102				entry->last_offset = defrag->last_offset;
 103			goto exists;
 104		}
 105	}
 106	BTRFS_I(inode)->in_defrag = 1;
 107	rb_link_node(&defrag->rb_node, parent, p);
 108	rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes);
 109	return;
 110
 111exists:
 112	kfree(defrag);
 113	return;
 114
 115}
 116
 117/*
 118 * insert a defrag record for this inode if auto defrag is
 119 * enabled
 120 */
 121int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
 122			   struct inode *inode)
 123{
 124	struct btrfs_root *root = BTRFS_I(inode)->root;
 125	struct inode_defrag *defrag;
 126	u64 transid;
 127
 128	if (!btrfs_test_opt(root, AUTO_DEFRAG))
 129		return 0;
 130
 131	if (btrfs_fs_closing(root->fs_info))
 132		return 0;
 133
 134	if (BTRFS_I(inode)->in_defrag)
 135		return 0;
 136
 137	if (trans)
 138		transid = trans->transid;
 139	else
 140		transid = BTRFS_I(inode)->root->last_trans;
 141
 142	defrag = kzalloc(sizeof(*defrag), GFP_NOFS);
 143	if (!defrag)
 144		return -ENOMEM;
 145
 146	defrag->ino = btrfs_ino(inode);
 147	defrag->transid = transid;
 148	defrag->root = root->root_key.objectid;
 149
 150	spin_lock(&root->fs_info->defrag_inodes_lock);
 151	if (!BTRFS_I(inode)->in_defrag)
 152		__btrfs_add_inode_defrag(inode, defrag);
 153	else
 154		kfree(defrag);
 155	spin_unlock(&root->fs_info->defrag_inodes_lock);
 156	return 0;
 157}
 158
 159/*
 160 * must be called with the defrag_inodes lock held
 161 */
 162struct inode_defrag *btrfs_find_defrag_inode(struct btrfs_fs_info *info, u64 ino,
 163					     struct rb_node **next)
 164{
 165	struct inode_defrag *entry = NULL;
 166	struct rb_node *p;
 167	struct rb_node *parent = NULL;
 168
 169	p = info->defrag_inodes.rb_node;
 170	while (p) {
 171		parent = p;
 172		entry = rb_entry(parent, struct inode_defrag, rb_node);
 173
 174		if (ino < entry->ino)
 175			p = parent->rb_left;
 176		else if (ino > entry->ino)
 177			p = parent->rb_right;
 178		else
 179			return entry;
 180	}
 181
 182	if (next) {
 183		while (parent && ino > entry->ino) {
 184			parent = rb_next(parent);
 185			entry = rb_entry(parent, struct inode_defrag, rb_node);
 186		}
 187		*next = parent;
 188	}
 189	return NULL;
 190}
 191
 192/*
 193 * run through the list of inodes in the FS that need
 194 * defragging
 195 */
 196int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info)
 197{
 198	struct inode_defrag *defrag;
 199	struct btrfs_root *inode_root;
 200	struct inode *inode;
 201	struct rb_node *n;
 202	struct btrfs_key key;
 203	struct btrfs_ioctl_defrag_range_args range;
 204	u64 first_ino = 0;
 205	int num_defrag;
 206	int defrag_batch = 1024;
 207
 208	memset(&range, 0, sizeof(range));
 209	range.len = (u64)-1;
 210
 211	atomic_inc(&fs_info->defrag_running);
 212	spin_lock(&fs_info->defrag_inodes_lock);
 213	while(1) {
 214		n = NULL;
 215
 216		/* find an inode to defrag */
 217		defrag = btrfs_find_defrag_inode(fs_info, first_ino, &n);
 218		if (!defrag) {
 219			if (n)
 220				defrag = rb_entry(n, struct inode_defrag, rb_node);
 221			else if (first_ino) {
 222				first_ino = 0;
 223				continue;
 224			} else {
 225				break;
 226			}
 227		}
 228
 229		/* remove it from the rbtree */
 230		first_ino = defrag->ino + 1;
 231		rb_erase(&defrag->rb_node, &fs_info->defrag_inodes);
 232
 233		if (btrfs_fs_closing(fs_info))
 234			goto next_free;
 235
 236		spin_unlock(&fs_info->defrag_inodes_lock);
 237
 238		/* get the inode */
 239		key.objectid = defrag->root;
 240		btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 241		key.offset = (u64)-1;
 242		inode_root = btrfs_read_fs_root_no_name(fs_info, &key);
 243		if (IS_ERR(inode_root))
 244			goto next;
 245
 246		key.objectid = defrag->ino;
 247		btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 248		key.offset = 0;
 249
 250		inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL);
 251		if (IS_ERR(inode))
 252			goto next;
 253
 254		/* do a chunk of defrag */
 255		BTRFS_I(inode)->in_defrag = 0;
 256		range.start = defrag->last_offset;
 257		num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid,
 258					       defrag_batch);
 259		/*
 260		 * if we filled the whole defrag batch, there
 261		 * must be more work to do.  Queue this defrag
 262		 * again
 263		 */
 264		if (num_defrag == defrag_batch) {
 265			defrag->last_offset = range.start;
 266			__btrfs_add_inode_defrag(inode, defrag);
 267			/*
 268			 * we don't want to kfree defrag, we added it back to
 269			 * the rbtree
 270			 */
 271			defrag = NULL;
 272		} else if (defrag->last_offset && !defrag->cycled) {
 273			/*
 274			 * we didn't fill our defrag batch, but
 275			 * we didn't start at zero.  Make sure we loop
 276			 * around to the start of the file.
 277			 */
 278			defrag->last_offset = 0;
 279			defrag->cycled = 1;
 280			__btrfs_add_inode_defrag(inode, defrag);
 281			defrag = NULL;
 282		}
 283
 284		iput(inode);
 285next:
 286		spin_lock(&fs_info->defrag_inodes_lock);
 287next_free:
 288		kfree(defrag);
 289	}
 290	spin_unlock(&fs_info->defrag_inodes_lock);
 291
 292	atomic_dec(&fs_info->defrag_running);
 293
 294	/*
 295	 * during unmount, we use the transaction_wait queue to
 296	 * wait for the defragger to stop
 297	 */
 298	wake_up(&fs_info->transaction_wait);
 299	return 0;
 300}
 301
 302/* simple helper to fault in pages and copy.  This should go away
 303 * and be replaced with calls into generic code.
 304 */
 305static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
 306					 size_t write_bytes,
 307					 struct page **prepared_pages,
 308					 struct iov_iter *i)
 309{
 310	size_t copied = 0;
 311	size_t total_copied = 0;
 312	int pg = 0;
 313	int offset = pos & (PAGE_CACHE_SIZE - 1);
 314
 315	while (write_bytes > 0) {
 316		size_t count = min_t(size_t,
 317				     PAGE_CACHE_SIZE - offset, write_bytes);
 318		struct page *page = prepared_pages[pg];
 319		/*
 320		 * Copy data from userspace to the current page
 321		 *
 322		 * Disable pagefault to avoid recursive lock since
 323		 * the pages are already locked
 324		 */
 325		pagefault_disable();
 326		copied = iov_iter_copy_from_user_atomic(page, i, offset, count);
 327		pagefault_enable();
 328
 329		/* Flush processor's dcache for this page */
 330		flush_dcache_page(page);
 331
 332		/*
 333		 * if we get a partial write, we can end up with
 334		 * partially up to date pages.  These add
 335		 * a lot of complexity, so make sure they don't
 336		 * happen by forcing this copy to be retried.
 337		 *
 338		 * The rest of the btrfs_file_write code will fall
 339		 * back to page at a time copies after we return 0.
 340		 */
 341		if (!PageUptodate(page) && copied < count)
 342			copied = 0;
 343
 344		iov_iter_advance(i, copied);
 345		write_bytes -= copied;
 346		total_copied += copied;
 347
 348		/* Return to btrfs_file_aio_write to fault page */
 349		if (unlikely(copied == 0))
 350			break;
 351
 352		if (unlikely(copied < PAGE_CACHE_SIZE - offset)) {
 353			offset += copied;
 354		} else {
 355			pg++;
 356			offset = 0;
 357		}
 358	}
 359	return total_copied;
 360}
 361
 362/*
 363 * unlocks pages after btrfs_file_write is done with them
 364 */
 365void btrfs_drop_pages(struct page **pages, size_t num_pages)
 366{
 367	size_t i;
 368	for (i = 0; i < num_pages; i++) {
 369		/* page checked is some magic around finding pages that
 370		 * have been modified without going through btrfs_set_page_dirty
 371		 * clear it here
 372		 */
 373		ClearPageChecked(pages[i]);
 374		unlock_page(pages[i]);
 375		mark_page_accessed(pages[i]);
 376		page_cache_release(pages[i]);
 377	}
 378}
 379
 380/*
 381 * after copy_from_user, pages need to be dirtied and we need to make
 382 * sure holes are created between the current EOF and the start of
 383 * any next extents (if required).
 384 *
 385 * this also makes the decision about creating an inline extent vs
 386 * doing real data extents, marking pages dirty and delalloc as required.
 387 */
 388int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode,
 389		      struct page **pages, size_t num_pages,
 390		      loff_t pos, size_t write_bytes,
 391		      struct extent_state **cached)
 392{
 393	int err = 0;
 394	int i;
 395	u64 num_bytes;
 396	u64 start_pos;
 397	u64 end_of_last_block;
 398	u64 end_pos = pos + write_bytes;
 399	loff_t isize = i_size_read(inode);
 400
 401	start_pos = pos & ~((u64)root->sectorsize - 1);
 402	num_bytes = (write_bytes + pos - start_pos +
 403		    root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 404
 405	end_of_last_block = start_pos + num_bytes - 1;
 406	err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block,
 407					cached);
 408	if (err)
 409		return err;
 410
 411	for (i = 0; i < num_pages; i++) {
 412		struct page *p = pages[i];
 413		SetPageUptodate(p);
 414		ClearPageChecked(p);
 415		set_page_dirty(p);
 416	}
 417
 418	/*
 419	 * we've only changed i_size in ram, and we haven't updated
 420	 * the disk i_size.  There is no need to log the inode
 421	 * at this time.
 422	 */
 423	if (end_pos > isize)
 424		i_size_write(inode, end_pos);
 425	return 0;
 426}
 427
 428/*
 429 * this drops all the extents in the cache that intersect the range
 430 * [start, end].  Existing extents are split as required.
 431 */
 432int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 433			    int skip_pinned)
 434{
 435	struct extent_map *em;
 436	struct extent_map *split = NULL;
 437	struct extent_map *split2 = NULL;
 438	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 439	u64 len = end - start + 1;
 440	int ret;
 441	int testend = 1;
 442	unsigned long flags;
 443	int compressed = 0;
 444
 445	WARN_ON(end < start);
 446	if (end == (u64)-1) {
 447		len = (u64)-1;
 448		testend = 0;
 449	}
 450	while (1) {
 451		if (!split)
 452			split = alloc_extent_map();
 453		if (!split2)
 454			split2 = alloc_extent_map();
 455		BUG_ON(!split || !split2);
 456
 457		write_lock(&em_tree->lock);
 458		em = lookup_extent_mapping(em_tree, start, len);
 459		if (!em) {
 460			write_unlock(&em_tree->lock);
 461			break;
 462		}
 463		flags = em->flags;
 464		if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 465			if (testend && em->start + em->len >= start + len) {
 466				free_extent_map(em);
 467				write_unlock(&em_tree->lock);
 468				break;
 469			}
 470			start = em->start + em->len;
 471			if (testend)
 472				len = start + len - (em->start + em->len);
 473			free_extent_map(em);
 474			write_unlock(&em_tree->lock);
 475			continue;
 476		}
 477		compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 478		clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 479		remove_extent_mapping(em_tree, em);
 480
 481		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 482		    em->start < start) {
 483			split->start = em->start;
 484			split->len = start - em->start;
 485			split->orig_start = em->orig_start;
 486			split->block_start = em->block_start;
 487
 488			if (compressed)
 489				split->block_len = em->block_len;
 490			else
 491				split->block_len = split->len;
 492
 493			split->bdev = em->bdev;
 494			split->flags = flags;
 495			split->compress_type = em->compress_type;
 496			ret = add_extent_mapping(em_tree, split);
 497			BUG_ON(ret);
 498			free_extent_map(split);
 499			split = split2;
 500			split2 = NULL;
 501		}
 502		if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 503		    testend && em->start + em->len > start + len) {
 504			u64 diff = start + len - em->start;
 505
 506			split->start = start + len;
 507			split->len = em->start + em->len - (start + len);
 508			split->bdev = em->bdev;
 509			split->flags = flags;
 510			split->compress_type = em->compress_type;
 511
 512			if (compressed) {
 513				split->block_len = em->block_len;
 514				split->block_start = em->block_start;
 515				split->orig_start = em->orig_start;
 516			} else {
 517				split->block_len = split->len;
 518				split->block_start = em->block_start + diff;
 519				split->orig_start = split->start;
 520			}
 521
 522			ret = add_extent_mapping(em_tree, split);
 523			BUG_ON(ret);
 524			free_extent_map(split);
 525			split = NULL;
 526		}
 527		write_unlock(&em_tree->lock);
 528
 529		/* once for us */
 530		free_extent_map(em);
 531		/* once for the tree*/
 532		free_extent_map(em);
 533	}
 534	if (split)
 535		free_extent_map(split);
 536	if (split2)
 537		free_extent_map(split2);
 538	return 0;
 539}
 540
 541/*
 542 * this is very complex, but the basic idea is to drop all extents
 543 * in the range start - end.  hint_block is filled in with a block number
 544 * that would be a good hint to the block allocator for this file.
 545 *
 546 * If an extent intersects the range but is not entirely inside the range
 547 * it is either truncated or split.  Anything entirely inside the range
 548 * is deleted from the tree.
 549 */
 550int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 551		       u64 start, u64 end, u64 *hint_byte, int drop_cache)
 552{
 553	struct btrfs_root *root = BTRFS_I(inode)->root;
 554	struct extent_buffer *leaf;
 555	struct btrfs_file_extent_item *fi;
 556	struct btrfs_path *path;
 557	struct btrfs_key key;
 558	struct btrfs_key new_key;
 559	u64 ino = btrfs_ino(inode);
 560	u64 search_start = start;
 561	u64 disk_bytenr = 0;
 562	u64 num_bytes = 0;
 563	u64 extent_offset = 0;
 564	u64 extent_end = 0;
 565	int del_nr = 0;
 566	int del_slot = 0;
 567	int extent_type;
 568	int recow;
 569	int ret;
 570
 571	if (drop_cache)
 572		btrfs_drop_extent_cache(inode, start, end - 1, 0);
 573
 574	path = btrfs_alloc_path();
 575	if (!path)
 576		return -ENOMEM;
 577
 578	while (1) {
 579		recow = 0;
 580		ret = btrfs_lookup_file_extent(trans, root, path, ino,
 581					       search_start, -1);
 582		if (ret < 0)
 583			break;
 584		if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 585			leaf = path->nodes[0];
 586			btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 587			if (key.objectid == ino &&
 588			    key.type == BTRFS_EXTENT_DATA_KEY)
 589				path->slots[0]--;
 590		}
 591		ret = 0;
 592next_slot:
 593		leaf = path->nodes[0];
 594		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 595			BUG_ON(del_nr > 0);
 596			ret = btrfs_next_leaf(root, path);
 597			if (ret < 0)
 598				break;
 599			if (ret > 0) {
 600				ret = 0;
 601				break;
 602			}
 603			leaf = path->nodes[0];
 604			recow = 1;
 605		}
 606
 607		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 608		if (key.objectid > ino ||
 609		    key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 610			break;
 611
 612		fi = btrfs_item_ptr(leaf, path->slots[0],
 613				    struct btrfs_file_extent_item);
 614		extent_type = btrfs_file_extent_type(leaf, fi);
 615
 616		if (extent_type == BTRFS_FILE_EXTENT_REG ||
 617		    extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 618			disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 619			num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 620			extent_offset = btrfs_file_extent_offset(leaf, fi);
 621			extent_end = key.offset +
 622				btrfs_file_extent_num_bytes(leaf, fi);
 623		} else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 624			extent_end = key.offset +
 625				btrfs_file_extent_inline_len(leaf, fi);
 626		} else {
 627			WARN_ON(1);
 628			extent_end = search_start;
 629		}
 630
 631		if (extent_end <= search_start) {
 632			path->slots[0]++;
 633			goto next_slot;
 634		}
 635
 636		search_start = max(key.offset, start);
 637		if (recow) {
 638			btrfs_release_path(path);
 639			continue;
 640		}
 641
 642		/*
 643		 *     | - range to drop - |
 644		 *  | -------- extent -------- |
 645		 */
 646		if (start > key.offset && end < extent_end) {
 647			BUG_ON(del_nr > 0);
 648			BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 649
 650			memcpy(&new_key, &key, sizeof(new_key));
 651			new_key.offset = start;
 652			ret = btrfs_duplicate_item(trans, root, path,
 653						   &new_key);
 654			if (ret == -EAGAIN) {
 655				btrfs_release_path(path);
 656				continue;
 657			}
 658			if (ret < 0)
 659				break;
 660
 661			leaf = path->nodes[0];
 662			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 663					    struct btrfs_file_extent_item);
 664			btrfs_set_file_extent_num_bytes(leaf, fi,
 665							start - key.offset);
 666
 667			fi = btrfs_item_ptr(leaf, path->slots[0],
 668					    struct btrfs_file_extent_item);
 669
 670			extent_offset += start - key.offset;
 671			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 672			btrfs_set_file_extent_num_bytes(leaf, fi,
 673							extent_end - start);
 674			btrfs_mark_buffer_dirty(leaf);
 675
 676			if (disk_bytenr > 0) {
 677				ret = btrfs_inc_extent_ref(trans, root,
 678						disk_bytenr, num_bytes, 0,
 679						root->root_key.objectid,
 680						new_key.objectid,
 681						start - extent_offset);
 682				BUG_ON(ret);
 683				*hint_byte = disk_bytenr;
 684			}
 685			key.offset = start;
 686		}
 687		/*
 688		 *  | ---- range to drop ----- |
 689		 *      | -------- extent -------- |
 690		 */
 691		if (start <= key.offset && end < extent_end) {
 692			BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 693
 694			memcpy(&new_key, &key, sizeof(new_key));
 695			new_key.offset = end;
 696			btrfs_set_item_key_safe(trans, root, path, &new_key);
 697
 698			extent_offset += end - key.offset;
 699			btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 700			btrfs_set_file_extent_num_bytes(leaf, fi,
 701							extent_end - end);
 702			btrfs_mark_buffer_dirty(leaf);
 703			if (disk_bytenr > 0) {
 704				inode_sub_bytes(inode, end - key.offset);
 705				*hint_byte = disk_bytenr;
 706			}
 707			break;
 708		}
 709
 710		search_start = extent_end;
 711		/*
 712		 *       | ---- range to drop ----- |
 713		 *  | -------- extent -------- |
 714		 */
 715		if (start > key.offset && end >= extent_end) {
 716			BUG_ON(del_nr > 0);
 717			BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 718
 719			btrfs_set_file_extent_num_bytes(leaf, fi,
 720							start - key.offset);
 721			btrfs_mark_buffer_dirty(leaf);
 722			if (disk_bytenr > 0) {
 723				inode_sub_bytes(inode, extent_end - start);
 724				*hint_byte = disk_bytenr;
 725			}
 726			if (end == extent_end)
 727				break;
 728
 729			path->slots[0]++;
 730			goto next_slot;
 731		}
 732
 733		/*
 734		 *  | ---- range to drop ----- |
 735		 *    | ------ extent ------ |
 736		 */
 737		if (start <= key.offset && end >= extent_end) {
 738			if (del_nr == 0) {
 739				del_slot = path->slots[0];
 740				del_nr = 1;
 741			} else {
 742				BUG_ON(del_slot + del_nr != path->slots[0]);
 743				del_nr++;
 744			}
 745
 746			if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 747				inode_sub_bytes(inode,
 748						extent_end - key.offset);
 749				extent_end = ALIGN(extent_end,
 750						   root->sectorsize);
 751			} else if (disk_bytenr > 0) {
 752				ret = btrfs_free_extent(trans, root,
 753						disk_bytenr, num_bytes, 0,
 754						root->root_key.objectid,
 755						key.objectid, key.offset -
 756						extent_offset);
 757				BUG_ON(ret);
 758				inode_sub_bytes(inode,
 759						extent_end - key.offset);
 760				*hint_byte = disk_bytenr;
 761			}
 762
 763			if (end == extent_end)
 764				break;
 765
 766			if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 767				path->slots[0]++;
 768				goto next_slot;
 769			}
 770
 771			ret = btrfs_del_items(trans, root, path, del_slot,
 772					      del_nr);
 773			BUG_ON(ret);
 774
 775			del_nr = 0;
 776			del_slot = 0;
 777
 778			btrfs_release_path(path);
 779			continue;
 780		}
 781
 782		BUG_ON(1);
 783	}
 784
 785	if (del_nr > 0) {
 786		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 787		BUG_ON(ret);
 788	}
 789
 790	btrfs_free_path(path);
 791	return ret;
 792}
 793
 794static int extent_mergeable(struct extent_buffer *leaf, int slot,
 795			    u64 objectid, u64 bytenr, u64 orig_offset,
 796			    u64 *start, u64 *end)
 797{
 798	struct btrfs_file_extent_item *fi;
 799	struct btrfs_key key;
 800	u64 extent_end;
 801
 802	if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 803		return 0;
 804
 805	btrfs_item_key_to_cpu(leaf, &key, slot);
 806	if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 807		return 0;
 808
 809	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 810	if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 811	    btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 812	    btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset ||
 813	    btrfs_file_extent_compression(leaf, fi) ||
 814	    btrfs_file_extent_encryption(leaf, fi) ||
 815	    btrfs_file_extent_other_encoding(leaf, fi))
 816		return 0;
 817
 818	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 819	if ((*start && *start != key.offset) || (*end && *end != extent_end))
 820		return 0;
 821
 822	*start = key.offset;
 823	*end = extent_end;
 824	return 1;
 825}
 826
 827/*
 828 * Mark extent in the range start - end as written.
 829 *
 830 * This changes extent type from 'pre-allocated' to 'regular'. If only
 831 * part of extent is marked as written, the extent will be split into
 832 * two or three.
 833 */
 834int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 835			      struct inode *inode, u64 start, u64 end)
 836{
 837	struct btrfs_root *root = BTRFS_I(inode)->root;
 838	struct extent_buffer *leaf;
 839	struct btrfs_path *path;
 840	struct btrfs_file_extent_item *fi;
 841	struct btrfs_key key;
 842	struct btrfs_key new_key;
 843	u64 bytenr;
 844	u64 num_bytes;
 845	u64 extent_end;
 846	u64 orig_offset;
 847	u64 other_start;
 848	u64 other_end;
 849	u64 split;
 850	int del_nr = 0;
 851	int del_slot = 0;
 852	int recow;
 853	int ret;
 854	u64 ino = btrfs_ino(inode);
 855
 856	btrfs_drop_extent_cache(inode, start, end - 1, 0);
 857
 858	path = btrfs_alloc_path();
 859	if (!path)
 860		return -ENOMEM;
 861again:
 862	recow = 0;
 863	split = start;
 864	key.objectid = ino;
 865	key.type = BTRFS_EXTENT_DATA_KEY;
 866	key.offset = split;
 867
 868	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 869	if (ret < 0)
 870		goto out;
 871	if (ret > 0 && path->slots[0] > 0)
 872		path->slots[0]--;
 873
 874	leaf = path->nodes[0];
 875	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 876	BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY);
 877	fi = btrfs_item_ptr(leaf, path->slots[0],
 878			    struct btrfs_file_extent_item);
 879	BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 880	       BTRFS_FILE_EXTENT_PREALLOC);
 881	extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 882	BUG_ON(key.offset > start || extent_end < end);
 883
 884	bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 885	num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 886	orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 887	memcpy(&new_key, &key, sizeof(new_key));
 888
 889	if (start == key.offset && end < extent_end) {
 890		other_start = 0;
 891		other_end = start;
 892		if (extent_mergeable(leaf, path->slots[0] - 1,
 893				     ino, bytenr, orig_offset,
 894				     &other_start, &other_end)) {
 895			new_key.offset = end;
 896			btrfs_set_item_key_safe(trans, root, path, &new_key);
 897			fi = btrfs_item_ptr(leaf, path->slots[0],
 898					    struct btrfs_file_extent_item);
 899			btrfs_set_file_extent_num_bytes(leaf, fi,
 900							extent_end - end);
 901			btrfs_set_file_extent_offset(leaf, fi,
 902						     end - orig_offset);
 903			fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 904					    struct btrfs_file_extent_item);
 905			btrfs_set_file_extent_num_bytes(leaf, fi,
 906							end - other_start);
 907			btrfs_mark_buffer_dirty(leaf);
 908			goto out;
 909		}
 910	}
 911
 912	if (start > key.offset && end == extent_end) {
 913		other_start = end;
 914		other_end = 0;
 915		if (extent_mergeable(leaf, path->slots[0] + 1,
 916				     ino, bytenr, orig_offset,
 917				     &other_start, &other_end)) {
 918			fi = btrfs_item_ptr(leaf, path->slots[0],
 919					    struct btrfs_file_extent_item);
 920			btrfs_set_file_extent_num_bytes(leaf, fi,
 921							start - key.offset);
 922			path->slots[0]++;
 923			new_key.offset = start;
 924			btrfs_set_item_key_safe(trans, root, path, &new_key);
 925
 926			fi = btrfs_item_ptr(leaf, path->slots[0],
 927					    struct btrfs_file_extent_item);
 928			btrfs_set_file_extent_num_bytes(leaf, fi,
 929							other_end - start);
 930			btrfs_set_file_extent_offset(leaf, fi,
 931						     start - orig_offset);
 932			btrfs_mark_buffer_dirty(leaf);
 933			goto out;
 934		}
 935	}
 936
 937	while (start > key.offset || end < extent_end) {
 938		if (key.offset == start)
 939			split = end;
 940
 941		new_key.offset = split;
 942		ret = btrfs_duplicate_item(trans, root, path, &new_key);
 943		if (ret == -EAGAIN) {
 944			btrfs_release_path(path);
 945			goto again;
 946		}
 947		BUG_ON(ret < 0);
 948
 949		leaf = path->nodes[0];
 950		fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 951				    struct btrfs_file_extent_item);
 952		btrfs_set_file_extent_num_bytes(leaf, fi,
 953						split - key.offset);
 954
 955		fi = btrfs_item_ptr(leaf, path->slots[0],
 956				    struct btrfs_file_extent_item);
 957
 958		btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 959		btrfs_set_file_extent_num_bytes(leaf, fi,
 960						extent_end - split);
 961		btrfs_mark_buffer_dirty(leaf);
 962
 963		ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 964					   root->root_key.objectid,
 965					   ino, orig_offset);
 966		BUG_ON(ret);
 967
 968		if (split == start) {
 969			key.offset = start;
 970		} else {
 971			BUG_ON(start != key.offset);
 972			path->slots[0]--;
 973			extent_end = end;
 974		}
 975		recow = 1;
 976	}
 977
 978	other_start = end;
 979	other_end = 0;
 980	if (extent_mergeable(leaf, path->slots[0] + 1,
 981			     ino, bytenr, orig_offset,
 982			     &other_start, &other_end)) {
 983		if (recow) {
 984			btrfs_release_path(path);
 985			goto again;
 986		}
 987		extent_end = other_end;
 988		del_slot = path->slots[0] + 1;
 989		del_nr++;
 990		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 991					0, root->root_key.objectid,
 992					ino, orig_offset);
 993		BUG_ON(ret);
 994	}
 995	other_start = 0;
 996	other_end = start;
 997	if (extent_mergeable(leaf, path->slots[0] - 1,
 998			     ino, bytenr, orig_offset,
 999			     &other_start, &other_end)) {
1000		if (recow) {
1001			btrfs_release_path(path);
1002			goto again;
1003		}
1004		key.offset = other_start;
1005		del_slot = path->slots[0];
1006		del_nr++;
1007		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1008					0, root->root_key.objectid,
1009					ino, orig_offset);
1010		BUG_ON(ret);
1011	}
1012	if (del_nr == 0) {
1013		fi = btrfs_item_ptr(leaf, path->slots[0],
1014			   struct btrfs_file_extent_item);
1015		btrfs_set_file_extent_type(leaf, fi,
1016					   BTRFS_FILE_EXTENT_REG);
1017		btrfs_mark_buffer_dirty(leaf);
1018	} else {
1019		fi = btrfs_item_ptr(leaf, del_slot - 1,
1020			   struct btrfs_file_extent_item);
1021		btrfs_set_file_extent_type(leaf, fi,
1022					   BTRFS_FILE_EXTENT_REG);
1023		btrfs_set_file_extent_num_bytes(leaf, fi,
1024						extent_end - key.offset);
1025		btrfs_mark_buffer_dirty(leaf);
1026
1027		ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
1028		BUG_ON(ret);
1029	}
1030out:
1031	btrfs_free_path(path);
1032	return 0;
1033}
1034
1035/*
1036 * on error we return an unlocked page and the error value
1037 * on success we return a locked page and 0
1038 */
1039static int prepare_uptodate_page(struct page *page, u64 pos,
1040				 bool force_uptodate)
1041{
1042	int ret = 0;
1043
1044	if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) &&
1045	    !PageUptodate(page)) {
1046		ret = btrfs_readpage(NULL, page);
1047		if (ret)
1048			return ret;
1049		lock_page(page);
1050		if (!PageUptodate(page)) {
1051			unlock_page(page);
1052			return -EIO;
1053		}
1054	}
1055	return 0;
1056}
1057
1058/*
1059 * this gets pages into the page cache and locks them down, it also properly
1060 * waits for data=ordered extents to finish before allowing the pages to be
1061 * modified.
1062 */
1063static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
1064			 struct page **pages, size_t num_pages,
1065			 loff_t pos, unsigned long first_index,
1066			 size_t write_bytes, bool force_uptodate)
1067{
1068	struct extent_state *cached_state = NULL;
1069	int i;
1070	unsigned long index = pos >> PAGE_CACHE_SHIFT;
1071	struct inode *inode = fdentry(file)->d_inode;
1072	int err = 0;
1073	int faili = 0;
1074	u64 start_pos;
1075	u64 last_pos;
1076
1077	start_pos = pos & ~((u64)root->sectorsize - 1);
1078	last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
1079
1080again:
1081	for (i = 0; i < num_pages; i++) {
1082		pages[i] = find_or_create_page(inode->i_mapping, index + i,
1083					       GFP_NOFS);
1084		if (!pages[i]) {
1085			faili = i - 1;
1086			err = -ENOMEM;
1087			goto fail;
1088		}
1089
1090		if (i == 0)
1091			err = prepare_uptodate_page(pages[i], pos,
1092						    force_uptodate);
1093		if (i == num_pages - 1)
1094			err = prepare_uptodate_page(pages[i],
1095						    pos + write_bytes, false);
1096		if (err) {
1097			page_cache_release(pages[i]);
1098			faili = i - 1;
1099			goto fail;
1100		}
1101		wait_on_page_writeback(pages[i]);
1102	}
1103	err = 0;
1104	if (start_pos < inode->i_size) {
1105		struct btrfs_ordered_extent *ordered;
1106		lock_extent_bits(&BTRFS_I(inode)->io_tree,
1107				 start_pos, last_pos - 1, 0, &cached_state,
1108				 GFP_NOFS);
1109		ordered = btrfs_lookup_first_ordered_extent(inode,
1110							    last_pos - 1);
1111		if (ordered &&
1112		    ordered->file_offset + ordered->len > start_pos &&
1113		    ordered->file_offset < last_pos) {
1114			btrfs_put_ordered_extent(ordered);
1115			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1116					     start_pos, last_pos - 1,
1117					     &cached_state, GFP_NOFS);
1118			for (i = 0; i < num_pages; i++) {
1119				unlock_page(pages[i]);
1120				page_cache_release(pages[i]);
1121			}
1122			btrfs_wait_ordered_range(inode, start_pos,
1123						 last_pos - start_pos);
1124			goto again;
1125		}
1126		if (ordered)
1127			btrfs_put_ordered_extent(ordered);
1128
1129		clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos,
1130				  last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
1131				  EXTENT_DO_ACCOUNTING, 0, 0, &cached_state,
1132				  GFP_NOFS);
1133		unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1134				     start_pos, last_pos - 1, &cached_state,
1135				     GFP_NOFS);
1136	}
1137	for (i = 0; i < num_pages; i++) {
1138		clear_page_dirty_for_io(pages[i]);
1139		set_page_extent_mapped(pages[i]);
1140		WARN_ON(!PageLocked(pages[i]));
1141	}
1142	return 0;
1143fail:
1144	while (faili >= 0) {
1145		unlock_page(pages[faili]);
1146		page_cache_release(pages[faili]);
1147		faili--;
1148	}
1149	return err;
1150
1151}
1152
1153static noinline ssize_t __btrfs_buffered_write(struct file *file,
1154					       struct iov_iter *i,
1155					       loff_t pos)
1156{
1157	struct inode *inode = fdentry(file)->d_inode;
1158	struct btrfs_root *root = BTRFS_I(inode)->root;
1159	struct page **pages = NULL;
1160	unsigned long first_index;
1161	size_t num_written = 0;
1162	int nrptrs;
1163	int ret = 0;
1164	bool force_page_uptodate = false;
1165
1166	nrptrs = min((iov_iter_count(i) + PAGE_CACHE_SIZE - 1) /
1167		     PAGE_CACHE_SIZE, PAGE_CACHE_SIZE /
1168		     (sizeof(struct page *)));
1169	pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
1170	if (!pages)
1171		return -ENOMEM;
1172
1173	first_index = pos >> PAGE_CACHE_SHIFT;
1174
1175	while (iov_iter_count(i) > 0) {
1176		size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1177		size_t write_bytes = min(iov_iter_count(i),
1178					 nrptrs * (size_t)PAGE_CACHE_SIZE -
1179					 offset);
1180		size_t num_pages = (write_bytes + offset +
1181				    PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1182		size_t dirty_pages;
1183		size_t copied;
1184
1185		WARN_ON(num_pages > nrptrs);
1186
1187		/*
1188		 * Fault pages before locking them in prepare_pages
1189		 * to avoid recursive lock
1190		 */
1191		if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) {
1192			ret = -EFAULT;
1193			break;
1194		}
1195
1196		ret = btrfs_delalloc_reserve_space(inode,
1197					num_pages << PAGE_CACHE_SHIFT);
1198		if (ret)
1199			break;
1200
1201		/*
1202		 * This is going to setup the pages array with the number of
1203		 * pages we want, so we don't really need to worry about the
1204		 * contents of pages from loop to loop
1205		 */
1206		ret = prepare_pages(root, file, pages, num_pages,
1207				    pos, first_index, write_bytes,
1208				    force_page_uptodate);
1209		if (ret) {
1210			btrfs_delalloc_release_space(inode,
1211					num_pages << PAGE_CACHE_SHIFT);
1212			break;
1213		}
1214
1215		copied = btrfs_copy_from_user(pos, num_pages,
1216					   write_bytes, pages, i);
1217
1218		/*
1219		 * if we have trouble faulting in the pages, fall
1220		 * back to one page at a time
1221		 */
1222		if (copied < write_bytes)
1223			nrptrs = 1;
1224
1225		if (copied == 0) {
1226			force_page_uptodate = true;
1227			dirty_pages = 0;
1228		} else {
1229			force_page_uptodate = false;
1230			dirty_pages = (copied + offset +
1231				       PAGE_CACHE_SIZE - 1) >>
1232				       PAGE_CACHE_SHIFT;
1233		}
1234
1235		/*
1236		 * If we had a short copy we need to release the excess delaloc
1237		 * bytes we reserved.  We need to increment outstanding_extents
1238		 * because btrfs_delalloc_release_space will decrement it, but
1239		 * we still have an outstanding extent for the chunk we actually
1240		 * managed to copy.
1241		 */
1242		if (num_pages > dirty_pages) {
1243			if (copied > 0) {
1244				spin_lock(&BTRFS_I(inode)->lock);
1245				BTRFS_I(inode)->outstanding_extents++;
1246				spin_unlock(&BTRFS_I(inode)->lock);
1247			}
1248			btrfs_delalloc_release_space(inode,
1249					(num_pages - dirty_pages) <<
1250					PAGE_CACHE_SHIFT);
1251		}
1252
1253		if (copied > 0) {
1254			ret = btrfs_dirty_pages(root, inode, pages,
1255						dirty_pages, pos, copied,
1256						NULL);
1257			if (ret) {
1258				btrfs_delalloc_release_space(inode,
1259					dirty_pages << PAGE_CACHE_SHIFT);
1260				btrfs_drop_pages(pages, num_pages);
1261				break;
1262			}
1263		}
1264
1265		btrfs_drop_pages(pages, num_pages);
1266
1267		cond_resched();
1268
1269		balance_dirty_pages_ratelimited_nr(inode->i_mapping,
1270						   dirty_pages);
1271		if (dirty_pages < (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
1272			btrfs_btree_balance_dirty(root, 1);
1273		btrfs_throttle(root);
1274
1275		pos += copied;
1276		num_written += copied;
1277	}
1278
1279	kfree(pages);
1280
1281	return num_written ? num_written : ret;
1282}
1283
1284static ssize_t __btrfs_direct_write(struct kiocb *iocb,
1285				    const struct iovec *iov,
1286				    unsigned long nr_segs, loff_t pos,
1287				    loff_t *ppos, size_t count, size_t ocount)
1288{
1289	struct file *file = iocb->ki_filp;
1290	struct inode *inode = fdentry(file)->d_inode;
1291	struct iov_iter i;
1292	ssize_t written;
1293	ssize_t written_buffered;
1294	loff_t endbyte;
1295	int err;
1296
1297	written = generic_file_direct_write(iocb, iov, &nr_segs, pos, ppos,
1298					    count, ocount);
1299
1300	/*
1301	 * the generic O_DIRECT will update in-memory i_size after the
1302	 * DIOs are done.  But our endio handlers that update the on
1303	 * disk i_size never update past the in memory i_size.  So we
1304	 * need one more update here to catch any additions to the
1305	 * file
1306	 */
1307	if (inode->i_size != BTRFS_I(inode)->disk_i_size) {
1308		btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
1309		mark_inode_dirty(inode);
1310	}
1311
1312	if (written < 0 || written == count)
1313		return written;
1314
1315	pos += written;
1316	count -= written;
1317	iov_iter_init(&i, iov, nr_segs, count, written);
1318	written_buffered = __btrfs_buffered_write(file, &i, pos);
1319	if (written_buffered < 0) {
1320		err = written_buffered;
1321		goto out;
1322	}
1323	endbyte = pos + written_buffered - 1;
1324	err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
1325	if (err)
1326		goto out;
1327	written += written_buffered;
1328	*ppos = pos + written_buffered;
1329	invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT,
1330				 endbyte >> PAGE_CACHE_SHIFT);
1331out:
1332	return written ? written : err;
1333}
1334
1335static ssize_t btrfs_file_aio_write(struct kiocb *iocb,
1336				    const struct iovec *iov,
1337				    unsigned long nr_segs, loff_t pos)
1338{
1339	struct file *file = iocb->ki_filp;
1340	struct inode *inode = fdentry(file)->d_inode;
1341	struct btrfs_root *root = BTRFS_I(inode)->root;
1342	loff_t *ppos = &iocb->ki_pos;
1343	u64 start_pos;
1344	ssize_t num_written = 0;
1345	ssize_t err = 0;
1346	size_t count, ocount;
1347
1348	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1349
1350	mutex_lock(&inode->i_mutex);
1351
1352	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
1353	if (err) {
1354		mutex_unlock(&inode->i_mutex);
1355		goto out;
1356	}
1357	count = ocount;
1358
1359	current->backing_dev_info = inode->i_mapping->backing_dev_info;
1360	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1361	if (err) {
1362		mutex_unlock(&inode->i_mutex);
1363		goto out;
1364	}
1365
1366	if (count == 0) {
1367		mutex_unlock(&inode->i_mutex);
1368		goto out;
1369	}
1370
1371	err = file_remove_suid(file);
1372	if (err) {
1373		mutex_unlock(&inode->i_mutex);
1374		goto out;
1375	}
1376
1377	/*
1378	 * If BTRFS flips readonly due to some impossible error
1379	 * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR),
1380	 * although we have opened a file as writable, we have
1381	 * to stop this write operation to ensure FS consistency.
1382	 */
1383	if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
1384		mutex_unlock(&inode->i_mutex);
1385		err = -EROFS;
1386		goto out;
1387	}
1388
1389	file_update_time(file);
1390	BTRFS_I(inode)->sequence++;
1391
1392	start_pos = round_down(pos, root->sectorsize);
1393	if (start_pos > i_size_read(inode)) {
1394		err = btrfs_cont_expand(inode, i_size_read(inode), start_pos);
1395		if (err) {
1396			mutex_unlock(&inode->i_mutex);
1397			goto out;
1398		}
1399	}
1400
1401	if (unlikely(file->f_flags & O_DIRECT)) {
1402		num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1403						   pos, ppos, count, ocount);
1404	} else {
1405		struct iov_iter i;
1406
1407		iov_iter_init(&i, iov, nr_segs, count, num_written);
1408
1409		num_written = __btrfs_buffered_write(file, &i, pos);
1410		if (num_written > 0)
1411			*ppos = pos + num_written;
1412	}
1413
1414	mutex_unlock(&inode->i_mutex);
1415
1416	/*
1417	 * we want to make sure fsync finds this change
1418	 * but we haven't joined a transaction running right now.
1419	 *
1420	 * Later on, someone is sure to update the inode and get the
1421	 * real transid recorded.
1422	 *
1423	 * We set last_trans now to the fs_info generation + 1,
1424	 * this will either be one more than the running transaction
1425	 * or the generation used for the next transaction if there isn't
1426	 * one running right now.
1427	 */
1428	BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
1429	if (num_written > 0 || num_written == -EIOCBQUEUED) {
1430		err = generic_write_sync(file, pos, num_written);
1431		if (err < 0 && num_written > 0)
1432			num_written = err;
1433	}
1434out:
1435	current->backing_dev_info = NULL;
1436	return num_written ? num_written : err;
1437}
1438
1439int btrfs_release_file(struct inode *inode, struct file *filp)
1440{
1441	/*
1442	 * ordered_data_close is set by settattr when we are about to truncate
1443	 * a file from a non-zero size to a zero size.  This tries to
1444	 * flush down new bytes that may have been written if the
1445	 * application were using truncate to replace a file in place.
1446	 */
1447	if (BTRFS_I(inode)->ordered_data_close) {
1448		BTRFS_I(inode)->ordered_data_close = 0;
1449		btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
1450		if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
1451			filemap_flush(inode->i_mapping);
1452	}
1453	if (filp->private_data)
1454		btrfs_ioctl_trans_end(filp);
1455	return 0;
1456}
1457
1458/*
1459 * fsync call for both files and directories.  This logs the inode into
1460 * the tree log instead of forcing full commits whenever possible.
1461 *
1462 * It needs to call filemap_fdatawait so that all ordered extent updates are
1463 * in the metadata btree are up to date for copying to the log.
1464 *
1465 * It drops the inode mutex before doing the tree log commit.  This is an
1466 * important optimization for directories because holding the mutex prevents
1467 * new operations on the dir while we write to disk.
1468 */
1469int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
1470{
1471	struct dentry *dentry = file->f_path.dentry;
1472	struct inode *inode = dentry->d_inode;
1473	struct btrfs_root *root = BTRFS_I(inode)->root;
1474	int ret = 0;
1475	struct btrfs_trans_handle *trans;
1476
1477	trace_btrfs_sync_file(file, datasync);
1478
1479	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1480	if (ret)
1481		return ret;
1482	mutex_lock(&inode->i_mutex);
1483
1484	/* we wait first, since the writeback may change the inode */
1485	root->log_batch++;
1486	btrfs_wait_ordered_range(inode, 0, (u64)-1);
1487	root->log_batch++;
1488
1489	/*
1490	 * check the transaction that last modified this inode
1491	 * and see if its already been committed
1492	 */
1493	if (!BTRFS_I(inode)->last_trans) {
1494		mutex_unlock(&inode->i_mutex);
1495		goto out;
1496	}
1497
1498	/*
1499	 * if the last transaction that changed this file was before
1500	 * the current transaction, we can bail out now without any
1501	 * syncing
1502	 */
1503	smp_mb();
1504	if (BTRFS_I(inode)->last_trans <=
1505	    root->fs_info->last_trans_committed) {
1506		BTRFS_I(inode)->last_trans = 0;
1507		mutex_unlock(&inode->i_mutex);
1508		goto out;
1509	}
1510
1511	/*
1512	 * ok we haven't committed the transaction yet, lets do a commit
1513	 */
1514	if (file->private_data)
1515		btrfs_ioctl_trans_end(file);
1516
1517	trans = btrfs_start_transaction(root, 0);
1518	if (IS_ERR(trans)) {
1519		ret = PTR_ERR(trans);
1520		mutex_unlock(&inode->i_mutex);
1521		goto out;
1522	}
1523
1524	ret = btrfs_log_dentry_safe(trans, root, dentry);
1525	if (ret < 0) {
1526		mutex_unlock(&inode->i_mutex);
1527		goto out;
1528	}
1529
1530	/* we've logged all the items and now have a consistent
1531	 * version of the file in the log.  It is possible that
1532	 * someone will come in and modify the file, but that's
1533	 * fine because the log is consistent on disk, and we
1534	 * have references to all of the file's extents
1535	 *
1536	 * It is possible that someone will come in and log the
1537	 * file again, but that will end up using the synchronization
1538	 * inside btrfs_sync_log to keep things safe.
1539	 */
1540	mutex_unlock(&inode->i_mutex);
1541
1542	if (ret != BTRFS_NO_LOG_SYNC) {
1543		if (ret > 0) {
1544			ret = btrfs_commit_transaction(trans, root);
1545		} else {
1546			ret = btrfs_sync_log(trans, root);
1547			if (ret == 0)
1548				ret = btrfs_end_transaction(trans, root);
1549			else
1550				ret = btrfs_commit_transaction(trans, root);
1551		}
1552	} else {
1553		ret = btrfs_end_transaction(trans, root);
1554	}
1555out:
1556	return ret > 0 ? -EIO : ret;
1557}
1558
1559static const struct vm_operations_struct btrfs_file_vm_ops = {
1560	.fault		= filemap_fault,
1561	.page_mkwrite	= btrfs_page_mkwrite,
1562};
1563
1564static int btrfs_file_mmap(struct file	*filp, struct vm_area_struct *vma)
1565{
1566	struct address_space *mapping = filp->f_mapping;
1567
1568	if (!mapping->a_ops->readpage)
1569		return -ENOEXEC;
1570
1571	file_accessed(filp);
1572	vma->vm_ops = &btrfs_file_vm_ops;
1573	vma->vm_flags |= VM_CAN_NONLINEAR;
1574
1575	return 0;
1576}
1577
1578static long btrfs_fallocate(struct file *file, int mode,
1579			    loff_t offset, loff_t len)
1580{
1581	struct inode *inode = file->f_path.dentry->d_inode;
1582	struct extent_state *cached_state = NULL;
1583	u64 cur_offset;
1584	u64 last_byte;
1585	u64 alloc_start;
1586	u64 alloc_end;
1587	u64 alloc_hint = 0;
1588	u64 locked_end;
1589	u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
1590	struct extent_map *em;
1591	int ret;
1592
1593	alloc_start = offset & ~mask;
1594	alloc_end =  (offset + len + mask) & ~mask;
1595
1596	/* We only support the FALLOC_FL_KEEP_SIZE mode */
1597	if (mode & ~FALLOC_FL_KEEP_SIZE)
1598		return -EOPNOTSUPP;
1599
1600	/*
1601	 * wait for ordered IO before we have any locks.  We'll loop again
1602	 * below with the locks held.
1603	 */
1604	btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
1605
1606	mutex_lock(&inode->i_mutex);
1607	ret = inode_newsize_ok(inode, alloc_end);
1608	if (ret)
1609		goto out;
1610
1611	if (alloc_start > inode->i_size) {
1612		ret = btrfs_cont_expand(inode, i_size_read(inode),
1613					alloc_start);
1614		if (ret)
1615			goto out;
1616	}
1617
1618	ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
1619	if (ret)
1620		goto out;
1621
1622	locked_end = alloc_end - 1;
1623	while (1) {
1624		struct btrfs_ordered_extent *ordered;
1625
1626		/* the extent lock is ordered inside the running
1627		 * transaction
1628		 */
1629		lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
1630				 locked_end, 0, &cached_state, GFP_NOFS);
1631		ordered = btrfs_lookup_first_ordered_extent(inode,
1632							    alloc_end - 1);
1633		if (ordered &&
1634		    ordered->file_offset + ordered->len > alloc_start &&
1635		    ordered->file_offset < alloc_end) {
1636			btrfs_put_ordered_extent(ordered);
1637			unlock_extent_cached(&BTRFS_I(inode)->io_tree,
1638					     alloc_start, locked_end,
1639					     &cached_state, GFP_NOFS);
1640			/*
1641			 * we can't wait on the range with the transaction
1642			 * running or with the extent lock held
1643			 */
1644			btrfs_wait_ordered_range(inode, alloc_start,
1645						 alloc_end - alloc_start);
1646		} else {
1647			if (ordered)
1648				btrfs_put_ordered_extent(ordered);
1649			break;
1650		}
1651	}
1652
1653	cur_offset = alloc_start;
1654	while (1) {
1655		u64 actual_end;
1656
1657		em = btrfs_get_extent(inode, NULL, 0, cur_offset,
1658				      alloc_end - cur_offset, 0);
1659		BUG_ON(IS_ERR_OR_NULL(em));
1660		last_byte = min(extent_map_end(em), alloc_end);
1661		actual_end = min_t(u64, extent_map_end(em), offset + len);
1662		last_byte = (last_byte + mask) & ~mask;
1663
1664		if (em->block_start == EXTENT_MAP_HOLE ||
1665		    (cur_offset >= inode->i_size &&
1666		     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
1667			ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
1668							last_byte - cur_offset,
1669							1 << inode->i_blkbits,
1670							offset + len,
1671							&alloc_hint);
1672			if (ret < 0) {
1673				free_extent_map(em);
1674				break;
1675			}
1676		} else if (actual_end > inode->i_size &&
1677			   !(mode & FALLOC_FL_KEEP_SIZE)) {
1678			/*
1679			 * We didn't need to allocate any more space, but we
1680			 * still extended the size of the file so we need to
1681			 * update i_size.
1682			 */
1683			inode->i_ctime = CURRENT_TIME;
1684			i_size_write(inode, actual_end);
1685			btrfs_ordered_update_i_size(inode, actual_end, NULL);
1686		}
1687		free_extent_map(em);
1688
1689		cur_offset = last_byte;
1690		if (cur_offset >= alloc_end) {
1691			ret = 0;
1692			break;
1693		}
1694	}
1695	unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
1696			     &cached_state, GFP_NOFS);
1697
1698	btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
1699out:
1700	mutex_unlock(&inode->i_mutex);
1701	return ret;
1702}
1703
1704static int find_desired_extent(struct inode *inode, loff_t *offset, int origin)
1705{
1706	struct btrfs_root *root = BTRFS_I(inode)->root;
1707	struct extent_map *em;
1708	struct extent_state *cached_state = NULL;
1709	u64 lockstart = *offset;
1710	u64 lockend = i_size_read(inode);
1711	u64 start = *offset;
1712	u64 orig_start = *offset;
1713	u64 len = i_size_read(inode);
1714	u64 last_end = 0;
1715	int ret = 0;
1716
1717	lockend = max_t(u64, root->sectorsize, lockend);
1718	if (lockend <= lockstart)
1719		lockend = lockstart + root->sectorsize;
1720
1721	len = lockend - lockstart + 1;
1722
1723	len = max_t(u64, len, root->sectorsize);
1724	if (inode->i_size == 0)
1725		return -ENXIO;
1726
1727	lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0,
1728			 &cached_state, GFP_NOFS);
1729
1730	/*
1731	 * Delalloc is such a pain.  If we have a hole and we have pending
1732	 * delalloc for a portion of the hole we will get back a hole that
1733	 * exists for the entire range since it hasn't been actually written
1734	 * yet.  So to take care of this case we need to look for an extent just
1735	 * before the position we want in case there is outstanding delalloc
1736	 * going on here.
1737	 */
1738	if (origin == SEEK_HOLE && start != 0) {
1739		if (start <= root->sectorsize)
1740			em = btrfs_get_extent_fiemap(inode, NULL, 0, 0,
1741						     root->sectorsize, 0);
1742		else
1743			em = btrfs_get_extent_fiemap(inode, NULL, 0,
1744						     start - root->sectorsize,
1745						     root->sectorsize, 0);
1746		if (IS_ERR(em)) {
1747			ret = -ENXIO;
1748			goto out;
1749		}
1750		last_end = em->start + em->len;
1751		if (em->block_start == EXTENT_MAP_DELALLOC)
1752			last_end = min_t(u64, last_end, inode->i_size);
1753		free_extent_map(em);
1754	}
1755
1756	while (1) {
1757		em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0);
1758		if (IS_ERR(em)) {
1759			ret = -ENXIO;
1760			break;
1761		}
1762
1763		if (em->block_start == EXTENT_MAP_HOLE) {
1764			if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1765				if (last_end <= orig_start) {
1766					free_extent_map(em);
1767					ret = -ENXIO;
1768					break;
1769				}
1770			}
1771
1772			if (origin == SEEK_HOLE) {
1773				*offset = start;
1774				free_extent_map(em);
1775				break;
1776			}
1777		} else {
1778			if (origin == SEEK_DATA) {
1779				if (em->block_start == EXTENT_MAP_DELALLOC) {
1780					if (start >= inode->i_size) {
1781						free_extent_map(em);
1782						ret = -ENXIO;
1783						break;
1784					}
1785				}
1786
1787				*offset = start;
1788				free_extent_map(em);
1789				break;
1790			}
1791		}
1792
1793		start = em->start + em->len;
1794		last_end = em->start + em->len;
1795
1796		if (em->block_start == EXTENT_MAP_DELALLOC)
1797			last_end = min_t(u64, last_end, inode->i_size);
1798
1799		if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
1800			free_extent_map(em);
1801			ret = -ENXIO;
1802			break;
1803		}
1804		free_extent_map(em);
1805		cond_resched();
1806	}
1807	if (!ret)
1808		*offset = min(*offset, inode->i_size);
1809out:
1810	unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
1811			     &cached_state, GFP_NOFS);
1812	return ret;
1813}
1814
1815static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int origin)
1816{
1817	struct inode *inode = file->f_mapping->host;
1818	int ret;
1819
1820	mutex_lock(&inode->i_mutex);
1821	switch (origin) {
1822	case SEEK_END:
1823	case SEEK_CUR:
1824		offset = generic_file_llseek_unlocked(file, offset, origin);
1825		goto out;
1826	case SEEK_DATA:
1827	case SEEK_HOLE:
1828		if (offset >= i_size_read(inode)) {
1829			mutex_unlock(&inode->i_mutex);
1830			return -ENXIO;
1831		}
1832
1833		ret = find_desired_extent(inode, &offset, origin);
1834		if (ret) {
1835			mutex_unlock(&inode->i_mutex);
1836			return ret;
1837		}
1838	}
1839
1840	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
1841		offset = -EINVAL;
1842		goto out;
1843	}
1844	if (offset > inode->i_sb->s_maxbytes) {
1845		offset = -EINVAL;
1846		goto out;
1847	}
1848
1849	/* Special lock needed here? */
1850	if (offset != file->f_pos) {
1851		file->f_pos = offset;
1852		file->f_version = 0;
1853	}
1854out:
1855	mutex_unlock(&inode->i_mutex);
1856	return offset;
1857}
1858
1859const struct file_operations btrfs_file_operations = {
1860	.llseek		= btrfs_file_llseek,
1861	.read		= do_sync_read,
1862	.write		= do_sync_write,
1863	.aio_read       = generic_file_aio_read,
1864	.splice_read	= generic_file_splice_read,
1865	.aio_write	= btrfs_file_aio_write,
1866	.mmap		= btrfs_file_mmap,
1867	.open		= generic_file_open,
1868	.release	= btrfs_release_file,
1869	.fsync		= btrfs_sync_file,
1870	.fallocate	= btrfs_fallocate,
1871	.unlocked_ioctl	= btrfs_ioctl,
1872#ifdef CONFIG_COMPAT
1873	.compat_ioctl	= btrfs_ioctl,
1874#endif
1875};